tighten

Reverts openai/codex#6485

.github/pull_request_template.md

@@ -4,3 +4,5 @@ Before opening this Pull Request, please read the dedicated "Contributing" markd
 https://github.com/openai/codex/blob/main/docs/contributing.md
 
 If your PR conforms to our contribution guidelines, replace this text with a detailed and high quality description of your changes.
+
+Include a link to a bug report or enhancement request.

.github/workflows/ci.yml

@@ -46,7 +46,7 @@ jobs:
           echo "pack_output=$PACK_OUTPUT" >> "$GITHUB_OUTPUT"
 
       - name: Upload staged npm package artifact
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v5
         with:
           name: codex-npm-staging
           path: ${{ steps.stage_npm_package.outputs.pack_output }}

.github/workflows/cla.yml

@@ -13,13 +13,33 @@ permissions:
 
 jobs:
   cla:
+    # Only run the CLA assistant for the canonical openai repo so forks are not blocked
+    # and contributors who signed previously do not receive duplicate CLA notifications.
+    if: ${{ github.repository_owner == 'openai' }}
     runs-on: ubuntu-latest
     steps:
       - uses: contributor-assistant/github-action@v2.6.1
+        # Run on close only if the PR was merged. This will lock the PR to preserve
+        # the CLA agreement. We don't want to lock PRs that have been closed without
+        # merging because the contributor may want to respond with additional comments.
+        # This action has a "lock-pullrequest-aftermerge" option that can be set to false,
+        # but that would unconditionally skip locking even in cases where the PR was merged.
         if: |
-          github.event_name == 'pull_request_target' ||
-          github.event.comment.body == 'recheck' ||
-          github.event.comment.body == 'I have read the CLA Document and I hereby sign the CLA'
+          (
+            github.event_name == 'pull_request_target' &&
+            (
+              github.event.action == 'opened' ||
+              github.event.action == 'synchronize' ||
+              (github.event.action == 'closed' && github.event.pull_request.merged == true)
+            )
+          ) ||
+          (
+            github.event_name == 'issue_comment' &&
+            (
+              github.event.comment.body == 'recheck' ||
+              github.event.comment.body == 'I have read the CLA Document and I hereby sign the CLA'
+            )
+          )
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         with:

.github/workflows/close-stale-contributor-prs.yml

@@ -0,0 +1,105 @@
+name: Close stale contributor PRs
+
+on:
+  workflow_dispatch:
+  schedule:
+    - cron: "0 6 * * *"
+
+permissions:
+  contents: read
+  issues: write
+  pull-requests: write
+
+jobs:
+  close-stale-contributor-prs:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Close inactive PRs from contributors
+        uses: actions/github-script@v7
+        with:
+          github-token: ${{ secrets.GITHUB_TOKEN }}
+          script: |
+            const DAYS_INACTIVE = 14;
+            const cutoff = new Date(Date.now() - DAYS_INACTIVE * 24 * 60 * 60 * 1000);
+            const { owner, repo } = context.repo;
+            const dryRun = false;
+            const stalePrs = [];
+
+            core.info(`Dry run mode: ${dryRun}`);
+
+            const prs = await github.paginate(github.rest.pulls.list, {
+              owner,
+              repo,
+              state: "open",
+              per_page: 100,
+              sort: "updated",
+              direction: "asc",
+            });
+
+            for (const pr of prs) {
+              const lastUpdated = new Date(pr.updated_at);
+              if (lastUpdated > cutoff) {
+                core.info(`PR ${pr.number} is fresh`);
+                continue;
+              }
+
+              if (!pr.user || pr.user.type !== "User") {
+                core.info(`PR ${pr.number} wasn't created by a user`);
+                continue;
+              }
+
+              let permission;
+              try {
+                const permissionResponse = await github.rest.repos.getCollaboratorPermissionLevel({
+                  owner,
+                  repo,
+                  username: pr.user.login,
+                });
+                permission = permissionResponse.data.permission;
+              } catch (error) {
+                if (error.status === 404) {
+                  core.info(`Author ${pr.user.login} is not a collaborator; skipping #${pr.number}`);
+                  continue;
+                }
+                throw error;
+              }
+
+              const hasContributorAccess = ["admin", "maintain", "write"].includes(permission);
+              if (!hasContributorAccess) {
+                core.info(`Author ${pr.user.login} has ${permission} access; skipping #${pr.number}`);
+                continue;
+              }
+
+              stalePrs.push(pr);
+            }
+
+            if (!stalePrs.length) {
+              core.info("No stale contributor pull requests found.");
+              return;
+            }
+
+            for (const pr of stalePrs) {
+              const issue_number = pr.number;
+              const closeComment = `Closing this pull request because it has had no updates for more than ${DAYS_INACTIVE} days. If you plan to continue working on it, feel free to reopen or open a new PR.`;
+
+              if (dryRun) {
+                core.info(`[dry-run] Would close contributor PR #${issue_number} from ${pr.user.login}`);
+                continue;
+              }
+
+              await github.rest.issues.createComment({
+                owner,
+                repo,
+                issue_number,
+                body: closeComment,
+              });
+
+              await github.rest.pulls.update({
+                owner,
+                repo,
+                pull_number: issue_number,
+                state: "closed",
+              });
+
+              core.info(`Closed contributor PR #${issue_number} from ${pr.user.login}`);
+            }

.github/workflows/codespell.yml

@@ -22,6 +22,6 @@ jobs:
       - name: Annotate locations with typos
         uses: codespell-project/codespell-problem-matcher@b80729f885d32f78a716c2f107b4db1025001c42 # v1
       - name: Codespell
-        uses: codespell-project/actions-codespell@406322ec52dd7b488e48c1c4b82e2a8b3a1bf630 # v2.1
+        uses: codespell-project/actions-codespell@8f01853be192eb0f849a5c7d721450e7a467c579 # v2.2
         with:
           ignore_words_file: .codespellignore

.github/workflows/issue-labeler.yml

@@ -26,21 +26,36 @@ jobs:
           prompt: |
             You are an assistant that reviews GitHub issues for the repository.
 
-            Your job is to choose the most appropriate existing labels for the issue described later in this prompt.
+            Your job is to choose the most appropriate labels for the issue described later in this prompt.
             Follow these rules:
-            - Only pick labels out of the list below.
-            - Prefer a small set of precise labels over many broad ones.
 
-            Labels to apply:
+            - Add one (and only one) of the following three labels to distinguish the type of issue. Default to "bug" if unsure.
             1. bug — Reproducible defects in Codex products (CLI, VS Code extension, web, auth).
             2. enhancement — Feature requests or usability improvements that ask for new capabilities, better ergonomics, or quality-of-life tweaks.
-            3. extension — VS Code (or other IDE) extension-specific issues.
-            4. windows-os — Bugs or friction specific to Windows environments (always when PowerShell is mentioned, path handling, copy/paste, OS-specific auth or tooling failures).
-            5. mcp — Topics involving Model Context Protocol servers/clients.
-            6. codex-web — Issues targeting the Codex web UI/Cloud experience.
-            8. azure — Problems or requests tied to Azure OpenAI deployments.
-            9. documentation — Updates or corrections needed in docs/README/config references (broken links, missing examples, outdated keys, clarification requests).
-            10. model-behavior — Undesirable LLM behavior: forgetting goals, refusing work, hallucinating environment details, quota misreports, or other reasoning/performance anomalies.
+            3. documentation — Updates or corrections needed in docs/README/config references (broken links, missing examples, outdated keys, clarification requests).
+
+            - If applicable, add one of the following labels to specify which sub-product or product surface the issue relates to.
+            1. CLI — the Codex command line interface.
+            2. extension — VS Code (or other IDE) extension-specific issues.
+            3. codex-web — Issues targeting the Codex web UI/Cloud experience.
+            4. github-action — Issues with the Codex GitHub action.
+            5. iOS — Issues with the Codex iOS app.
+
+            - Additionally add zero or more of the following labels that are relevant to the issue content. Prefer a small set of precise labels over many broad ones.
+            1. windows-os — Bugs or friction specific to Windows environments (always when PowerShell is mentioned, path handling, copy/paste, OS-specific auth or tooling failures).
+            2. mcp — Topics involving Model Context Protocol servers/clients.
+            3. mcp-server — Problems related to the codex mcp-server command, where codex runs as an MCP server.
+            4. azure — Problems or requests tied to Azure OpenAI deployments.
+            5. model-behavior — Undesirable LLM behavior: forgetting goals, refusing work, hallucinating environment details, quota misreports, or other reasoning/performance anomalies.
+            6. code-review — Issues related to the code review feature or functionality.
+            7. auth - Problems related to authentication, login, or access tokens.
+            8. codex-exec - Problems related to the "codex exec" command or functionality.
+            9. context-management - Problems related to compaction, context windows, or available context reporting.
+            10. custom-model - Problems that involve using custom model providers, local models, or OSS models.
+            11. rate-limits - Problems related to token limits, rate limits, or token usage reporting.
+            12. sandbox - Issues related to local sandbox environments or tool call approvals to override sandbox restrictions.
+            13. tool-calls - Problems related to specific tool call invocations including unexpected errors, failures, or hangs.
+            14. TUI - Problems with the terminal user interface (TUI) including keyboard shortcuts, copy & pasting, menus, or screen update issues.
 
             Issue number: ${{ github.event.issue.number }}
 

.github/workflows/rust-ci.yml

@@ -76,7 +76,7 @@ jobs:
     steps:
       - uses: actions/checkout@v5
       - uses: dtolnay/rust-toolchain@1.90
-      - uses: taiki-e/install-action@0c5db7f7f897c03b771660e91d065338615679f4 # v2
+      - uses: taiki-e/install-action@44c6d64aa62cd779e873306675c7a58e86d6d532 # v2
         with:
           tool: cargo-shear
           version: 1.5.1
@@ -95,8 +95,9 @@ jobs:
       run:
         working-directory: codex-rs
     env:
-      # Speed up repeated builds across CI runs by caching compiled objects.
-      RUSTC_WRAPPER: sccache
+      # Speed up repeated builds across CI runs by caching compiled objects (non-Windows).
+      USE_SCCACHE: ${{ startsWith(matrix.runner, 'windows') && 'false' || 'true' }}
+      RUSTC_WRAPPER: ${{ startsWith(matrix.runner, 'windows') && '' || 'sccache' }}
       CARGO_INCREMENTAL: "0"
       SCCACHE_CACHE_SIZE: 10G
 
@@ -170,12 +171,14 @@ jobs:
 
       # Install and restore sccache cache
       - name: Install sccache
-        uses: taiki-e/install-action@0c5db7f7f897c03b771660e91d065338615679f4 # v2
+        if: ${{ env.USE_SCCACHE == 'true' }}
+        uses: taiki-e/install-action@44c6d64aa62cd779e873306675c7a58e86d6d532 # v2
         with:
           tool: sccache
           version: 0.7.5
 
       - name: Configure sccache backend
+        if: ${{ env.USE_SCCACHE == 'true' }}
         shell: bash
         run: |
           set -euo pipefail
@@ -189,7 +192,7 @@ jobs:
           fi
 
       - name: Restore sccache cache (fallback)
-        if: ${{ env.SCCACHE_GHA_ENABLED != 'true' }}
+        if: ${{ env.USE_SCCACHE == 'true' && env.SCCACHE_GHA_ENABLED != 'true' }}
         id: cache_sccache_restore
         uses: actions/cache/restore@v4
         with:
@@ -228,7 +231,7 @@ jobs:
 
       - name: Install cargo-chef
         if: ${{ matrix.profile == 'release' }}
-        uses: taiki-e/install-action@0c5db7f7f897c03b771660e91d065338615679f4 # v2
+        uses: taiki-e/install-action@44c6d64aa62cd779e873306675c7a58e86d6d532 # v2
         with:
           tool: cargo-chef
           version: 0.1.71
@@ -274,7 +277,7 @@ jobs:
           key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ hashFiles('codex-rs/rust-toolchain.toml') }}
 
       - name: Save sccache cache (fallback)
-        if: always() && !cancelled() && env.SCCACHE_GHA_ENABLED != 'true'
+        if: always() && !cancelled() && env.USE_SCCACHE == 'true' && env.SCCACHE_GHA_ENABLED != 'true'
         continue-on-error: true
         uses: actions/cache/save@v4
         with:
@@ -282,12 +285,12 @@ jobs:
           key: sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ github.run_id }}
 
       - name: sccache stats
-        if: always()
+        if: always() && env.USE_SCCACHE == 'true'
         continue-on-error: true
         run: sccache --show-stats || true
 
       - name: sccache summary
-        if: always()
+        if: always() && env.USE_SCCACHE == 'true'
         shell: bash
         run: |
           {
@@ -326,7 +329,9 @@ jobs:
       run:
         working-directory: codex-rs
     env:
-      RUSTC_WRAPPER: sccache
+      # Speed up repeated builds across CI runs by caching compiled objects (non-Windows).
+      USE_SCCACHE: ${{ startsWith(matrix.runner, 'windows') && 'false' || 'true' }}
+      RUSTC_WRAPPER: ${{ startsWith(matrix.runner, 'windows') && '' || 'sccache' }}
       CARGO_INCREMENTAL: "0"
       SCCACHE_CACHE_SIZE: 10G
 
@@ -370,12 +375,14 @@ jobs:
             cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-
 
       - name: Install sccache
-        uses: taiki-e/install-action@0c5db7f7f897c03b771660e91d065338615679f4 # v2
+        if: ${{ env.USE_SCCACHE == 'true' }}
+        uses: taiki-e/install-action@44c6d64aa62cd779e873306675c7a58e86d6d532 # v2
         with:
           tool: sccache
           version: 0.7.5
 
       - name: Configure sccache backend
+        if: ${{ env.USE_SCCACHE == 'true' }}
         shell: bash
         run: |
           set -euo pipefail
@@ -389,7 +396,7 @@ jobs:
           fi
 
       - name: Restore sccache cache (fallback)
-        if: ${{ env.SCCACHE_GHA_ENABLED != 'true' }}
+        if: ${{ env.USE_SCCACHE == 'true' && env.SCCACHE_GHA_ENABLED != 'true' }}
         id: cache_sccache_restore
         uses: actions/cache/restore@v4
         with:
@@ -399,17 +406,17 @@ jobs:
             sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-
             sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-
 
-      - uses: taiki-e/install-action@0c5db7f7f897c03b771660e91d065338615679f4 # v2
+      - uses: taiki-e/install-action@44c6d64aa62cd779e873306675c7a58e86d6d532 # v2
         with:
           tool: nextest
           version: 0.9.103
 
       - name: tests
         id: test
-        continue-on-error: true
         run: cargo nextest run --all-features --no-fail-fast --target ${{ matrix.target }} --cargo-profile ci-test
         env:
           RUST_BACKTRACE: 1
+          NEXTEST_STATUS_LEVEL: leak
 
       - name: Save cargo home cache
         if: always() && !cancelled() && steps.cache_cargo_home_restore.outputs.cache-hit != 'true'
@@ -424,7 +431,7 @@ jobs:
           key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ hashFiles('codex-rs/rust-toolchain.toml') }}
 
       - name: Save sccache cache (fallback)
-        if: always() && !cancelled() && env.SCCACHE_GHA_ENABLED != 'true'
+        if: always() && !cancelled() && env.USE_SCCACHE == 'true' && env.SCCACHE_GHA_ENABLED != 'true'
         continue-on-error: true
         uses: actions/cache/save@v4
         with:
@@ -432,12 +439,12 @@ jobs:
           key: sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ github.run_id }}
 
       - name: sccache stats
-        if: always()
+        if: always() && env.USE_SCCACHE == 'true'
         continue-on-error: true
         run: sccache --show-stats || true
 
       - name: sccache summary
-        if: always()
+        if: always() && env.USE_SCCACHE == 'true'
         shell: bash
         run: |
           {

.github/workflows/rust-release.yml

@@ -295,6 +295,15 @@ jobs:
           # ${{ matrix.target }}
           dest="dist/${{ matrix.target }}"
 
+          # We want to ship the raw Windows executables in the GitHub Release
+          # in addition to the compressed archives. Keep the originals for
+          # Windows targets; remove them elsewhere to limit the number of
+          # artifacts that end up in the GitHub Release.
+          keep_originals=false
+          if [[ "${{ matrix.runner }}" == windows* ]]; then
+            keep_originals=true
+          fi
+
           # For compatibility with environments that lack the `zstd` tool we
           # additionally create a `.tar.gz` for all platforms and `.zip` for
           # Windows alongside every single binary that we publish. The end result is:
@@ -324,7 +333,11 @@ jobs:
 
             # Also create .zst (existing behaviour) *and* remove the original
             # uncompressed binary to keep the directory small.
-            zstd -T0 -19 --rm "$dest/$base"
+            zstd_args=(-T0 -19)
+            if [[ "${keep_originals}" == false ]]; then
+              zstd_args+=(--rm)
+            fi
+            zstd "${zstd_args[@]}" "$dest/$base"
           done
 
       - name: Remove signing keychain
@@ -350,7 +363,7 @@ jobs:
             fi
           fi
 
-      - uses: actions/upload-artifact@v4
+      - uses: actions/upload-artifact@v5
         with:
           name: ${{ matrix.target }}
           # Upload the per-binary .zst files as well as the new .tar.gz

AGENTS.md

@@ -84,6 +84,8 @@ If you don’t have the tool:
 - Use `ResponseMock::single_request()` when a test should only issue one POST, or `ResponseMock::requests()` to inspect every captured `ResponsesRequest`.
 - `ResponsesRequest` exposes helpers (`body_json`, `input`, `function_call_output`, `custom_tool_call_output`, `call_output`, `header`, `path`, `query_param`) so assertions can target structured payloads instead of manual JSON digging.
 - Build SSE payloads with the provided `ev_*` constructors and the `sse(...)`.
+- Prefer `wait_for_event` over `wait_for_event_with_timeout`.
+- Prefer `mount_sse_once` over `mount_sse_once_match` or `mount_sse_sequence`
 
 - Typical pattern:
 

README.md

@@ -75,11 +75,13 @@ Codex CLI supports a rich set of configuration options, with preferences stored
 
 - [**Getting started**](./docs/getting-started.md)
   - [CLI usage](./docs/getting-started.md#cli-usage)
+  - [Slash Commands](./docs/slash_commands.md)
   - [Running with a prompt as input](./docs/getting-started.md#running-with-a-prompt-as-input)
   - [Example prompts](./docs/getting-started.md#example-prompts)
   - [Custom prompts](./docs/prompts.md)
   - [Memory with AGENTS.md](./docs/getting-started.md#memory-with-agentsmd)
 - [**Configuration**](./docs/config.md)
+  - [Example config](./docs/example-config.md)
 - [**Sandbox & approvals**](./docs/sandbox.md)
 - [**Authentication**](./docs/authentication.md)
   - [Auth methods](./docs/authentication.md#forcing-a-specific-auth-method-advanced)

codex-rs/.config/nextest.toml

@@ -0,0 +1,9 @@
+[profile.default]
+# Do not increase, fix your test instead
+slow-timeout = { period = "15s", terminate-after = 2 }
+
+
+[[profile.default.overrides]]
+# Do not add new tests here
+filter = 'test(rmcp_client) | test(humanlike_typing_1000_chars_appears_live_no_placeholder)'
+slow-timeout = { period = "1m", terminate-after = 4 }

codex-rs/Cargo.lock

@@ -186,9 +186,11 @@ dependencies = [
  "chrono",
  "codex-app-server-protocol",
  "codex-core",
+ "codex-protocol",
  "serde",
  "serde_json",
  "tokio",
+ "uuid",
  "wiremock",
 ]
 
@@ -209,6 +211,7 @@ dependencies = [
  "parking_lot",
  "percent-encoding",
  "windows-sys 0.59.0",
+ "wl-clipboard-rs",
  "x11rb",
 ]
 
@@ -235,46 +238,44 @@ dependencies = [
 
 [[package]]
 name = "askama"
-version = "0.12.1"
+version = "0.14.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b79091df18a97caea757e28cd2d5fda49c6cd4bd01ddffd7ff01ace0c0ad2c28"
+checksum = "f75363874b771be265f4ffe307ca705ef6f3baa19011c149da8674a87f1b75c4"
 dependencies = [
  "askama_derive",
- "askama_escape",
- "humansize",
- "num-traits",
+ "itoa",
  "percent-encoding",
+ "serde",
+ "serde_json",
 ]
 
 [[package]]
 name = "askama_derive"
-version = "0.12.5"
+version = "0.14.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "19fe8d6cb13c4714962c072ea496f3392015f0989b1a2847bb4b2d9effd71d83"
+checksum = "129397200fe83088e8a68407a8e2b1f826cf0086b21ccdb866a722c8bcd3a94f"
 dependencies = [
  "askama_parser",
  "basic-toml",
- "mime",
- "mime_guess",
+ "memchr",
  "proc-macro2",
  "quote",
+ "rustc-hash 2.1.1",
  "serde",
+ "serde_derive",
  "syn 2.0.104",
 ]
 
-[[package]]
-name = "askama_escape"
-version = "0.10.3"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "619743e34b5ba4e9703bba34deac3427c72507c7159f5fd030aea8cac0cfe341"
-
 [[package]]
 name = "askama_parser"
-version = "0.2.1"
+version = "0.14.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "acb1161c6b64d1c3d83108213c2a2533a342ac225aabd0bda218278c2ddb00c0"
+checksum = "d6ab5630b3d5eaf232620167977f95eb51f3432fc76852328774afbd242d4358"
 dependencies = [
- "nom",
+ "memchr",
+ "serde",
+ "serde_derive",
+ "winnow",
 ]
 
 [[package]]
@@ -849,6 +850,7 @@ dependencies = [
  "codex-protocol",
  "codex-utils-json-to-toml",
  "core_test_support",
+ "mcp-types",
  "opentelemetry-appender-tracing",
  "os_info",
  "pretty_assertions",
@@ -871,6 +873,7 @@ dependencies = [
  "anyhow",
  "clap",
  "codex-protocol",
+ "mcp-types",
  "paste",
  "pretty_assertions",
  "schemars 0.8.22",
@@ -881,6 +884,19 @@ dependencies = [
  "uuid",
 ]
 
+[[package]]
+name = "codex-app-server-test-client"
+version = "0.0.0"
+dependencies = [
+ "anyhow",
+ "clap",
+ "codex-app-server-protocol",
+ "codex-protocol",
+ "serde",
+ "serde_json",
+ "uuid",
+]
+
 [[package]]
 name = "codex-apply-patch"
 version = "0.0.0"
@@ -978,20 +994,23 @@ dependencies = [
  "codex-mcp-server",
  "codex-process-hardening",
  "codex-protocol",
- "codex-protocol-ts",
  "codex-responses-api-proxy",
  "codex-rmcp-client",
  "codex-stdio-to-uds",
  "codex-tui",
  "codex-windows-sandbox",
  "ctor 0.5.0",
+ "libc",
  "owo-colors",
  "predicates",
  "pretty_assertions",
+ "regex-lite",
  "serde_json",
  "supports-color",
  "tempfile",
  "tokio",
+ "toml",
+ "tracing",
 ]
 
 [[package]]
@@ -1043,6 +1062,7 @@ dependencies = [
  "codex-app-server-protocol",
  "codex-core",
  "codex-protocol",
+ "once_cell",
  "serde",
  "toml",
 ]
@@ -1062,6 +1082,7 @@ dependencies = [
  "chrono",
  "codex-app-server-protocol",
  "codex-apply-patch",
+ "codex-arg0",
  "codex-async-utils",
  "codex-file-search",
  "codex-git",
@@ -1076,6 +1097,7 @@ dependencies = [
  "codex-windows-sandbox",
  "core-foundation 0.9.4",
  "core_test_support",
+ "ctor 0.5.0",
  "dirs",
  "dunce",
  "env-flags",
@@ -1107,6 +1129,7 @@ dependencies = [
  "similar",
  "strum_macros 0.27.2",
  "tempfile",
+ "test-case",
  "test-log",
  "thiserror 2.0.17",
  "time",
@@ -1361,16 +1384,6 @@ dependencies = [
  "uuid",
 ]
 
-[[package]]
-name = "codex-protocol-ts"
-version = "0.0.0"
-dependencies = [
- "anyhow",
- "clap",
- "codex-app-server-protocol",
- "ts-rs",
-]
-
 [[package]]
 name = "codex-responses-api-proxy"
 version = "0.0.0"
@@ -1448,8 +1461,10 @@ dependencies = [
  "codex-login",
  "codex-ollama",
  "codex-protocol",
+ "codex-windows-sandbox",
  "color-eyre",
  "crossterm",
+ "derive_more 2.0.1",
  "diffy",
  "dirs",
  "dunce",
@@ -1469,6 +1484,7 @@ dependencies = [
  "regex-lite",
  "serde",
  "serde_json",
+ "serial_test",
  "shlex",
  "strum 0.27.2",
  "strum_macros 0.27.2",
@@ -1548,9 +1564,11 @@ name = "codex-utils-tokenizer"
 version = "0.0.0"
 dependencies = [
  "anyhow",
+ "codex-utils-cache",
  "pretty_assertions",
  "thiserror 2.0.17",
  "tiktoken-rs",
+ "tokio",
 ]
 
 [[package]]
@@ -1559,6 +1577,7 @@ version = "0.1.0"
 dependencies = [
  "anyhow",
  "dirs-next",
+ "dunce",
  "rand 0.8.5",
  "serde",
  "serde_json",
@@ -1652,6 +1671,15 @@ dependencies = [
  "unicode-segmentation",
 ]
 
+[[package]]
+name = "convert_case"
+version = "0.7.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "bb402b8d4c85569410425650ce3eddc7d698ed96d39a73f941b08fb63082f1e7"
+dependencies = [
+ "unicode-segmentation",
+]
+
 [[package]]
 name = "core-foundation"
 version = "0.9.4"
@@ -1750,8 +1778,7 @@ checksum = "d0a5c400df2834b80a4c3327b3aad3a4c4cd4de0629063962b03235697506a28"
 [[package]]
 name = "crossterm"
 version = "0.28.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "829d955a0bb380ef178a640b91779e3987da38c9aea133b20614cfed8cdea9c6"
+source = "git+https://github.com/nornagon/crossterm?branch=nornagon%2Fcolor-query#87db8bfa6dc99427fd3b071681b07fc31c6ce995"
 dependencies = [
  "bitflags 2.10.0",
  "crossterm_winapi",
@@ -1998,7 +2025,7 @@ version = "1.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "cb7330aeadfbe296029522e6c40f315320aba36fc43a5b3632f3795348f3bd22"
 dependencies = [
- "convert_case",
+ "convert_case 0.6.0",
  "proc-macro2",
  "quote",
  "syn 2.0.104",
@@ -2011,6 +2038,7 @@ version = "2.0.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "bda628edc44c4bb645fbe0f758797143e4e07926f7ebf4e9bdfbd3d2ce621df3"
 dependencies = [
+ "convert_case 0.7.1",
  "proc-macro2",
  "quote",
  "syn 2.0.104",
@@ -2874,15 +2902,6 @@ version = "1.0.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "df3b46402a9d5adb4c86a0cf463f42e19994e3ee891101b1841f30a545cb49a9"
 
-[[package]]
-name = "humansize"
-version = "2.1.3"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6cb51c9a029ddc91b07a787f1d86b53ccfa49b0e86688c946ebe8d3555685dd7"
-dependencies = [
- "libm",
-]
-
 [[package]]
 name = "hyper"
 version = "1.7.0"
@@ -3526,12 +3545,6 @@ dependencies = [
  "pkg-config",
 ]
 
-[[package]]
-name = "libm"
-version = "0.2.15"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f9fbbcab51052fe104eb5e5d351cf728d30a5be1fe14d9be8a3b097481fb97de"
-
 [[package]]
 name = "libredox"
 version = "0.1.6"
@@ -4298,6 +4311,16 @@ dependencies = [
  "windows-sys 0.52.0",
 ]
 
+[[package]]
+name = "os_pipe"
+version = "1.2.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "db335f4760b14ead6290116f2427bf33a14d4f0617d49f78a246de10c1831224"
+dependencies = [
+ "libc",
+ "windows-sys 0.59.0",
+]
+
 [[package]]
 name = "owo-colors"
 version = "4.2.2"
@@ -4445,7 +4468,7 @@ checksum = "3af6b589e163c5a788fab00ce0c0366f6efbb9959c2f9874b224936af7fce7e1"
 dependencies = [
  "base64",
  "indexmap 2.12.0",
- "quick-xml",
+ "quick-xml 0.38.0",
  "serde",
  "time",
 ]
@@ -4674,6 +4697,15 @@ version = "2.0.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "a993555f31e5a609f617c12db6250dedcac1b0a85076912c436e6fc9b2c8e6a3"
 
+[[package]]
+name = "quick-xml"
+version = "0.37.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "331e97a1af0bf59823e6eadffe373d7b27f485be8748f71471c662c1f269b7fb"
+dependencies = [
+ "memchr",
+]
+
 [[package]]
 name = "quick-xml"
 version = "0.38.0"
@@ -5008,9 +5040,9 @@ dependencies = [
 
 [[package]]
 name = "rmcp"
-version = "0.8.3"
+version = "0.8.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "1fdad1258f7259fdc0f2dfc266939c82c3b5d1fd72bcde274d600cdc27e60243"
+checksum = "e5947688160b56fb6c827e3c20a72c90392a1d7e9dec74749197aa1780ac42ca"
 dependencies = [
  "base64",
  "bytes",
@@ -5042,9 +5074,9 @@ dependencies = [
 
 [[package]]
 name = "rmcp-macros"
-version = "0.8.3"
+version = "0.8.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ede0589a208cc7ce81d1be68aa7e74b917fcd03c81528408bab0457e187dcd9b"
+checksum = "01263441d3f8635c628e33856c468b96ebbce1af2d3699ea712ca71432d4ee7a"
 dependencies = [
  "darling 0.21.3",
  "proc-macro2",
@@ -5685,6 +5717,12 @@ dependencies = [
  "digest",
 ]
 
+[[package]]
+name = "sha1_smol"
+version = "1.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "bbfa15b3dddfee50a0fff136974b3e1bde555604ba463834a7eb7deb6417705d"
+
 [[package]]
 name = "sha2"
 version = "0.10.9"
@@ -6140,6 +6178,39 @@ version = "0.5.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "8f50febec83f5ee1df3015341d8bd429f2d1cc62bcba7ea2076759d315084683"
 
+[[package]]
+name = "test-case"
+version = "3.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "eb2550dd13afcd286853192af8601920d959b14c401fcece38071d53bf0768a8"
+dependencies = [
+ "test-case-macros",
+]
+
+[[package]]
+name = "test-case-core"
+version = "3.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "adcb7fd841cd518e279be3d5a3eb0636409487998a4aff22f3de87b81e88384f"
+dependencies = [
+ "cfg-if",
+ "proc-macro2",
+ "quote",
+ "syn 2.0.104",
+]
+
+[[package]]
+name = "test-case-macros"
+version = "3.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5c89e72a01ed4c579669add59014b9a524d609c0c88c6a585ce37485879f6ffb"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.104",
+ "test-case-core",
+]
+
 [[package]]
 name = "test-log"
 version = "0.2.18"
@@ -6247,9 +6318,9 @@ dependencies = [
 
 [[package]]
 name = "tiktoken-rs"
-version = "0.7.0"
+version = "0.9.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "25563eeba904d770acf527e8b370fe9a5547bacd20ff84a0b6c3bc41288e5625"
+checksum = "3a19830747d9034cd9da43a60eaa8e552dfda7712424aebf187b7a60126bae0d"
 dependencies = [
  "anyhow",
  "base64",
@@ -6705,6 +6776,18 @@ version = "0.1.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "c4013970217383f67b18aef68f6fb2e8d409bc5755227092d32efb0422ba24b8"
 
+[[package]]
+name = "tree_magic_mini"
+version = "3.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f943391d896cdfe8eec03a04d7110332d445be7df856db382dd96a730667562c"
+dependencies = [
+ "memchr",
+ "nom",
+ "once_cell",
+ "petgraph",
+]
+
 [[package]]
 name = "try-lock"
 version = "0.2.5"
@@ -6872,6 +6955,7 @@ dependencies = [
  "getrandom 0.3.3",
  "js-sys",
  "serde",
+ "sha1_smol",
  "wasm-bindgen",
 ]
 
@@ -7041,6 +7125,76 @@ dependencies = [
  "web-sys",
 ]
 
+[[package]]
+name = "wayland-backend"
+version = "0.3.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "673a33c33048a5ade91a6b139580fa174e19fb0d23f396dca9fa15f2e1e49b35"
+dependencies = [
+ "cc",
+ "downcast-rs",
+ "rustix 1.0.8",
+ "smallvec",
+ "wayland-sys",
+]
+
+[[package]]
+name = "wayland-client"
+version = "0.31.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c66a47e840dc20793f2264eb4b3e4ecb4b75d91c0dd4af04b456128e0bdd449d"
+dependencies = [
+ "bitflags 2.10.0",
+ "rustix 1.0.8",
+ "wayland-backend",
+ "wayland-scanner",
+]
+
+[[package]]
+name = "wayland-protocols"
+version = "0.32.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "efa790ed75fbfd71283bd2521a1cfdc022aabcc28bdcff00851f9e4ae88d9901"
+dependencies = [
+ "bitflags 2.10.0",
+ "wayland-backend",
+ "wayland-client",
+ "wayland-scanner",
+]
+
+[[package]]
+name = "wayland-protocols-wlr"
+version = "0.3.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "efd94963ed43cf9938a090ca4f7da58eb55325ec8200c3848963e98dc25b78ec"
+dependencies = [
+ "bitflags 2.10.0",
+ "wayland-backend",
+ "wayland-client",
+ "wayland-protocols",
+ "wayland-scanner",
+]
+
+[[package]]
+name = "wayland-scanner"
+version = "0.31.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "54cb1e9dc49da91950bdfd8b848c49330536d9d1fb03d4bfec8cae50caa50ae3"
+dependencies = [
+ "proc-macro2",
+ "quick-xml 0.37.5",
+ "quote",
+]
+
+[[package]]
+name = "wayland-sys"
+version = "0.31.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "34949b42822155826b41db8e5d0c1be3a2bd296c747577a43a3e6daefc296142"
+dependencies = [
+ "pkg-config",
+]
+
 [[package]]
 name = "web-sys"
 version = "0.3.77"
@@ -7612,6 +7766,25 @@ dependencies = [
  "bitflags 2.10.0",
 ]
 
+[[package]]
+name = "wl-clipboard-rs"
+version = "0.9.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8e5ff8d0e60065f549fafd9d6cb626203ea64a798186c80d8e7df4f8af56baeb"
+dependencies = [
+ "libc",
+ "log",
+ "os_pipe",
+ "rustix 0.38.44",
+ "tempfile",
+ "thiserror 2.0.17",
+ "tree_magic_mini",
+ "wayland-backend",
+ "wayland-client",
+ "wayland-protocols",
+ "wayland-protocols-wlr",
+]
+
 [[package]]
 name = "writeable"
 version = "0.6.2"
@@ -7780,9 +7953,9 @@ dependencies = [
 
 [[package]]
 name = "zeroize"
-version = "1.8.1"
+version = "1.8.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ced3678a2879b30306d323f4542626697a464a97c0a07c9aebf7ebca65cd4dde"
+checksum = "b97154e67e32c85465826e8bcc1c59429aaaf107c1e4a9e53c8d8ccd5eff88d0"
 dependencies = [
  "zeroize_derive",
 ]

codex-rs/Cargo.toml

@@ -5,6 +5,7 @@ members = [
     "async-utils",
     "app-server",
     "app-server-protocol",
+    "app-server-test-client",
     "apply-patch",
     "arg0",
     "feedback",
@@ -25,7 +26,6 @@ members = [
     "ollama",
     "process-hardening",
     "protocol",
-    "protocol-ts",
     "rmcp-client",
     "responses-api-proxy",
     "stdio-to-uds",
@@ -75,7 +75,6 @@ codex-ollama = { path = "ollama" }
 codex-otel = { path = "otel" }
 codex-process-hardening = { path = "process-hardening" }
 codex-protocol = { path = "protocol" }
-codex-protocol-ts = { path = "protocol-ts" }
 codex-responses-api-proxy = { path = "responses-api-proxy" }
 codex-rmcp-client = { path = "rmcp-client" }
 codex-stdio-to-uds = { path = "stdio-to-uds" }
@@ -87,7 +86,7 @@ codex-utils-pty = { path = "utils/pty" }
 codex-utils-readiness = { path = "utils/readiness" }
 codex-utils-string = { path = "utils/string" }
 codex-utils-tokenizer = { path = "utils/tokenizer" }
-codex-windows-sandbox = { path = "windows-sandbox" }
+codex-windows-sandbox = { path = "windows-sandbox-rs" }
 core_test_support = { path = "core/tests/common" }
 mcp-types = { path = "mcp-types" }
 mcp_test_support = { path = "mcp-server/tests/common" }
@@ -96,8 +95,8 @@ mcp_test_support = { path = "mcp-server/tests/common" }
 allocative = "0.3.3"
 ansi-to-tui = "7.0.0"
 anyhow = "1"
-arboard = "3"
-askama = "0.12"
+arboard = { version = "3", features = ["wayland-data-control"] }
+askama = "0.14"
 assert_cmd = "2"
 assert_matches = "1.5.0"
 async-channel = "2.3.1"
@@ -124,8 +123,8 @@ eventsource-stream = "0.2.3"
 futures = { version = "0.3", default-features = false }
 http = "1.3.1"
 icu_decimal = "2.1"
-icu_provider = { version = "2.1", features = ["sync"] }
 icu_locale_core = "2.1"
+icu_provider = { version = "2.1", features = ["sync"] }
 ignore = "0.4.23"
 image = { version = "^0.25.8", default-features = false }
 indexmap = "2.12.0"
@@ -142,6 +141,7 @@ mime_guess = "2.0.5"
 multimap = "0.10.0"
 notify = "8.2.0"
 nucleo-matcher = "0.3.1"
+once_cell = "1"
 openssl-sys = "*"
 opentelemetry = "0.30.0"
 opentelemetry-appender-tracing = "0.30.0"
@@ -162,7 +162,7 @@ ratatui = "0.29.0"
 ratatui-macros = "0.6.0"
 regex-lite = "0.1.7"
 reqwest = "0.12"
-rmcp = { version = "0.8.3", default-features = false }
+rmcp = { version = "0.8.5", default-features = false }
 schemars = "0.8.22"
 seccompiler = "0.5.0"
 sentry = "0.34.0"
@@ -183,6 +183,7 @@ tempfile = "3.23.0"
 test-log = "0.2.18"
 textwrap = "0.16.2"
 thiserror = "2.0.17"
+tiktoken-rs = "0.9"
 time = "0.3"
 tiny_http = "0.12"
 tokio = "1"
@@ -213,7 +214,7 @@ which = "6"
 wildmatch = "2.5.0"
 
 wiremock = "0.6"
-zeroize = "1.8.1"
+zeroize = "1.8.2"
 
 [workspace.lints]
 rust = {}
@@ -256,7 +257,12 @@ unwrap_used = "deny"
 # cargo-shear cannot see the platform-specific openssl-sys usage, so we
 # silence the false positive here instead of deleting a real dependency.
 [workspace.metadata.cargo-shear]
-ignored = ["icu_provider", "openssl-sys", "codex-utils-readiness", "codex-utils-tokenizer"]
+ignored = [
+    "icu_provider",
+    "openssl-sys",
+    "codex-utils-readiness",
+    "codex-utils-tokenizer",
+]
 
 [profile.release]
 lto = "fat"
@@ -275,6 +281,7 @@ opt-level = 0
 [patch.crates-io]
 # Uncomment to debug local changes.
 # ratatui = { path = "../../ratatui" }
+crossterm = { git = "https://github.com/nornagon/crossterm", branch = "nornagon/color-query" }
 ratatui = { git = "https://github.com/nornagon/ratatui", branch = "nornagon-v0.29.0-patch" }
 
 # Uncomment to debug local changes.

codex-rs/README.md

@@ -58,7 +58,7 @@ To test to see what happens when a command is run under the sandbox provided by
 
 ```
 # macOS
-codex sandbox macos [--full-auto] [COMMAND]...
+codex sandbox macos [--full-auto] [--log-denials] [COMMAND]...
 
 # Linux
 codex sandbox linux [--full-auto] [COMMAND]...
@@ -67,7 +67,7 @@ codex sandbox linux [--full-auto] [COMMAND]...
 codex sandbox windows [--full-auto] [COMMAND]...
 
 # Legacy aliases
-codex debug seatbelt [--full-auto] [COMMAND]...
+codex debug seatbelt [--full-auto] [--log-denials] [COMMAND]...
 codex debug landlock [--full-auto] [COMMAND]...
 ```
 

codex-rs/app-server-protocol/Cargo.toml

@@ -14,6 +14,7 @@ workspace = true
 anyhow = { workspace = true }
 clap = { workspace = true, features = ["derive"] }
 codex-protocol = { workspace = true }
+mcp-types = { workspace = true }
 paste = { workspace = true }
 schemars = { workspace = true }
 serde = { workspace = true, features = ["derive"] }

codex-rs/app-server-protocol/src/export.rs

@@ -2,20 +2,24 @@ use crate::ClientNotification;
 use crate::ClientRequest;
 use crate::ServerNotification;
 use crate::ServerRequest;
+use crate::export_client_notification_schemas;
+use crate::export_client_param_schemas;
 use crate::export_client_response_schemas;
 use crate::export_client_responses;
+use crate::export_server_notification_schemas;
+use crate::export_server_param_schemas;
 use crate::export_server_response_schemas;
 use crate::export_server_responses;
 use anyhow::Context;
 use anyhow::Result;
 use anyhow::anyhow;
+use codex_protocol::protocol::EventMsg;
 use schemars::JsonSchema;
-use schemars::schema::RootSchema;
 use schemars::schema_for;
 use serde::Serialize;
 use serde_json::Map;
 use serde_json::Value;
-use std::collections::BTreeMap;
+use std::collections::HashMap;
 use std::collections::HashSet;
 use std::ffi::OsStr;
 use std::fs;
@@ -28,83 +32,29 @@ use ts_rs::TS;
 
 const HEADER: &str = "// GENERATED CODE! DO NOT MODIFY BY HAND!\n\n";
 
-macro_rules! for_each_schema_type {
-    ($macro:ident) => {
-        $macro!(crate::RequestId);
-        $macro!(crate::JSONRPCMessage);
-        $macro!(crate::JSONRPCRequest);
-        $macro!(crate::JSONRPCNotification);
-        $macro!(crate::JSONRPCResponse);
-        $macro!(crate::JSONRPCError);
-        $macro!(crate::JSONRPCErrorError);
-        $macro!(crate::AddConversationListenerParams);
-        $macro!(crate::AddConversationSubscriptionResponse);
-        $macro!(crate::ApplyPatchApprovalParams);
-        $macro!(crate::ApplyPatchApprovalResponse);
-        $macro!(crate::ArchiveConversationParams);
-        $macro!(crate::ArchiveConversationResponse);
-        $macro!(crate::AuthMode);
-        $macro!(crate::AuthStatusChangeNotification);
-        $macro!(crate::CancelLoginChatGptParams);
-        $macro!(crate::CancelLoginChatGptResponse);
-        $macro!(crate::ClientInfo);
-        $macro!(crate::ClientNotification);
-        $macro!(crate::ClientRequest);
-        $macro!(crate::ConversationSummary);
-        $macro!(crate::ExecCommandApprovalParams);
-        $macro!(crate::ExecCommandApprovalResponse);
-        $macro!(crate::ExecOneOffCommandParams);
-        $macro!(crate::ExecOneOffCommandResponse);
-        $macro!(crate::FuzzyFileSearchParams);
-        $macro!(crate::FuzzyFileSearchResponse);
-        $macro!(crate::FuzzyFileSearchResult);
-        $macro!(crate::GetAuthStatusParams);
-        $macro!(crate::GetAuthStatusResponse);
-        $macro!(crate::GetUserAgentResponse);
-        $macro!(crate::GetUserSavedConfigResponse);
-        $macro!(crate::GitDiffToRemoteParams);
-        $macro!(crate::GitDiffToRemoteResponse);
-        $macro!(crate::GitSha);
-        $macro!(crate::InitializeParams);
-        $macro!(crate::InitializeResponse);
-        $macro!(crate::InputItem);
-        $macro!(crate::InterruptConversationParams);
-        $macro!(crate::InterruptConversationResponse);
-        $macro!(crate::ListConversationsParams);
-        $macro!(crate::ListConversationsResponse);
-        $macro!(crate::LoginApiKeyParams);
-        $macro!(crate::LoginApiKeyResponse);
-        $macro!(crate::LoginChatGptCompleteNotification);
-        $macro!(crate::LoginChatGptResponse);
-        $macro!(crate::LogoutChatGptParams);
-        $macro!(crate::LogoutChatGptResponse);
-        $macro!(crate::NewConversationParams);
-        $macro!(crate::NewConversationResponse);
-        $macro!(crate::Profile);
-        $macro!(crate::RemoveConversationListenerParams);
-        $macro!(crate::RemoveConversationSubscriptionResponse);
-        $macro!(crate::ResumeConversationParams);
-        $macro!(crate::ResumeConversationResponse);
-        $macro!(crate::SandboxSettings);
-        $macro!(crate::SendUserMessageParams);
-        $macro!(crate::SendUserMessageResponse);
-        $macro!(crate::SendUserTurnParams);
-        $macro!(crate::SendUserTurnResponse);
-        $macro!(crate::ServerNotification);
-        $macro!(crate::ServerRequest);
-        $macro!(crate::SessionConfiguredNotification);
-        $macro!(crate::SetDefaultModelParams);
-        $macro!(crate::SetDefaultModelResponse);
-        $macro!(crate::Tools);
-        $macro!(crate::UserInfoResponse);
-        $macro!(crate::UserSavedConfig);
-        $macro!(codex_protocol::protocol::EventMsg);
-        $macro!(codex_protocol::protocol::FileChange);
-        $macro!(codex_protocol::parse_command::ParsedCommand);
-        $macro!(codex_protocol::protocol::SandboxPolicy);
-    };
+#[derive(Clone)]
+pub struct GeneratedSchema {
+    namespace: Option<String>,
+    logical_name: String,
+    value: Value,
+    in_v1_dir: bool,
 }
 
+impl GeneratedSchema {
+    fn namespace(&self) -> Option<&str> {
+        self.namespace.as_deref()
+    }
+
+    fn logical_name(&self) -> &str {
+        &self.logical_name
+    }
+
+    fn value(&self) -> &Value {
+        &self.value
+    }
+}
+
+type JsonSchemaEmitter = fn(&Path) -> Result<GeneratedSchema>;
 pub fn generate_types(out_dir: &Path, prettier: Option<&Path>) -> Result<()> {
     generate_ts(out_dir, prettier)?;
     generate_json(out_dir)?;
@@ -112,7 +62,9 @@ pub fn generate_types(out_dir: &Path, prettier: Option<&Path>) -> Result<()> {
 }
 
 pub fn generate_ts(out_dir: &Path, prettier: Option<&Path>) -> Result<()> {
+    let v2_out_dir = out_dir.join("v2");
     ensure_dir(out_dir)?;
+    ensure_dir(&v2_out_dir)?;
 
     ClientRequest::export_all_to(out_dir)?;
     export_client_responses(out_dir)?;
@@ -123,17 +75,22 @@ pub fn generate_ts(out_dir: &Path, prettier: Option<&Path>) -> Result<()> {
     ServerNotification::export_all_to(out_dir)?;
 
     generate_index_ts(out_dir)?;
+    generate_index_ts(&v2_out_dir)?;
 
-    let ts_files = ts_files_in(out_dir)?;
+    // Ensure our header is present on all TS files (root + subdirs like v2/).
+    let ts_files = ts_files_in_recursive(out_dir)?;
     for file in &ts_files {
         prepend_header_if_missing(file)?;
     }
 
+    // Optionally run Prettier on all generated TS files.
     if let Some(prettier_bin) = prettier
         && !ts_files.is_empty()
     {
         let status = Command::new(prettier_bin)
             .arg("--write")
+            .arg("--log-level")
+            .arg("warn")
             .args(ts_files.iter().map(|p| p.as_os_str()))
             .status()
             .with_context(|| format!("Failed to invoke Prettier at {}", prettier_bin.display()))?;
@@ -147,51 +104,115 @@ pub fn generate_ts(out_dir: &Path, prettier: Option<&Path>) -> Result<()> {
 
 pub fn generate_json(out_dir: &Path) -> Result<()> {
     ensure_dir(out_dir)?;
-    let mut bundle: BTreeMap<String, RootSchema> = BTreeMap::new();
+    let envelope_emitters: &[JsonSchemaEmitter] = &[
+        |d| write_json_schema_with_return::<crate::RequestId>(d, "RequestId"),
+        |d| write_json_schema_with_return::<crate::JSONRPCMessage>(d, "JSONRPCMessage"),
+        |d| write_json_schema_with_return::<crate::JSONRPCRequest>(d, "JSONRPCRequest"),
+        |d| write_json_schema_with_return::<crate::JSONRPCNotification>(d, "JSONRPCNotification"),
+        |d| write_json_schema_with_return::<crate::JSONRPCResponse>(d, "JSONRPCResponse"),
+        |d| write_json_schema_with_return::<crate::JSONRPCError>(d, "JSONRPCError"),
+        |d| write_json_schema_with_return::<crate::JSONRPCErrorError>(d, "JSONRPCErrorError"),
+        |d| write_json_schema_with_return::<crate::ClientRequest>(d, "ClientRequest"),
+        |d| write_json_schema_with_return::<crate::ServerRequest>(d, "ServerRequest"),
+        |d| write_json_schema_with_return::<crate::ClientNotification>(d, "ClientNotification"),
+        |d| write_json_schema_with_return::<crate::ServerNotification>(d, "ServerNotification"),
+        |d| write_json_schema_with_return::<EventMsg>(d, "EventMsg"),
+    ];
 
-    macro_rules! add_schema {
-        ($ty:path) => {{
-            let name = type_basename(stringify!($ty));
-            let schema = write_json_schema_with_return::<$ty>(out_dir, &name)?;
-            bundle.insert(name, schema);
-        }};
+    let mut schemas: Vec<GeneratedSchema> = Vec::new();
+    for emit in envelope_emitters {
+        schemas.push(emit(out_dir)?);
     }
 
-    for_each_schema_type!(add_schema);
+    schemas.extend(export_client_param_schemas(out_dir)?);
+    schemas.extend(export_client_response_schemas(out_dir)?);
+    schemas.extend(export_server_param_schemas(out_dir)?);
+    schemas.extend(export_server_response_schemas(out_dir)?);
+    schemas.extend(export_client_notification_schemas(out_dir)?);
+    schemas.extend(export_server_notification_schemas(out_dir)?);
 
-    export_client_response_schemas(out_dir)?;
-    export_server_response_schemas(out_dir)?;
+    let bundle = build_schema_bundle(schemas)?;
+    write_pretty_json(
+        out_dir.join("codex_app_server_protocol.schemas.json"),
+        &bundle,
+    )?;
 
-    let mut definitions = Map::new();
+    Ok(())
+}
 
+fn build_schema_bundle(schemas: Vec<GeneratedSchema>) -> Result<Value> {
     const SPECIAL_DEFINITIONS: &[&str] = &[
         "ClientNotification",
         "ClientRequest",
         "EventMsg",
-        "FileChange",
-        "InputItem",
-        "ParsedCommand",
-        "SandboxPolicy",
         "ServerNotification",
         "ServerRequest",
     ];
+    const IGNORED_DEFINITIONS: &[&str] = &["Option<()>"];
 
-    for (name, schema) in bundle {
-        let mut schema_value = serde_json::to_value(schema)?;
-        annotate_schema(&mut schema_value, Some(name.as_str()));
+    let namespaced_types = collect_namespaced_types(&schemas);
+    let mut definitions = Map::new();
 
-        if let Value::Object(ref mut obj) = schema_value
+    for schema in schemas {
+        let GeneratedSchema {
+            namespace,
+            logical_name,
+            mut value,
+            in_v1_dir,
+        } = schema;
+
+        if IGNORED_DEFINITIONS.contains(&logical_name.as_str()) {
+            continue;
+        }
+
+        if let Some(ref ns) = namespace {
+            rewrite_refs_to_namespace(&mut value, ns);
+        }
+
+        let mut forced_namespace_refs: Vec<(String, String)> = Vec::new();
+        if let Value::Object(ref mut obj) = value
             && let Some(defs) = obj.remove("definitions")
             && let Value::Object(defs_obj) = defs
         {
             for (def_name, mut def_schema) in defs_obj {
-                if !SPECIAL_DEFINITIONS.contains(&def_name.as_str()) {
-                    annotate_schema(&mut def_schema, Some(def_name.as_str()));
+                if IGNORED_DEFINITIONS.contains(&def_name.as_str()) {
+                    continue;
+                }
+                if SPECIAL_DEFINITIONS.contains(&def_name.as_str()) {
+                    continue;
+                }
+                annotate_schema(&mut def_schema, Some(def_name.as_str()));
+                let target_namespace = match namespace {
+                    Some(ref ns) => Some(ns.clone()),
+                    None => namespace_for_definition(&def_name, &namespaced_types)
+                        .cloned()
+                        .filter(|_| !in_v1_dir),
+                };
+                if let Some(ref ns) = target_namespace {
+                    if namespace.as_deref() == Some(ns.as_str()) {
+                        rewrite_refs_to_namespace(&mut def_schema, ns);
+                        insert_into_namespace(&mut definitions, ns, def_name.clone(), def_schema)?;
+                    } else if !forced_namespace_refs
+                        .iter()
+                        .any(|(name, existing_ns)| name == &def_name && existing_ns == ns)
+                    {
+                        forced_namespace_refs.push((def_name.clone(), ns.clone()));
+                    }
+                } else {
                     definitions.insert(def_name, def_schema);
                 }
             }
         }
-        definitions.insert(name, schema_value);
+
+        for (name, ns) in forced_namespace_refs {
+            rewrite_named_ref_to_namespace(&mut value, &ns, &name);
+        }
+
+        if let Some(ref ns) = namespace {
+            insert_into_namespace(&mut definitions, ns, logical_name.clone(), value)?;
+        } else {
+            definitions.insert(logical_name, value);
+        }
     }
 
     let mut root = Map::new();
@@ -206,15 +227,28 @@ pub fn generate_json(out_dir: &Path) -> Result<()> {
     root.insert("type".to_string(), Value::String("object".into()));
     root.insert("definitions".to_string(), Value::Object(definitions));
 
-    write_pretty_json(
-        out_dir.join("codex_app_server_protocol.schemas.json"),
-        &Value::Object(root),
-    )?;
-
-    Ok(())
+    Ok(Value::Object(root))
 }
 
-fn write_json_schema_with_return<T>(out_dir: &Path, name: &str) -> Result<RootSchema>
+fn insert_into_namespace(
+    definitions: &mut Map<String, Value>,
+    namespace: &str,
+    name: String,
+    schema: Value,
+) -> Result<()> {
+    let entry = definitions
+        .entry(namespace.to_string())
+        .or_insert_with(|| Value::Object(Map::new()));
+    match entry {
+        Value::Object(map) => {
+            map.insert(name, schema);
+            Ok(())
+        }
+        _ => Err(anyhow!("expected namespace {namespace} to be an object")),
+    }
+}
+
+fn write_json_schema_with_return<T>(out_dir: &Path, name: &str) -> Result<GeneratedSchema>
 where
     T: JsonSchema,
 {
@@ -222,17 +256,37 @@ where
     let schema = schema_for!(T);
     let mut schema_value = serde_json::to_value(schema)?;
     annotate_schema(&mut schema_value, Some(file_stem));
-    write_pretty_json(out_dir.join(format!("{file_stem}.json")), &schema_value)
+    // If the name looks like a namespaced path (e.g., "v2::Type"), mirror
+    // the TypeScript layout and write to out_dir/v2/Type.json. Otherwise
+    // write alongside the legacy files.
+    let (raw_namespace, logical_name) = split_namespace(file_stem);
+    let out_path = if let Some(ns) = raw_namespace {
+        let dir = out_dir.join(ns);
+        ensure_dir(&dir)?;
+        dir.join(format!("{logical_name}.json"))
+    } else {
+        out_dir.join(format!("{file_stem}.json"))
+    };
+
+    write_pretty_json(out_path, &schema_value)
         .with_context(|| format!("Failed to write JSON schema for {file_stem}"))?;
-    let annotated_schema = serde_json::from_value(schema_value)?;
-    Ok(annotated_schema)
+    let namespace = match raw_namespace {
+        Some("v1") | None => None,
+        Some(ns) => Some(ns.to_string()),
+    };
+    Ok(GeneratedSchema {
+        in_v1_dir: raw_namespace == Some("v1"),
+        namespace,
+        logical_name: logical_name.to_string(),
+        value: schema_value,
+    })
 }
 
-pub(crate) fn write_json_schema<T>(out_dir: &Path, name: &str) -> Result<()>
+pub(crate) fn write_json_schema<T>(out_dir: &Path, name: &str) -> Result<GeneratedSchema>
 where
     T: JsonSchema,
 {
-    write_json_schema_with_return::<T>(out_dir, name).map(|_| ())
+    write_json_schema_with_return::<T>(out_dir, name)
 }
 
 fn write_pretty_json(path: PathBuf, value: &impl Serialize) -> Result<()> {
@@ -241,13 +295,73 @@ fn write_pretty_json(path: PathBuf, value: &impl Serialize) -> Result<()> {
     fs::write(&path, json).with_context(|| format!("Failed to write {}", path.display()))?;
     Ok(())
 }
-fn type_basename(type_path: &str) -> String {
-    type_path
-        .rsplit_once("::")
-        .map(|(_, name)| name)
-        .unwrap_or(type_path)
-        .trim()
-        .to_string()
+
+/// Split a fully-qualified type name like "v2::Type" into its namespace and logical name.
+fn split_namespace(name: &str) -> (Option<&str>, &str) {
+    name.split_once("::")
+        .map_or((None, name), |(ns, rest)| (Some(ns), rest))
+}
+
+/// Recursively rewrite $ref values that point at "#/definitions/..." so that
+/// they point to a namespaced location under the bundle.
+fn rewrite_refs_to_namespace(value: &mut Value, ns: &str) {
+    match value {
+        Value::Object(obj) => {
+            if let Some(Value::String(r)) = obj.get_mut("$ref")
+                && let Some(suffix) = r.strip_prefix("#/definitions/")
+            {
+                let prefix = format!("{ns}/");
+                if !suffix.starts_with(&prefix) {
+                    *r = format!("#/definitions/{ns}/{suffix}");
+                }
+            }
+            for v in obj.values_mut() {
+                rewrite_refs_to_namespace(v, ns);
+            }
+        }
+        Value::Array(items) => {
+            for v in items.iter_mut() {
+                rewrite_refs_to_namespace(v, ns);
+            }
+        }
+        _ => {}
+    }
+}
+
+fn collect_namespaced_types(schemas: &[GeneratedSchema]) -> HashMap<String, String> {
+    let mut types = HashMap::new();
+    for schema in schemas {
+        if let Some(ns) = schema.namespace() {
+            types
+                .entry(schema.logical_name().to_string())
+                .or_insert_with(|| ns.to_string());
+            if let Some(Value::Object(defs)) = schema.value().get("definitions") {
+                for key in defs.keys() {
+                    types.entry(key.clone()).or_insert_with(|| ns.to_string());
+                }
+            }
+            if let Some(Value::Object(defs)) = schema.value().get("$defs") {
+                for key in defs.keys() {
+                    types.entry(key.clone()).or_insert_with(|| ns.to_string());
+                }
+            }
+        }
+    }
+    types
+}
+
+fn namespace_for_definition<'a>(
+    name: &str,
+    types: &'a HashMap<String, String>,
+) -> Option<&'a String> {
+    if let Some(ns) = types.get(name) {
+        return Some(ns);
+    }
+    let trimmed = name.trim_end_matches(|c: char| c.is_ascii_digit());
+    if trimmed != name {
+        return types.get(trimmed);
+    }
+    None
 }
 
 fn variant_definition_name(base: &str, variant: &Value) -> Option<String> {
@@ -269,14 +383,6 @@ fn variant_definition_name(base: &str, variant: &Value) -> Option<String> {
             });
         }
 
-        if let Some(mode_literal) = literal_from_property(props, "mode") {
-            let pascal = to_pascal_case(mode_literal);
-            return Some(match base {
-                "SandboxPolicy" => format!("{pascal}SandboxPolicy"),
-                _ => format!("{pascal}{base}"),
-            });
-        }
-
         if props.len() == 1
             && let Some(key) = props.keys().next()
         {
@@ -467,6 +573,33 @@ fn ensure_dir(dir: &Path) -> Result<()> {
         .with_context(|| format!("Failed to create output directory {}", dir.display()))
 }
 
+fn rewrite_named_ref_to_namespace(value: &mut Value, ns: &str, name: &str) {
+    let direct = format!("#/definitions/{name}");
+    let prefixed = format!("{direct}/");
+    let replacement = format!("#/definitions/{ns}/{name}");
+    let replacement_prefixed = format!("{replacement}/");
+    match value {
+        Value::Object(obj) => {
+            if let Some(Value::String(reference)) = obj.get_mut("$ref") {
+                if reference == &direct {
+                    *reference = replacement;
+                } else if let Some(rest) = reference.strip_prefix(&prefixed) {
+                    *reference = format!("{replacement_prefixed}{rest}");
+                }
+            }
+            for child in obj.values_mut() {
+                rewrite_named_ref_to_namespace(child, ns, name);
+            }
+        }
+        Value::Array(items) => {
+            for child in items {
+                rewrite_named_ref_to_namespace(child, ns, name);
+            }
+        }
+        _ => {}
+    }
+}
+
 fn prepend_header_if_missing(path: &Path) -> Result<()> {
     let mut content = String::new();
     {
@@ -504,6 +637,28 @@ fn ts_files_in(dir: &Path) -> Result<Vec<PathBuf>> {
     Ok(files)
 }
 
+fn ts_files_in_recursive(dir: &Path) -> Result<Vec<PathBuf>> {
+    let mut files = Vec::new();
+    let mut stack = vec![dir.to_path_buf()];
+    while let Some(d) = stack.pop() {
+        for entry in
+            fs::read_dir(&d).with_context(|| format!("Failed to read dir {}", d.display()))?
+        {
+            let entry = entry?;
+            let path = entry.path();
+            if path.is_dir() {
+                stack.push(path);
+            } else if path.is_file() && path.extension() == Some(OsStr::new("ts")) {
+                files.push(path);
+            }
+        }
+    }
+    files.sort();
+    Ok(files)
+}
+
+/// Generate an index.ts file that re-exports all generated types.
+/// This allows consumers to import all types from a single file.
 fn generate_index_ts(out_dir: &Path) -> Result<PathBuf> {
     let mut entries: Vec<String> = Vec::new();
     let mut stems: Vec<String> = ts_files_in(out_dir)?
@@ -520,6 +675,14 @@ fn generate_index_ts(out_dir: &Path) -> Result<PathBuf> {
         entries.push(format!("export type {{ {name} }} from \"./{name}\";\n"));
     }
 
+    // If this is the root out_dir and a ./v2 folder exists with TS files,
+    // expose it as a namespace to avoid symbol collisions at the root.
+    let v2_dir = out_dir.join("v2");
+    let has_v2_ts = ts_files_in(&v2_dir).map(|v| !v.is_empty()).unwrap_or(false);
+    if has_v2_ts {
+        entries.push("export * as v2 from \"./v2\";\n".to_string());
+    }
+
     let mut content =
         String::with_capacity(HEADER.len() + entries.iter().map(String::len).sum::<usize>());
     content.push_str(HEADER);
@@ -546,6 +709,7 @@ mod tests {
 
     #[test]
     fn generated_ts_has_no_optional_nullable_fields() -> Result<()> {
+        // Assert that there are no types of the form "?: T | null" in the generated TS files.
         let output_dir = std::env::temp_dir().join(format!("codex_ts_types_{}", Uuid::now_v7()));
         fs::create_dir(&output_dir)?;
 

codex-rs/app-server-protocol/src/protocol/common.rs

@@ -1,15 +1,17 @@
 use std::collections::HashMap;
+use std::path::Path;
 use std::path::PathBuf;
 
 use crate::JSONRPCNotification;
 use crate::JSONRPCRequest;
 use crate::RequestId;
+use crate::export::GeneratedSchema;
+use crate::export::write_json_schema;
 use crate::protocol::v1;
 use crate::protocol::v2;
 use codex_protocol::ConversationId;
 use codex_protocol::parse_command::ParsedCommand;
 use codex_protocol::protocol::FileChange;
-use codex_protocol::protocol::RateLimitSnapshot;
 use codex_protocol::protocol::ReviewDecision;
 use codex_protocol::protocol::SandboxCommandAssessment;
 use paste::paste;
@@ -44,7 +46,7 @@ macro_rules! client_request_definitions {
     (
         $(
             $(#[$variant_meta:meta])*
-            $variant:ident {
+            $variant:ident $(=> $wire:literal)? {
                 params: $(#[$params_meta:meta])* $params:ty,
                 response: $response:ty,
             }
@@ -56,6 +58,7 @@ macro_rules! client_request_definitions {
         pub enum ClientRequest {
             $(
                 $(#[$variant_meta])*
+                $(#[serde(rename = $wire)] #[ts(rename = $wire)])?
                 $variant {
                     #[serde(rename = "id")]
                     request_id: RequestId,
@@ -74,66 +77,103 @@ macro_rules! client_request_definitions {
             Ok(())
         }
 
+        #[allow(clippy::vec_init_then_push)]
         pub fn export_client_response_schemas(
             out_dir: &::std::path::Path,
-        ) -> ::anyhow::Result<()> {
+        ) -> ::anyhow::Result<Vec<GeneratedSchema>> {
+            let mut schemas = Vec::new();
             $(
-                crate::export::write_json_schema::<$response>(out_dir, stringify!($response))?;
+                schemas.push(write_json_schema::<$response>(out_dir, stringify!($response))?);
             )*
-            Ok(())
+            Ok(schemas)
+        }
+
+        #[allow(clippy::vec_init_then_push)]
+        pub fn export_client_param_schemas(
+            out_dir: &::std::path::Path,
+        ) -> ::anyhow::Result<Vec<GeneratedSchema>> {
+            let mut schemas = Vec::new();
+            $(
+                schemas.push(write_json_schema::<$params>(out_dir, stringify!($params))?);
+            )*
+            Ok(schemas)
         }
     };
 }
 
 client_request_definitions! {
-    /// NEW APIs
-    #[serde(rename = "model/list")]
-    #[ts(rename = "model/list")]
-    ListModels {
-        params: v2::ListModelsParams,
-        response: v2::ListModelsResponse,
-    },
-
-    #[serde(rename = "account/login")]
-    #[ts(rename = "account/login")]
-    LoginAccount {
-        params: v2::LoginAccountParams,
-        response: v2::LoginAccountResponse,
-    },
-
-    #[serde(rename = "account/logout")]
-    #[ts(rename = "account/logout")]
-    LogoutAccount {
-        params: #[ts(type = "undefined")] #[serde(skip_serializing_if = "Option::is_none")] Option<()>,
-        response: v2::LogoutAccountResponse,
-    },
-
-    #[serde(rename = "account/rateLimits/read")]
-    #[ts(rename = "account/rateLimits/read")]
-    GetAccountRateLimits {
-        params: #[ts(type = "undefined")] #[serde(skip_serializing_if = "Option::is_none")] Option<()>,
-        response: v2::GetAccountRateLimitsResponse,
-    },
-
-    #[serde(rename = "feedback/upload")]
-    #[ts(rename = "feedback/upload")]
-    UploadFeedback {
-        params: v2::UploadFeedbackParams,
-        response: v2::UploadFeedbackResponse,
-    },
-
-    #[serde(rename = "account/read")]
-    #[ts(rename = "account/read")]
-    GetAccount {
-        params: #[ts(type = "undefined")] #[serde(skip_serializing_if = "Option::is_none")] Option<()>,
-        response: v2::GetAccountResponse,
-    },
-
-    /// DEPRECATED APIs below
     Initialize {
         params: v1::InitializeParams,
         response: v1::InitializeResponse,
     },
+
+    /// NEW APIs
+    // Thread lifecycle
+    ThreadStart => "thread/start" {
+        params: v2::ThreadStartParams,
+        response: v2::ThreadStartResponse,
+    },
+    ThreadResume => "thread/resume" {
+        params: v2::ThreadResumeParams,
+        response: v2::ThreadResumeResponse,
+    },
+    ThreadArchive => "thread/archive" {
+        params: v2::ThreadArchiveParams,
+        response: v2::ThreadArchiveResponse,
+    },
+    ThreadList => "thread/list" {
+        params: v2::ThreadListParams,
+        response: v2::ThreadListResponse,
+    },
+    ThreadCompact => "thread/compact" {
+        params: v2::ThreadCompactParams,
+        response: v2::ThreadCompactResponse,
+    },
+    TurnStart => "turn/start" {
+        params: v2::TurnStartParams,
+        response: v2::TurnStartResponse,
+    },
+    TurnInterrupt => "turn/interrupt" {
+        params: v2::TurnInterruptParams,
+        response: v2::TurnInterruptResponse,
+    },
+
+    ModelList => "model/list" {
+        params: v2::ModelListParams,
+        response: v2::ModelListResponse,
+    },
+
+    LoginAccount => "account/login/start" {
+        params: v2::LoginAccountParams,
+        response: v2::LoginAccountResponse,
+    },
+
+    CancelLoginAccount => "account/login/cancel" {
+        params: v2::CancelLoginAccountParams,
+        response: v2::CancelLoginAccountResponse,
+    },
+
+    LogoutAccount => "account/logout" {
+        params: #[ts(type = "undefined")] #[serde(skip_serializing_if = "Option::is_none")] Option<()>,
+        response: v2::LogoutAccountResponse,
+    },
+
+    GetAccountRateLimits => "account/rateLimits/read" {
+        params: #[ts(type = "undefined")] #[serde(skip_serializing_if = "Option::is_none")] Option<()>,
+        response: v2::GetAccountRateLimitsResponse,
+    },
+
+    FeedbackUpload => "feedback/upload" {
+        params: v2::FeedbackUploadParams,
+        response: v2::FeedbackUploadResponse,
+    },
+
+    GetAccount => "account/read" {
+        params: v2::GetAccountParams,
+        response: v2::GetAccountResponse,
+    },
+
+    /// DEPRECATED APIs below
     NewConversation {
         params: v1::NewConversationParams,
         response: v1::NewConversationResponse,
@@ -188,6 +228,7 @@ client_request_definitions! {
         params: #[ts(type = "undefined")] #[serde(skip_serializing_if = "Option::is_none")] Option<()>,
         response: v1::LoginChatGptResponse,
     },
+    // DEPRECATED in favor of CancelLoginAccount
     CancelLoginChatGpt {
         params: v1::CancelLoginChatGptParams,
         response: v1::CancelLoginChatGptResponse,
@@ -196,6 +237,7 @@ client_request_definitions! {
         params: #[ts(type = "undefined")] #[serde(skip_serializing_if = "Option::is_none")] Option<()>,
         response: v1::LogoutChatGptResponse,
     },
+    /// DEPRECATED in favor of GetAccount
     GetAuthStatus {
         params: v1::GetAuthStatusParams,
         response: v1::GetAuthStatusResponse,
@@ -276,13 +318,101 @@ macro_rules! server_request_definitions {
             Ok(())
         }
 
+        #[allow(clippy::vec_init_then_push)]
         pub fn export_server_response_schemas(
-            out_dir: &::std::path::Path,
-        ) -> ::anyhow::Result<()> {
+            out_dir: &Path,
+        ) -> ::anyhow::Result<Vec<GeneratedSchema>> {
+            let mut schemas = Vec::new();
             paste! {
-                $(crate::export::write_json_schema::<[<$variant Response>]>(out_dir, stringify!([<$variant Response>]))?;)*
+                $(schemas.push(crate::export::write_json_schema::<[<$variant Response>]>(out_dir, stringify!([<$variant Response>]))?);)*
             }
-            Ok(())
+            Ok(schemas)
+        }
+
+        #[allow(clippy::vec_init_then_push)]
+        pub fn export_server_param_schemas(
+            out_dir: &Path,
+        ) -> ::anyhow::Result<Vec<GeneratedSchema>> {
+            let mut schemas = Vec::new();
+            paste! {
+                $(schemas.push(crate::export::write_json_schema::<[<$variant Params>]>(out_dir, stringify!([<$variant Params>]))?);)*
+            }
+            Ok(schemas)
+        }
+    };
+}
+
+/// Generates `ServerNotification` enum and helpers, including a JSON Schema
+/// exporter for each notification.
+macro_rules! server_notification_definitions {
+    (
+        $(
+            $(#[$variant_meta:meta])*
+            $variant:ident $(=> $wire:literal)? ( $payload:ty )
+        ),* $(,)?
+    ) => {
+        /// Notification sent from the server to the client.
+        #[derive(Serialize, Deserialize, Debug, Clone, JsonSchema, TS, Display)]
+        #[serde(tag = "method", content = "params", rename_all = "camelCase")]
+        #[strum(serialize_all = "camelCase")]
+        pub enum ServerNotification {
+            $(
+                $(#[$variant_meta])*
+                $(#[serde(rename = $wire)] #[ts(rename = $wire)] #[strum(serialize = $wire)])?
+                $variant($payload),
+            )*
+        }
+
+        impl ServerNotification {
+            pub fn to_params(self) -> Result<serde_json::Value, serde_json::Error> {
+                match self {
+                    $(Self::$variant(params) => serde_json::to_value(params),)*
+                }
+            }
+        }
+
+        impl TryFrom<JSONRPCNotification> for ServerNotification {
+            type Error = serde_json::Error;
+
+            fn try_from(value: JSONRPCNotification) -> Result<Self, Self::Error> {
+                serde_json::from_value(serde_json::to_value(value)?)
+            }
+        }
+
+        #[allow(clippy::vec_init_then_push)]
+        pub fn export_server_notification_schemas(
+            out_dir: &::std::path::Path,
+        ) -> ::anyhow::Result<Vec<GeneratedSchema>> {
+            let mut schemas = Vec::new();
+            $(schemas.push(crate::export::write_json_schema::<$payload>(out_dir, stringify!($payload))?);)*
+            Ok(schemas)
+        }
+    };
+}
+/// Notifications sent from the client to the server.
+macro_rules! client_notification_definitions {
+    (
+        $(
+            $(#[$variant_meta:meta])*
+            $variant:ident $( ( $payload:ty ) )?
+        ),* $(,)?
+    ) => {
+        #[derive(Serialize, Deserialize, Debug, Clone, JsonSchema, TS, Display)]
+        #[serde(tag = "method", content = "params", rename_all = "camelCase")]
+        #[strum(serialize_all = "camelCase")]
+        pub enum ClientNotification {
+            $(
+                $(#[$variant_meta])*
+                $variant $( ( $payload ) )?,
+            )*
+        }
+
+        pub fn export_client_notification_schemas(
+            _out_dir: &::std::path::Path,
+        ) -> ::anyhow::Result<Vec<GeneratedSchema>> {
+            let schemas = Vec::new();
+            $( $(schemas.push(crate::export::write_json_schema::<$payload>(_out_dir, stringify!($payload))?);)? )*
+            Ok(schemas)
         }
     };
 }
@@ -366,52 +496,36 @@ pub struct FuzzyFileSearchResponse {
     pub files: Vec<FuzzyFileSearchResult>,
 }
 
-/// Notification sent from the server to the client.
-#[derive(Serialize, Deserialize, Debug, Clone, JsonSchema, TS, Display)]
-#[serde(tag = "method", content = "params", rename_all = "camelCase")]
-#[strum(serialize_all = "camelCase")]
-pub enum ServerNotification {
+server_notification_definitions! {
     /// NEW NOTIFICATIONS
-    #[serde(rename = "account/rateLimits/updated")]
-    #[ts(rename = "account/rateLimits/updated")]
-    #[strum(serialize = "account/rateLimits/updated")]
-    AccountRateLimitsUpdated(RateLimitSnapshot),
+    ThreadStarted => "thread/started" (v2::ThreadStartedNotification),
+    TurnStarted => "turn/started" (v2::TurnStartedNotification),
+    TurnCompleted => "turn/completed" (v2::TurnCompletedNotification),
+    ItemStarted => "item/started" (v2::ItemStartedNotification),
+    ItemCompleted => "item/completed" (v2::ItemCompletedNotification),
+    AgentMessageDelta => "item/agentMessage/delta" (v2::AgentMessageDeltaNotification),
+    CommandExecutionOutputDelta => "item/commandExecution/outputDelta" (v2::CommandExecutionOutputDeltaNotification),
+    McpToolCallProgress => "item/mcpToolCall/progress" (v2::McpToolCallProgressNotification),
+    AccountUpdated => "account/updated" (v2::AccountUpdatedNotification),
+    AccountRateLimitsUpdated => "account/rateLimits/updated" (v2::AccountRateLimitsUpdatedNotification),
+    ReasoningSummaryTextDelta => "item/reasoning/summaryTextDelta" (v2::ReasoningSummaryTextDeltaNotification),
+    ReasoningSummaryPartAdded => "item/reasoning/summaryPartAdded" (v2::ReasoningSummaryPartAddedNotification),
+    ReasoningTextDelta => "item/reasoning/textDelta" (v2::ReasoningTextDeltaNotification),
+
+    #[serde(rename = "account/login/completed")]
+    #[ts(rename = "account/login/completed")]
+    #[strum(serialize = "account/login/completed")]
+    AccountLoginCompleted(v2::AccountLoginCompletedNotification),
 
     /// DEPRECATED NOTIFICATIONS below
-    /// Authentication status changed
     AuthStatusChange(v1::AuthStatusChangeNotification),
 
-    /// ChatGPT login flow completed
+    /// Deprecated: use `account/login/completed` instead.
     LoginChatGptComplete(v1::LoginChatGptCompleteNotification),
-
-    /// The special session configured event for a new or resumed conversation.
     SessionConfigured(v1::SessionConfiguredNotification),
 }
 
-impl ServerNotification {
-    pub fn to_params(self) -> Result<serde_json::Value, serde_json::Error> {
-        match self {
-            ServerNotification::AccountRateLimitsUpdated(params) => serde_json::to_value(params),
-            ServerNotification::AuthStatusChange(params) => serde_json::to_value(params),
-            ServerNotification::LoginChatGptComplete(params) => serde_json::to_value(params),
-            ServerNotification::SessionConfigured(params) => serde_json::to_value(params),
-        }
-    }
-}
-
-impl TryFrom<JSONRPCNotification> for ServerNotification {
-    type Error = serde_json::Error;
-
-    fn try_from(value: JSONRPCNotification) -> Result<Self, Self::Error> {
-        serde_json::from_value(serde_json::to_value(value)?)
-    }
-}
-
-/// Notification sent from the client to the server.
-#[derive(Serialize, Deserialize, Debug, Clone, JsonSchema, TS, Display)]
-#[serde(tag = "method", content = "params", rename_all = "camelCase")]
-#[strum(serialize_all = "camelCase")]
-pub enum ClientNotification {
+client_notification_definitions! {
     Initialized,
 }
 
@@ -571,7 +685,7 @@ mod tests {
         };
         assert_eq!(
             json!({
-                "method": "account/login",
+                "method": "account/login/start",
                 "id": 2,
                 "params": {
                     "type": "apiKey",
@@ -587,11 +701,11 @@ mod tests {
     fn serialize_account_login_chatgpt() -> Result<()> {
         let request = ClientRequest::LoginAccount {
             request_id: RequestId::Integer(3),
-            params: v2::LoginAccountParams::ChatGpt,
+            params: v2::LoginAccountParams::Chatgpt,
         };
         assert_eq!(
             json!({
-                "method": "account/login",
+                "method": "account/login/start",
                 "id": 3,
                 "params": {
                     "type": "chatgpt"
@@ -622,12 +736,17 @@ mod tests {
     fn serialize_get_account() -> Result<()> {
         let request = ClientRequest::GetAccount {
             request_id: RequestId::Integer(5),
-            params: None,
+            params: v2::GetAccountParams {
+                refresh_token: false,
+            },
         };
         assert_eq!(
             json!({
                 "method": "account/read",
                 "id": 5,
+                "params": {
+                    "refreshToken": false
+                }
             }),
             serde_json::to_value(&request)?,
         );
@@ -636,19 +755,16 @@ mod tests {
 
     #[test]
     fn account_serializes_fields_in_camel_case() -> Result<()> {
-        let api_key = v2::Account::ApiKey {
-            api_key: "secret".to_string(),
-        };
+        let api_key = v2::Account::ApiKey {};
         assert_eq!(
             json!({
                 "type": "apiKey",
-                "apiKey": "secret",
             }),
             serde_json::to_value(&api_key)?,
         );
 
-        let chatgpt = v2::Account::ChatGpt {
-            email: Some("user@example.com".to_string()),
+        let chatgpt = v2::Account::Chatgpt {
+            email: "user@example.com".to_string(),
             plan_type: PlanType::Plus,
         };
         assert_eq!(
@@ -665,16 +781,16 @@ mod tests {
 
     #[test]
     fn serialize_list_models() -> Result<()> {
-        let request = ClientRequest::ListModels {
+        let request = ClientRequest::ModelList {
             request_id: RequestId::Integer(6),
-            params: v2::ListModelsParams::default(),
+            params: v2::ModelListParams::default(),
         };
         assert_eq!(
             json!({
                 "method": "model/list",
                 "id": 6,
                 "params": {
-                    "pageSize": null,
+                    "limit": null,
                     "cursor": null
                 }
             }),

codex-rs/app-server-protocol/src/protocol/v1.rs

@@ -11,6 +11,7 @@ use codex_protocol::models::ResponseItem;
 use codex_protocol::protocol::AskForApproval;
 use codex_protocol::protocol::EventMsg;
 use codex_protocol::protocol::SandboxPolicy;
+use codex_protocol::protocol::SessionSource;
 use codex_protocol::protocol::TurnAbortReason;
 use schemars::JsonSchema;
 use serde::Deserialize;
@@ -113,6 +114,18 @@ pub struct ConversationSummary {
     pub preview: String,
     pub timestamp: Option<String>,
     pub model_provider: String,
+    pub cwd: PathBuf,
+    pub cli_version: String,
+    pub source: SessionSource,
+    pub git_info: Option<ConversationGitInfo>,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "snake_case")]
+pub struct ConversationGitInfo {
+    pub sha: Option<String>,
+    pub branch: Option<String>,
+    pub origin_url: Option<String>,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
@@ -374,10 +387,9 @@ pub enum InputItem {
     LocalImage { path: PathBuf },
 }
 
-// Deprecated notifications (v1)
-
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
+/// Deprecated in favor of AccountLoginCompletedNotification.
 pub struct LoginChatGptCompleteNotification {
     #[schemars(with = "String")]
     pub login_id: Uuid,
@@ -400,6 +412,7 @@ pub struct SessionConfiguredNotification {
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
+/// Deprecated notification. Use AccountUpdatedNotification instead.
 pub struct AuthStatusChangeNotification {
     pub auth_method: Option<AuthMode>,
 }

codex-rs/app-server-protocol/src/protocol/v2.rs

@@ -1,35 +1,145 @@
+use std::collections::HashMap;
+use std::path::PathBuf;
+
+use crate::protocol::common::AuthMode;
 use codex_protocol::ConversationId;
 use codex_protocol::account::PlanType;
 use codex_protocol::config_types::ReasoningEffort;
-use codex_protocol::protocol::RateLimitSnapshot;
+use codex_protocol::config_types::ReasoningSummary;
+use codex_protocol::items::AgentMessageContent as CoreAgentMessageContent;
+use codex_protocol::items::TurnItem as CoreTurnItem;
+use codex_protocol::models::ResponseItem;
+use codex_protocol::protocol::RateLimitSnapshot as CoreRateLimitSnapshot;
+use codex_protocol::protocol::RateLimitWindow as CoreRateLimitWindow;
+use codex_protocol::user_input::UserInput as CoreUserInput;
+use mcp_types::ContentBlock as McpContentBlock;
 use schemars::JsonSchema;
 use serde::Deserialize;
 use serde::Serialize;
+use serde_json::Value as JsonValue;
 use ts_rs::TS;
-use uuid::Uuid;
+
+// Macro to declare a camelCased API v2 enum mirroring a core enum which
+// tends to use kebab-case.
+macro_rules! v2_enum_from_core {
+    (
+        pub enum $Name:ident from $Src:path { $( $Variant:ident ),+ $(,)? }
+    ) => {
+        #[derive(Serialize, Deserialize, Debug, Clone, Copy, PartialEq, Eq, JsonSchema, TS)]
+        #[serde(rename_all = "camelCase")]
+        #[ts(export_to = "v2/")]
+        pub enum $Name { $( $Variant ),+ }
+
+        impl $Name {
+            pub fn to_core(self) -> $Src {
+                match self { $( $Name::$Variant => <$Src>::$Variant ),+ }
+            }
+        }
+
+        impl From<$Src> for $Name {
+            fn from(value: $Src) -> Self {
+                match value { $( <$Src>::$Variant => $Name::$Variant ),+ }
+            }
+        }
+    };
+}
+
+v2_enum_from_core!(
+    pub enum AskForApproval from codex_protocol::protocol::AskForApproval {
+        UnlessTrusted, OnFailure, OnRequest, Never
+    }
+);
+
+v2_enum_from_core!(
+    pub enum SandboxMode from codex_protocol::config_types::SandboxMode {
+        ReadOnly, WorkspaceWrite, DangerFullAccess
+    }
+);
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq, JsonSchema, TS)]
+#[serde(tag = "type", rename_all = "camelCase")]
+#[ts(tag = "type")]
+#[ts(export_to = "v2/")]
+pub enum SandboxPolicy {
+    DangerFullAccess,
+    ReadOnly,
+    WorkspaceWrite {
+        #[serde(default)]
+        writable_roots: Vec<PathBuf>,
+        #[serde(default)]
+        network_access: bool,
+        #[serde(default)]
+        exclude_tmpdir_env_var: bool,
+        #[serde(default)]
+        exclude_slash_tmp: bool,
+    },
+}
+
+impl SandboxPolicy {
+    pub fn to_core(&self) -> codex_protocol::protocol::SandboxPolicy {
+        match self {
+            SandboxPolicy::DangerFullAccess => {
+                codex_protocol::protocol::SandboxPolicy::DangerFullAccess
+            }
+            SandboxPolicy::ReadOnly => codex_protocol::protocol::SandboxPolicy::ReadOnly,
+            SandboxPolicy::WorkspaceWrite {
+                writable_roots,
+                network_access,
+                exclude_tmpdir_env_var,
+                exclude_slash_tmp,
+            } => codex_protocol::protocol::SandboxPolicy::WorkspaceWrite {
+                writable_roots: writable_roots.clone(),
+                network_access: *network_access,
+                exclude_tmpdir_env_var: *exclude_tmpdir_env_var,
+                exclude_slash_tmp: *exclude_slash_tmp,
+            },
+        }
+    }
+}
+
+impl From<codex_protocol::protocol::SandboxPolicy> for SandboxPolicy {
+    fn from(value: codex_protocol::protocol::SandboxPolicy) -> Self {
+        match value {
+            codex_protocol::protocol::SandboxPolicy::DangerFullAccess => {
+                SandboxPolicy::DangerFullAccess
+            }
+            codex_protocol::protocol::SandboxPolicy::ReadOnly => SandboxPolicy::ReadOnly,
+            codex_protocol::protocol::SandboxPolicy::WorkspaceWrite {
+                writable_roots,
+                network_access,
+                exclude_tmpdir_env_var,
+                exclude_slash_tmp,
+            } => SandboxPolicy::WorkspaceWrite {
+                writable_roots,
+                network_access,
+                exclude_tmpdir_env_var,
+                exclude_slash_tmp,
+            },
+        }
+    }
+}
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(tag = "type", rename_all = "camelCase")]
 #[ts(tag = "type")]
+#[ts(export_to = "v2/")]
 pub enum Account {
     #[serde(rename = "apiKey", rename_all = "camelCase")]
     #[ts(rename = "apiKey", rename_all = "camelCase")]
-    ApiKey { api_key: String },
+    ApiKey {},
 
     #[serde(rename = "chatgpt", rename_all = "camelCase")]
     #[ts(rename = "chatgpt", rename_all = "camelCase")]
-    ChatGpt {
-        email: Option<String>,
-        plan_type: PlanType,
-    },
+    Chatgpt { email: String, plan_type: PlanType },
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(tag = "type")]
 #[ts(tag = "type")]
+#[ts(export_to = "v2/")]
 pub enum LoginAccountParams {
-    #[serde(rename = "apiKey")]
-    #[ts(rename = "apiKey")]
+    #[serde(rename = "apiKey", rename_all = "camelCase")]
+    #[ts(rename = "apiKey", rename_all = "camelCase")]
     ApiKey {
         #[serde(rename = "apiKey")]
         #[ts(rename = "apiKey")]
@@ -37,48 +147,81 @@ pub enum LoginAccountParams {
     },
     #[serde(rename = "chatgpt")]
     #[ts(rename = "chatgpt")]
-    ChatGpt,
+    Chatgpt,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(tag = "type", rename_all = "camelCase")]
+#[ts(tag = "type")]
+#[ts(export_to = "v2/")]
+pub enum LoginAccountResponse {
+    #[serde(rename = "apiKey", rename_all = "camelCase")]
+    #[ts(rename = "apiKey", rename_all = "camelCase")]
+    ApiKey {},
+    #[serde(rename = "chatgpt", rename_all = "camelCase")]
+    #[ts(rename = "chatgpt", rename_all = "camelCase")]
+    Chatgpt {
+        // Use plain String for identifiers to avoid TS/JSON Schema quirks around uuid-specific types.
+        // Convert to/from UUIDs at the application layer as needed.
+        login_id: String,
+        /// URL the client should open in a browser to initiate the OAuth flow.
+        auth_url: String,
+    },
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
-pub struct LoginAccountResponse {
-    /// Only set if the login method is ChatGPT.
-    #[schemars(with = "String")]
-    pub login_id: Option<Uuid>,
-
-    /// URL the client should open in a browser to initiate the OAuth flow.
-    /// Only set if the login method is ChatGPT.
-    pub auth_url: Option<String>,
+#[ts(export_to = "v2/")]
+pub struct CancelLoginAccountParams {
+    pub login_id: String,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct CancelLoginAccountResponse {}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
 pub struct LogoutAccountResponse {}
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
 pub struct GetAccountRateLimitsResponse {
     pub rate_limits: RateLimitSnapshot,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
-pub struct GetAccountResponse {
-    pub account: Account,
-}
-
-#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Default, JsonSchema, TS)]
-#[serde(rename_all = "camelCase")]
-pub struct ListModelsParams {
-    /// Optional page size; defaults to a reasonable server-side value.
-    pub page_size: Option<usize>,
-    /// Opaque pagination cursor returned by a previous call.
-    pub cursor: Option<String>,
+#[ts(export_to = "v2/")]
+pub struct GetAccountParams {
+    #[serde(default)]
+    pub refresh_token: bool,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct GetAccountResponse {
+    pub account: Option<Account>,
+    pub requires_openai_auth: bool,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Default, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ModelListParams {
+    /// Opaque pagination cursor returned by a previous call.
+    pub cursor: Option<String>,
+    /// Optional page size; defaults to a reasonable server-side value.
+    pub limit: Option<u32>,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
 pub struct Model {
     pub id: String,
     pub model: String,
@@ -92,6 +235,7 @@ pub struct Model {
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
 pub struct ReasoningEffortOption {
     pub reasoning_effort: ReasoningEffort,
     pub description: String,
@@ -99,16 +243,18 @@ pub struct ReasoningEffortOption {
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
-pub struct ListModelsResponse {
-    pub items: Vec<Model>,
+#[ts(export_to = "v2/")]
+pub struct ModelListResponse {
+    pub data: Vec<Model>,
     /// Opaque cursor to pass to the next call to continue after the last item.
-    /// if None, there are no more items to return.
+    /// If None, there are no more items to return.
     pub next_cursor: Option<String>,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
-pub struct UploadFeedbackParams {
+#[ts(export_to = "v2/")]
+pub struct FeedbackUploadParams {
     pub classification: String,
     pub reason: Option<String>,
     pub conversation_id: Option<ConversationId>,
@@ -117,6 +263,649 @@ pub struct UploadFeedbackParams {
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
-pub struct UploadFeedbackResponse {
+#[ts(export_to = "v2/")]
+pub struct FeedbackUploadResponse {
     pub thread_id: String,
 }
+
+// === Threads, Turns, and Items ===
+// Thread APIs
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Default, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ThreadStartParams {
+    pub model: Option<String>,
+    pub model_provider: Option<String>,
+    pub cwd: Option<String>,
+    pub approval_policy: Option<AskForApproval>,
+    pub sandbox: Option<SandboxMode>,
+    pub config: Option<HashMap<String, serde_json::Value>>,
+    pub base_instructions: Option<String>,
+    pub developer_instructions: Option<String>,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ThreadStartResponse {
+    pub thread: Thread,
+}
+
+#[derive(Serialize, Deserialize, Debug, Default, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+/// There are three ways to resume a thread:
+/// 1. By thread_id: load the thread from disk by thread_id and resume it.
+/// 2. By history: instantiate the thread from memory and resume it.
+/// 3. By path: load the thread from disk by path and resume it.
+///
+/// The precedence is: history > path > thread_id.
+/// If using history or path, the thread_id param will be ignored.
+///
+/// Prefer using thread_id whenever possible.
+pub struct ThreadResumeParams {
+    pub thread_id: String,
+
+    /// [UNSTABLE] FOR CODEX CLOUD - DO NOT USE.
+    /// If specified, the thread will be resumed with the provided history
+    /// instead of loaded from disk.
+    pub history: Option<Vec<ResponseItem>>,
+
+    /// [UNSTABLE] Specify the rollout path to resume from.
+    /// If specified, the thread_id param will be ignored.
+    pub path: Option<PathBuf>,
+
+    /// Configuration overrides for the resumed thread, if any.
+    pub model: Option<String>,
+    pub model_provider: Option<String>,
+    pub cwd: Option<String>,
+    pub approval_policy: Option<AskForApproval>,
+    pub sandbox: Option<SandboxMode>,
+    pub config: Option<HashMap<String, serde_json::Value>>,
+    pub base_instructions: Option<String>,
+    pub developer_instructions: Option<String>,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ThreadResumeResponse {
+    pub thread: Thread,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ThreadArchiveParams {
+    pub thread_id: String,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ThreadArchiveResponse {}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ThreadListParams {
+    /// Opaque pagination cursor returned by a previous call.
+    pub cursor: Option<String>,
+    /// Optional page size; defaults to a reasonable server-side value.
+    pub limit: Option<u32>,
+    /// Optional provider filter; when set, only sessions recorded under these
+    /// providers are returned. When present but empty, includes all providers.
+    pub model_providers: Option<Vec<String>>,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ThreadListResponse {
+    pub data: Vec<Thread>,
+    /// Opaque cursor to pass to the next call to continue after the last item.
+    /// if None, there are no more items to return.
+    pub next_cursor: Option<String>,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ThreadCompactParams {
+    pub thread_id: String,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ThreadCompactResponse {}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct Thread {
+    pub id: String,
+    /// Usually the first user message in the thread, if available.
+    pub preview: String,
+    pub model_provider: String,
+    /// Unix timestamp (in seconds) when the thread was created.
+    pub created_at: i64,
+    /// [UNSTABLE] Path to the thread on disk.
+    pub path: PathBuf,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct AccountUpdatedNotification {
+    pub auth_mode: Option<AuthMode>,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct Turn {
+    pub id: String,
+    pub items: Vec<ThreadItem>,
+    pub status: TurnStatus,
+    pub error: Option<TurnError>,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct TurnError {
+    pub message: String,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub enum TurnStatus {
+    Completed,
+    Interrupted,
+    Failed,
+    InProgress,
+}
+
+// Turn APIs
+#[derive(Serialize, Deserialize, Debug, Default, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct TurnStartParams {
+    pub thread_id: String,
+    pub input: Vec<UserInput>,
+    /// Override the working directory for this turn and subsequent turns.
+    pub cwd: Option<PathBuf>,
+    /// Override the approval policy for this turn and subsequent turns.
+    pub approval_policy: Option<AskForApproval>,
+    /// Override the sandbox policy for this turn and subsequent turns.
+    pub sandbox_policy: Option<SandboxPolicy>,
+    /// Override the model for this turn and subsequent turns.
+    pub model: Option<String>,
+    /// Override the reasoning effort for this turn and subsequent turns.
+    pub effort: Option<ReasoningEffort>,
+    /// Override the reasoning summary for this turn and subsequent turns.
+    pub summary: Option<ReasoningSummary>,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct TurnStartResponse {
+    pub turn: Turn,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct TurnInterruptParams {
+    pub thread_id: String,
+    pub turn_id: String,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct TurnInterruptResponse {}
+
+// User input types
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(tag = "type", rename_all = "camelCase")]
+#[ts(tag = "type")]
+#[ts(export_to = "v2/")]
+pub enum UserInput {
+    Text { text: String },
+    Image { url: String },
+    LocalImage { path: PathBuf },
+}
+
+impl UserInput {
+    pub fn into_core(self) -> CoreUserInput {
+        match self {
+            UserInput::Text { text } => CoreUserInput::Text { text },
+            UserInput::Image { url } => CoreUserInput::Image { image_url: url },
+            UserInput::LocalImage { path } => CoreUserInput::LocalImage { path },
+        }
+    }
+}
+
+impl From<CoreUserInput> for UserInput {
+    fn from(value: CoreUserInput) -> Self {
+        match value {
+            CoreUserInput::Text { text } => UserInput::Text { text },
+            CoreUserInput::Image { image_url } => UserInput::Image { url: image_url },
+            CoreUserInput::LocalImage { path } => UserInput::LocalImage { path },
+            _ => unreachable!("unsupported user input variant"),
+        }
+    }
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(tag = "type", rename_all = "camelCase")]
+#[ts(tag = "type")]
+#[ts(export_to = "v2/")]
+pub enum ThreadItem {
+    UserMessage {
+        id: String,
+        content: Vec<UserInput>,
+    },
+    AgentMessage {
+        id: String,
+        text: String,
+    },
+    Reasoning {
+        id: String,
+        #[serde(default)]
+        summary: Vec<String>,
+        #[serde(default)]
+        content: Vec<String>,
+    },
+    CommandExecution {
+        id: String,
+        command: String,
+        aggregated_output: String,
+        exit_code: Option<i32>,
+        status: CommandExecutionStatus,
+        duration_ms: Option<i64>,
+    },
+    FileChange {
+        id: String,
+        changes: Vec<FileUpdateChange>,
+        status: PatchApplyStatus,
+    },
+    McpToolCall {
+        id: String,
+        server: String,
+        tool: String,
+        status: McpToolCallStatus,
+        arguments: JsonValue,
+        result: Option<McpToolCallResult>,
+        error: Option<McpToolCallError>,
+    },
+    WebSearch {
+        id: String,
+        query: String,
+    },
+    TodoList {
+        id: String,
+        items: Vec<TodoItem>,
+    },
+    ImageView {
+        id: String,
+        path: String,
+    },
+    CodeReview {
+        id: String,
+        review: String,
+    },
+}
+
+impl From<CoreTurnItem> for ThreadItem {
+    fn from(value: CoreTurnItem) -> Self {
+        match value {
+            CoreTurnItem::UserMessage(user) => ThreadItem::UserMessage {
+                id: user.id,
+                content: user.content.into_iter().map(UserInput::from).collect(),
+            },
+            CoreTurnItem::AgentMessage(agent) => {
+                let text = agent
+                    .content
+                    .into_iter()
+                    .map(|entry| match entry {
+                        CoreAgentMessageContent::Text { text } => text,
+                    })
+                    .collect::<String>();
+                ThreadItem::AgentMessage { id: agent.id, text }
+            }
+            CoreTurnItem::Reasoning(reasoning) => ThreadItem::Reasoning {
+                id: reasoning.id,
+                summary: reasoning.summary_text,
+                content: reasoning.raw_content,
+            },
+            CoreTurnItem::WebSearch(search) => ThreadItem::WebSearch {
+                id: search.id,
+                query: search.query,
+            },
+        }
+    }
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub enum CommandExecutionStatus {
+    InProgress,
+    Completed,
+    Failed,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct FileUpdateChange {
+    pub path: String,
+    pub kind: PatchChangeKind,
+    pub diff: String,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub enum PatchChangeKind {
+    Add,
+    Delete,
+    Update,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub enum PatchApplyStatus {
+    Completed,
+    Failed,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub enum McpToolCallStatus {
+    InProgress,
+    Completed,
+    Failed,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct McpToolCallResult {
+    pub content: Vec<McpContentBlock>,
+    pub structured_content: Option<JsonValue>,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct McpToolCallError {
+    pub message: String,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct TodoItem {
+    pub id: String,
+    pub text: String,
+    pub completed: bool,
+}
+
+// === Server Notifications ===
+// Thread/Turn lifecycle notifications and item progress events
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ThreadStartedNotification {
+    pub thread: Thread,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct TurnStartedNotification {
+    pub turn: Turn,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct Usage {
+    pub input_tokens: i32,
+    pub cached_input_tokens: i32,
+    pub output_tokens: i32,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct TurnCompletedNotification {
+    pub turn: Turn,
+    // TODO: should usage be stored on the Turn object, and we return that instead?
+    pub usage: Usage,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ItemStartedNotification {
+    pub item: ThreadItem,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ItemCompletedNotification {
+    pub item: ThreadItem,
+}
+
+// Item-specific progress notifications
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct AgentMessageDeltaNotification {
+    pub item_id: String,
+    pub delta: String,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ReasoningSummaryTextDeltaNotification {
+    pub item_id: String,
+    pub delta: String,
+    pub summary_index: i64,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ReasoningSummaryPartAddedNotification {
+    pub item_id: String,
+    pub summary_index: i64,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ReasoningTextDeltaNotification {
+    pub item_id: String,
+    pub delta: String,
+    pub content_index: i64,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct CommandExecutionOutputDeltaNotification {
+    pub item_id: String,
+    pub delta: String,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct McpToolCallProgressNotification {
+    pub item_id: String,
+    pub message: String,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct AccountRateLimitsUpdatedNotification {
+    pub rate_limits: RateLimitSnapshot,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct RateLimitSnapshot {
+    pub primary: Option<RateLimitWindow>,
+    pub secondary: Option<RateLimitWindow>,
+}
+
+impl From<CoreRateLimitSnapshot> for RateLimitSnapshot {
+    fn from(value: CoreRateLimitSnapshot) -> Self {
+        Self {
+            primary: value.primary.map(RateLimitWindow::from),
+            secondary: value.secondary.map(RateLimitWindow::from),
+        }
+    }
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct RateLimitWindow {
+    pub used_percent: i32,
+    pub window_duration_mins: Option<i64>,
+    pub resets_at: Option<i64>,
+}
+
+impl From<CoreRateLimitWindow> for RateLimitWindow {
+    fn from(value: CoreRateLimitWindow) -> Self {
+        Self {
+            used_percent: value.used_percent.round() as i32,
+            window_duration_mins: value.window_minutes,
+            resets_at: value.resets_at,
+        }
+    }
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct AccountLoginCompletedNotification {
+    // Use plain String for identifiers to avoid TS/JSON Schema quirks around uuid-specific types.
+    // Convert to/from UUIDs at the application layer as needed.
+    pub login_id: Option<String>,
+    pub success: bool,
+    pub error: Option<String>,
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use codex_protocol::items::AgentMessageContent;
+    use codex_protocol::items::AgentMessageItem;
+    use codex_protocol::items::ReasoningItem;
+    use codex_protocol::items::TurnItem;
+    use codex_protocol::items::UserMessageItem;
+    use codex_protocol::items::WebSearchItem;
+    use codex_protocol::user_input::UserInput as CoreUserInput;
+    use pretty_assertions::assert_eq;
+    use std::path::PathBuf;
+
+    #[test]
+    fn core_turn_item_into_thread_item_converts_supported_variants() {
+        let user_item = TurnItem::UserMessage(UserMessageItem {
+            id: "user-1".to_string(),
+            content: vec![
+                CoreUserInput::Text {
+                    text: "hello".to_string(),
+                },
+                CoreUserInput::Image {
+                    image_url: "https://example.com/image.png".to_string(),
+                },
+                CoreUserInput::LocalImage {
+                    path: PathBuf::from("local/image.png"),
+                },
+            ],
+        });
+
+        assert_eq!(
+            ThreadItem::from(user_item),
+            ThreadItem::UserMessage {
+                id: "user-1".to_string(),
+                content: vec![
+                    UserInput::Text {
+                        text: "hello".to_string(),
+                    },
+                    UserInput::Image {
+                        url: "https://example.com/image.png".to_string(),
+                    },
+                    UserInput::LocalImage {
+                        path: PathBuf::from("local/image.png"),
+                    },
+                ],
+            }
+        );
+
+        let agent_item = TurnItem::AgentMessage(AgentMessageItem {
+            id: "agent-1".to_string(),
+            content: vec![
+                AgentMessageContent::Text {
+                    text: "Hello ".to_string(),
+                },
+                AgentMessageContent::Text {
+                    text: "world".to_string(),
+                },
+            ],
+        });
+
+        assert_eq!(
+            ThreadItem::from(agent_item),
+            ThreadItem::AgentMessage {
+                id: "agent-1".to_string(),
+                text: "Hello world".to_string(),
+            }
+        );
+
+        let reasoning_item = TurnItem::Reasoning(ReasoningItem {
+            id: "reasoning-1".to_string(),
+            summary_text: vec!["line one".to_string(), "line two".to_string()],
+            raw_content: vec![],
+        });
+
+        assert_eq!(
+            ThreadItem::from(reasoning_item),
+            ThreadItem::Reasoning {
+                id: "reasoning-1".to_string(),
+                summary: vec!["line one".to_string(), "line two".to_string()],
+                content: vec![],
+            }
+        );
+
+        let search_item = TurnItem::WebSearch(WebSearchItem {
+            id: "search-1".to_string(),
+            query: "docs".to_string(),
+        });
+
+        assert_eq!(
+            ThreadItem::from(search_item),
+            ThreadItem::WebSearch {
+                id: "search-1".to_string(),
+                query: "docs".to_string(),
+            }
+        );
+    }
+}

codex-rs/app-server-test-client/Cargo.toml

@@ -0,0 +1,16 @@
+[package]
+name = "codex-app-server-test-client"
+version = { workspace = true }
+edition = "2024"
+
+[lints]
+workspace = true
+
+[dependencies]
+anyhow = { workspace = true }
+clap = { workspace = true, features = ["derive", "env"] }
+codex-app-server-protocol = { workspace = true }
+codex-protocol = { workspace = true }
+serde = { workspace = true, features = ["derive"] }
+serde_json = { workspace = true }
+uuid = { workspace = true, features = ["v4"] }

codex-rs/app-server-test-client/README.md

@@ -0,0 +1,2 @@
+# App Server Test Client
+Exercises simple `codex app-server` flows end-to-end, logging JSON-RPC messages sent between client and server to stdout.

codex-rs/app-server-test-client/src/main.rs

@@ -0,0 +1,691 @@
+use std::collections::VecDeque;
+use std::io::BufRead;
+use std::io::BufReader;
+use std::io::Write;
+use std::process::Child;
+use std::process::ChildStdin;
+use std::process::ChildStdout;
+use std::process::Command;
+use std::process::Stdio;
+use std::thread;
+use std::time::Duration;
+
+use anyhow::Context;
+use anyhow::Result;
+use anyhow::bail;
+use clap::Parser;
+use clap::Subcommand;
+use codex_app_server_protocol::AddConversationListenerParams;
+use codex_app_server_protocol::AddConversationSubscriptionResponse;
+use codex_app_server_protocol::AskForApproval;
+use codex_app_server_protocol::ClientInfo;
+use codex_app_server_protocol::ClientRequest;
+use codex_app_server_protocol::GetAccountRateLimitsResponse;
+use codex_app_server_protocol::InitializeParams;
+use codex_app_server_protocol::InitializeResponse;
+use codex_app_server_protocol::InputItem;
+use codex_app_server_protocol::JSONRPCMessage;
+use codex_app_server_protocol::JSONRPCNotification;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::LoginChatGptCompleteNotification;
+use codex_app_server_protocol::LoginChatGptResponse;
+use codex_app_server_protocol::NewConversationParams;
+use codex_app_server_protocol::NewConversationResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::SandboxPolicy;
+use codex_app_server_protocol::SendUserMessageParams;
+use codex_app_server_protocol::SendUserMessageResponse;
+use codex_app_server_protocol::ServerNotification;
+use codex_app_server_protocol::ThreadStartParams;
+use codex_app_server_protocol::ThreadStartResponse;
+use codex_app_server_protocol::TurnStartParams;
+use codex_app_server_protocol::TurnStartResponse;
+use codex_app_server_protocol::UserInput as V2UserInput;
+use codex_protocol::ConversationId;
+use codex_protocol::protocol::Event;
+use codex_protocol::protocol::EventMsg;
+use serde::de::DeserializeOwned;
+use serde_json::Value;
+use uuid::Uuid;
+
+/// Minimal launcher that initializes the Codex app-server and logs the handshake.
+#[derive(Parser)]
+#[command(author = "Codex", version, about = "Bootstrap Codex app-server", long_about = None)]
+struct Cli {
+    /// Path to the `codex` CLI binary.
+    #[arg(long, env = "CODEX_BIN", default_value = "codex")]
+    codex_bin: String,
+
+    #[command(subcommand)]
+    command: CliCommand,
+}
+
+#[derive(Subcommand)]
+enum CliCommand {
+    /// Send a user message through the Codex app-server.
+    SendMessage {
+        /// User message to send to Codex.
+        #[arg()]
+        user_message: String,
+    },
+    /// Send a user message through the app-server V2 thread/turn APIs.
+    SendMessageV2 {
+        /// User message to send to Codex.
+        #[arg()]
+        user_message: String,
+    },
+    /// Start a V2 turn that elicits an ExecCommand approval.
+    #[command(name = "trigger-cmd-approval")]
+    TriggerCmdApproval {
+        /// Optional prompt; defaults to a simple python command.
+        #[arg()]
+        user_message: Option<String>,
+    },
+    /// Start a V2 turn that elicits an ApplyPatch approval.
+    #[command(name = "trigger-patch-approval")]
+    TriggerPatchApproval {
+        /// Optional prompt; defaults to creating a file via apply_patch.
+        #[arg()]
+        user_message: Option<String>,
+    },
+    /// Start a V2 turn that should not elicit an ExecCommand approval.
+    #[command(name = "no-trigger-cmd-approval")]
+    NoTriggerCmdApproval,
+    /// Trigger the ChatGPT login flow and wait for completion.
+    TestLogin,
+    /// Fetch the current account rate limits from the Codex app-server.
+    GetAccountRateLimits,
+}
+
+fn main() -> Result<()> {
+    let Cli { codex_bin, command } = Cli::parse();
+
+    match command {
+        CliCommand::SendMessage { user_message } => send_message(codex_bin, user_message),
+        CliCommand::SendMessageV2 { user_message } => send_message_v2(codex_bin, user_message),
+        CliCommand::TriggerCmdApproval { user_message } => {
+            trigger_cmd_approval(codex_bin, user_message)
+        }
+        CliCommand::TriggerPatchApproval { user_message } => {
+            trigger_patch_approval(codex_bin, user_message)
+        }
+        CliCommand::NoTriggerCmdApproval => no_trigger_cmd_approval(codex_bin),
+        CliCommand::TestLogin => test_login(codex_bin),
+        CliCommand::GetAccountRateLimits => get_account_rate_limits(codex_bin),
+    }
+}
+
+fn send_message(codex_bin: String, user_message: String) -> Result<()> {
+    let mut client = CodexClient::spawn(codex_bin)?;
+
+    let initialize = client.initialize()?;
+    println!("< initialize response: {initialize:?}");
+
+    let conversation = client.new_conversation()?;
+    println!("< newConversation response: {conversation:?}");
+
+    let subscription = client.add_conversation_listener(&conversation.conversation_id)?;
+    println!("< addConversationListener response: {subscription:?}");
+
+    let send_response = client.send_user_message(&conversation.conversation_id, &user_message)?;
+    println!("< sendUserMessage response: {send_response:?}");
+
+    client.stream_conversation(&conversation.conversation_id)?;
+
+    client.remove_conversation_listener(subscription.subscription_id)?;
+
+    Ok(())
+}
+
+fn send_message_v2(codex_bin: String, user_message: String) -> Result<()> {
+    send_message_v2_with_policies(codex_bin, user_message, None, None)
+}
+
+fn trigger_cmd_approval(codex_bin: String, user_message: Option<String>) -> Result<()> {
+    let default_prompt =
+        "Run `touch /tmp/should-trigger-approval` so I can confirm the file exists.";
+    let message = user_message.unwrap_or_else(|| default_prompt.to_string());
+    send_message_v2_with_policies(
+        codex_bin,
+        message,
+        Some(AskForApproval::OnRequest),
+        Some(SandboxPolicy::ReadOnly),
+    )
+}
+
+fn trigger_patch_approval(codex_bin: String, user_message: Option<String>) -> Result<()> {
+    let default_prompt =
+        "Create a file named APPROVAL_DEMO.txt containing a short hello message using apply_patch.";
+    let message = user_message.unwrap_or_else(|| default_prompt.to_string());
+    send_message_v2_with_policies(
+        codex_bin,
+        message,
+        Some(AskForApproval::OnRequest),
+        Some(SandboxPolicy::ReadOnly),
+    )
+}
+
+fn no_trigger_cmd_approval(codex_bin: String) -> Result<()> {
+    let prompt = "Run `touch should_not_trigger_approval.txt`";
+    send_message_v2_with_policies(codex_bin, prompt.to_string(), None, None)
+}
+
+fn send_message_v2_with_policies(
+    codex_bin: String,
+    user_message: String,
+    approval_policy: Option<AskForApproval>,
+    sandbox_policy: Option<SandboxPolicy>,
+) -> Result<()> {
+    let mut client = CodexClient::spawn(codex_bin)?;
+
+    let initialize = client.initialize()?;
+    println!("< initialize response: {initialize:?}");
+
+    let thread_response = client.thread_start(ThreadStartParams::default())?;
+    println!("< thread/start response: {thread_response:?}");
+    let mut turn_params = TurnStartParams {
+        thread_id: thread_response.thread.id.clone(),
+        input: vec![V2UserInput::Text { text: user_message }],
+        ..Default::default()
+    };
+    turn_params.approval_policy = approval_policy;
+    turn_params.sandbox_policy = sandbox_policy;
+
+    let turn_response = client.turn_start(turn_params)?;
+    println!("< turn/start response: {turn_response:?}");
+
+    client.stream_turn(&thread_response.thread.id, &turn_response.turn.id)?;
+
+    Ok(())
+}
+
+fn test_login(codex_bin: String) -> Result<()> {
+    let mut client = CodexClient::spawn(codex_bin)?;
+
+    let initialize = client.initialize()?;
+    println!("< initialize response: {initialize:?}");
+
+    let login_response = client.login_chat_gpt()?;
+    println!("< loginChatGpt response: {login_response:?}");
+    println!(
+        "Open the following URL in your browser to continue:\n{}",
+        login_response.auth_url
+    );
+
+    let completion = client.wait_for_login_completion(&login_response.login_id)?;
+    println!("< loginChatGptComplete notification: {completion:?}");
+
+    if completion.success {
+        println!("Login succeeded.");
+        Ok(())
+    } else {
+        bail!(
+            "login failed: {}",
+            completion
+                .error
+                .as_deref()
+                .unwrap_or("unknown error from loginChatGptComplete")
+        );
+    }
+}
+
+fn get_account_rate_limits(codex_bin: String) -> Result<()> {
+    let mut client = CodexClient::spawn(codex_bin)?;
+
+    let initialize = client.initialize()?;
+    println!("< initialize response: {initialize:?}");
+
+    let response = client.get_account_rate_limits()?;
+    println!("< account/rateLimits/read response: {response:?}");
+
+    Ok(())
+}
+
+struct CodexClient {
+    child: Child,
+    stdin: Option<ChildStdin>,
+    stdout: BufReader<ChildStdout>,
+    pending_notifications: VecDeque<JSONRPCNotification>,
+}
+
+impl CodexClient {
+    fn spawn(codex_bin: String) -> Result<Self> {
+        let mut codex_app_server = Command::new(&codex_bin)
+            .arg("app-server")
+            .stdin(Stdio::piped())
+            .stdout(Stdio::piped())
+            .stderr(Stdio::inherit())
+            .spawn()
+            .with_context(|| format!("failed to start `{codex_bin}` app-server"))?;
+
+        let stdin = codex_app_server
+            .stdin
+            .take()
+            .context("codex app-server stdin unavailable")?;
+        let stdout = codex_app_server
+            .stdout
+            .take()
+            .context("codex app-server stdout unavailable")?;
+
+        Ok(Self {
+            child: codex_app_server,
+            stdin: Some(stdin),
+            stdout: BufReader::new(stdout),
+            pending_notifications: VecDeque::new(),
+        })
+    }
+
+    fn initialize(&mut self) -> Result<InitializeResponse> {
+        let request_id = self.request_id();
+        let request = ClientRequest::Initialize {
+            request_id: request_id.clone(),
+            params: InitializeParams {
+                client_info: ClientInfo {
+                    name: "codex-toy-app-server".to_string(),
+                    title: Some("Codex Toy App Server".to_string()),
+                    version: env!("CARGO_PKG_VERSION").to_string(),
+                },
+            },
+        };
+
+        self.send_request(request, request_id, "initialize")
+    }
+
+    fn new_conversation(&mut self) -> Result<NewConversationResponse> {
+        let request_id = self.request_id();
+        let request = ClientRequest::NewConversation {
+            request_id: request_id.clone(),
+            params: NewConversationParams::default(),
+        };
+
+        self.send_request(request, request_id, "newConversation")
+    }
+
+    fn add_conversation_listener(
+        &mut self,
+        conversation_id: &ConversationId,
+    ) -> Result<AddConversationSubscriptionResponse> {
+        let request_id = self.request_id();
+        let request = ClientRequest::AddConversationListener {
+            request_id: request_id.clone(),
+            params: AddConversationListenerParams {
+                conversation_id: *conversation_id,
+                experimental_raw_events: false,
+            },
+        };
+
+        self.send_request(request, request_id, "addConversationListener")
+    }
+
+    fn remove_conversation_listener(&mut self, subscription_id: Uuid) -> Result<()> {
+        let request_id = self.request_id();
+        let request = ClientRequest::RemoveConversationListener {
+            request_id: request_id.clone(),
+            params: codex_app_server_protocol::RemoveConversationListenerParams { subscription_id },
+        };
+
+        self.send_request::<codex_app_server_protocol::RemoveConversationSubscriptionResponse>(
+            request,
+            request_id,
+            "removeConversationListener",
+        )?;
+
+        Ok(())
+    }
+
+    fn send_user_message(
+        &mut self,
+        conversation_id: &ConversationId,
+        message: &str,
+    ) -> Result<SendUserMessageResponse> {
+        let request_id = self.request_id();
+        let request = ClientRequest::SendUserMessage {
+            request_id: request_id.clone(),
+            params: SendUserMessageParams {
+                conversation_id: *conversation_id,
+                items: vec![InputItem::Text {
+                    text: message.to_string(),
+                }],
+            },
+        };
+
+        self.send_request(request, request_id, "sendUserMessage")
+    }
+
+    fn thread_start(&mut self, params: ThreadStartParams) -> Result<ThreadStartResponse> {
+        let request_id = self.request_id();
+        let request = ClientRequest::ThreadStart {
+            request_id: request_id.clone(),
+            params,
+        };
+
+        self.send_request(request, request_id, "thread/start")
+    }
+
+    fn turn_start(&mut self, params: TurnStartParams) -> Result<TurnStartResponse> {
+        let request_id = self.request_id();
+        let request = ClientRequest::TurnStart {
+            request_id: request_id.clone(),
+            params,
+        };
+
+        self.send_request(request, request_id, "turn/start")
+    }
+
+    fn login_chat_gpt(&mut self) -> Result<LoginChatGptResponse> {
+        let request_id = self.request_id();
+        let request = ClientRequest::LoginChatGpt {
+            request_id: request_id.clone(),
+            params: None,
+        };
+
+        self.send_request(request, request_id, "loginChatGpt")
+    }
+
+    fn get_account_rate_limits(&mut self) -> Result<GetAccountRateLimitsResponse> {
+        let request_id = self.request_id();
+        let request = ClientRequest::GetAccountRateLimits {
+            request_id: request_id.clone(),
+            params: None,
+        };
+
+        self.send_request(request, request_id, "account/rateLimits/read")
+    }
+
+    fn stream_conversation(&mut self, conversation_id: &ConversationId) -> Result<()> {
+        loop {
+            let notification = self.next_notification()?;
+
+            if !notification.method.starts_with("codex/event/") {
+                continue;
+            }
+
+            if let Some(event) = self.extract_event(notification, conversation_id)? {
+                match &event.msg {
+                    EventMsg::AgentMessage(event) => {
+                        println!("{}", event.message);
+                    }
+                    EventMsg::AgentMessageDelta(event) => {
+                        print!("{}", event.delta);
+                        std::io::stdout().flush().ok();
+                    }
+                    EventMsg::TaskComplete(event) => {
+                        println!("\n[task complete: {event:?}]");
+                        break;
+                    }
+                    EventMsg::TurnAborted(event) => {
+                        println!("\n[turn aborted: {:?}]", event.reason);
+                        break;
+                    }
+                    EventMsg::Error(event) => {
+                        println!("[error] {event:?}");
+                    }
+                    _ => {
+                        println!("[UNKNOWN EVENT] {:?}", event.msg);
+                    }
+                }
+            }
+        }
+
+        Ok(())
+    }
+
+    fn wait_for_login_completion(
+        &mut self,
+        expected_login_id: &Uuid,
+    ) -> Result<LoginChatGptCompleteNotification> {
+        loop {
+            let notification = self.next_notification()?;
+
+            if let Ok(server_notification) = ServerNotification::try_from(notification) {
+                match server_notification {
+                    ServerNotification::LoginChatGptComplete(completion) => {
+                        if &completion.login_id == expected_login_id {
+                            return Ok(completion);
+                        }
+
+                        println!(
+                            "[ignoring loginChatGptComplete for unexpected login_id: {}]",
+                            completion.login_id
+                        );
+                    }
+                    ServerNotification::AuthStatusChange(status) => {
+                        println!("< authStatusChange notification: {status:?}");
+                    }
+                    ServerNotification::AccountRateLimitsUpdated(snapshot) => {
+                        println!("< accountRateLimitsUpdated notification: {snapshot:?}");
+                    }
+                    ServerNotification::SessionConfigured(_) => {
+                        // SessionConfigured notifications are unrelated to login; skip.
+                    }
+                    _ => {}
+                }
+            }
+
+            // Not a server notification (likely a conversation event); keep waiting.
+        }
+    }
+
+    fn stream_turn(&mut self, thread_id: &str, turn_id: &str) -> Result<()> {
+        loop {
+            let notification = self.next_notification()?;
+
+            let Ok(server_notification) = ServerNotification::try_from(notification) else {
+                continue;
+            };
+
+            match server_notification {
+                ServerNotification::ThreadStarted(payload) => {
+                    if payload.thread.id == thread_id {
+                        println!("< thread/started notification: {:?}", payload.thread);
+                    }
+                }
+                ServerNotification::TurnStarted(payload) => {
+                    if payload.turn.id == turn_id {
+                        println!("< turn/started notification: {:?}", payload.turn.status);
+                    }
+                }
+                ServerNotification::AgentMessageDelta(delta) => {
+                    print!("{}", delta.delta);
+                    std::io::stdout().flush().ok();
+                }
+                ServerNotification::CommandExecutionOutputDelta(delta) => {
+                    print!("{}", delta.delta);
+                    std::io::stdout().flush().ok();
+                }
+                ServerNotification::ItemStarted(payload) => {
+                    println!("\n< item started: {:?}", payload.item);
+                }
+                ServerNotification::ItemCompleted(payload) => {
+                    println!("< item completed: {:?}", payload.item);
+                }
+                ServerNotification::TurnCompleted(payload) => {
+                    if payload.turn.id == turn_id {
+                        println!("\n< turn/completed notification: {:?}", payload.turn.status);
+                        if let Some(error) = payload.turn.error {
+                            println!("[turn error] {}", error.message);
+                        }
+                        println!("< usage: {:?}", payload.usage);
+                        break;
+                    }
+                }
+                ServerNotification::McpToolCallProgress(payload) => {
+                    println!("< MCP tool progress: {}", payload.message);
+                }
+                _ => {
+                    println!("[UNKNOWN SERVER NOTIFICATION] {server_notification:?}");
+                }
+            }
+        }
+
+        Ok(())
+    }
+
+    fn extract_event(
+        &self,
+        notification: JSONRPCNotification,
+        conversation_id: &ConversationId,
+    ) -> Result<Option<Event>> {
+        let params = notification
+            .params
+            .context("event notification missing params")?;
+
+        let mut map = match params {
+            Value::Object(map) => map,
+            other => bail!("unexpected params shape: {other:?}"),
+        };
+
+        let conversation_value = map
+            .remove("conversationId")
+            .context("event missing conversationId")?;
+        let notification_conversation: ConversationId = serde_json::from_value(conversation_value)
+            .context("conversationId was not a valid UUID")?;
+
+        if &notification_conversation != conversation_id {
+            return Ok(None);
+        }
+
+        let event_value = Value::Object(map);
+        let event: Event =
+            serde_json::from_value(event_value).context("failed to decode event payload")?;
+        Ok(Some(event))
+    }
+
+    fn send_request<T>(
+        &mut self,
+        request: ClientRequest,
+        request_id: RequestId,
+        method: &str,
+    ) -> Result<T>
+    where
+        T: DeserializeOwned,
+    {
+        self.write_request(&request)?;
+        self.wait_for_response(request_id, method)
+    }
+
+    fn write_request(&mut self, request: &ClientRequest) -> Result<()> {
+        let request_json = serde_json::to_string(request)?;
+        let request_pretty = serde_json::to_string_pretty(request)?;
+        print_multiline_with_prefix("> ", &request_pretty);
+
+        if let Some(stdin) = self.stdin.as_mut() {
+            writeln!(stdin, "{request_json}")?;
+            stdin
+                .flush()
+                .context("failed to flush request to codex app-server")?;
+        } else {
+            bail!("codex app-server stdin closed");
+        }
+
+        Ok(())
+    }
+
+    fn wait_for_response<T>(&mut self, request_id: RequestId, method: &str) -> Result<T>
+    where
+        T: DeserializeOwned,
+    {
+        loop {
+            let message = self.read_jsonrpc_message()?;
+
+            match message {
+                JSONRPCMessage::Response(JSONRPCResponse { id, result }) => {
+                    if id == request_id {
+                        return serde_json::from_value(result)
+                            .with_context(|| format!("{method} response missing payload"));
+                    }
+                }
+                JSONRPCMessage::Error(err) => {
+                    if err.id == request_id {
+                        bail!("{method} failed: {err:?}");
+                    }
+                }
+                JSONRPCMessage::Notification(notification) => {
+                    self.pending_notifications.push_back(notification);
+                }
+                JSONRPCMessage::Request(_) => {
+                    bail!("unexpected request from codex app-server");
+                }
+            }
+        }
+    }
+
+    fn next_notification(&mut self) -> Result<JSONRPCNotification> {
+        if let Some(notification) = self.pending_notifications.pop_front() {
+            return Ok(notification);
+        }
+
+        loop {
+            let message = self.read_jsonrpc_message()?;
+
+            match message {
+                JSONRPCMessage::Notification(notification) => return Ok(notification),
+                JSONRPCMessage::Response(_) | JSONRPCMessage::Error(_) => {
+                    // No outstanding requests, so ignore stray responses/errors for now.
+                    continue;
+                }
+                JSONRPCMessage::Request(_) => {
+                    bail!("unexpected request from codex app-server");
+                }
+            }
+        }
+    }
+
+    fn read_jsonrpc_message(&mut self) -> Result<JSONRPCMessage> {
+        loop {
+            let mut response_line = String::new();
+            let bytes = self
+                .stdout
+                .read_line(&mut response_line)
+                .context("failed to read from codex app-server")?;
+
+            if bytes == 0 {
+                bail!("codex app-server closed stdout");
+            }
+
+            let trimmed = response_line.trim();
+            if trimmed.is_empty() {
+                continue;
+            }
+
+            let parsed: Value =
+                serde_json::from_str(trimmed).context("response was not valid JSON-RPC")?;
+            let pretty = serde_json::to_string_pretty(&parsed)?;
+            print_multiline_with_prefix("< ", &pretty);
+            let message: JSONRPCMessage = serde_json::from_value(parsed)
+                .context("response was not a valid JSON-RPC message")?;
+            return Ok(message);
+        }
+    }
+
+    fn request_id(&self) -> RequestId {
+        RequestId::String(Uuid::new_v4().to_string())
+    }
+}
+
+fn print_multiline_with_prefix(prefix: &str, payload: &str) {
+    for line in payload.lines() {
+        println!("{prefix}{line}");
+    }
+}
+
+impl Drop for CodexClient {
+    fn drop(&mut self) {
+        let _ = self.stdin.take();
+
+        if let Ok(Some(status)) = self.child.try_wait() {
+            println!("[codex app-server exited: {status}]");
+            return;
+        }
+
+        thread::sleep(Duration::from_millis(100));
+
+        if let Ok(Some(status)) = self.child.try_wait() {
+            println!("[codex app-server exited: {status}]");
+            return;
+        }
+
+        let _ = self.child.kill();
+        let _ = self.child.wait();
+    }
+}

codex-rs/app-server/Cargo.toml

@@ -46,6 +46,7 @@ app_test_support = { workspace = true }
 assert_cmd = { workspace = true }
 base64 = { workspace = true }
 core_test_support = { workspace = true }
+mcp-types = { workspace = true }
 os_info = { workspace = true }
 pretty_assertions = { workspace = true }
 serial_test = { workspace = true }

codex-rs/app-server/README.md

@@ -1,6 +1,6 @@
 # codex-app-server
 
-`codex app-server` is the harness Codex uses to power rich interfaces such as the [Codex VS Code extension](https://marketplace.visualstudio.com/items?itemName=openai.chatgpt). The message schema is currently unstable, but those who wish to build experimental UIs on top of Codex may find it valuable.
+`codex app-server` is the interface Codex uses to power rich interfaces such as the [Codex VS Code extension](https://marketplace.visualstudio.com/items?itemName=openai.chatgpt). The message schema is currently unstable, but those who wish to build experimental UIs on top of Codex may find it valuable.
 
 ## Protocol
 
@@ -8,8 +8,253 @@ Similar to [MCP](https://modelcontextprotocol.io/), `codex app-server` supports
 
 ## Message Schema
 
-Currently, you can dump a TypeScript version of the schema using `codex generate-ts`. It is specific to the version of Codex you used to run `generate-ts`, so the two are guaranteed to be compatible.
+Currently, you can dump a TypeScript version of the schema using `codex app-server generate-ts`, or a JSON Schema bundle via `codex app-server generate-json-schema`. Each output is specific to the version of Codex you used to run the command, so the generated artifacts are guaranteed to match that version.
 
 ```
-codex generate-ts --out DIR
+codex app-server generate-ts --out DIR
+codex app-server generate-json-schema --out DIR
 ```
+
+## Initialization
+
+Clients must send a single `initialize` request before invoking any other method, then acknowledge with an `initialized` notification. The server returns the user agent string it will present to upstream services; subsequent requests issued before initialization receive a `"Not initialized"` error, and repeated `initialize` calls receive an `"Already initialized"` error.
+
+Example:
+
+```json
+{ "method": "initialize", "id": 0, "params": {
+    "clientInfo": { "name": "codex-vscode", "title": "Codex VS Code Extension", "version": "0.1.0" }
+} }
+{ "id": 0, "result": { "userAgent": "codex-app-server/0.1.0 codex-vscode/0.1.0" } }
+{ "method": "initialized" }
+```
+
+## Core primitives
+
+We have 3 top level primitives:
+- Thread - a conversation between the Codex agent and a user. Each thread contains multiple turns.
+- Turn - one turn of the conversation, typically starting with a user message and finishing with an agent message. Each turn contains multiple items.
+- Item - represents user inputs and agent outputs as part of the turn, persisted and used as the context for future conversations.
+
+## Thread & turn endpoints
+
+The JSON-RPC API exposes dedicated methods for managing Codex conversations. Threads store long-lived conversation metadata, and turns store the per-message exchange (input → Codex output, including streamed items). Use the thread APIs to create, list, or archive sessions, then drive the conversation with turn APIs and notifications.
+
+### Quick reference
+- `thread/start` — create a new thread; emits `thread/started` and auto-subscribes you to turn/item events for that thread.
+- `thread/resume` — reopen an existing thread by id so subsequent `turn/start` calls append to it.
+- `thread/list` — page through stored rollouts; supports cursor-based pagination and optional `modelProviders` filtering.
+- `thread/archive` — move a thread’s rollout file into the archived directory; returns `{}` on success.
+- `turn/start` — add user input to a thread and begin Codex generation; responds with the initial `turn` object and streams `turn/started`, `item/*`, and `turn/completed` notifications.
+- `turn/interrupt` — request cancellation of an in-flight turn by `(thread_id, turn_id)`; success is an empty `{}` response and the turn finishes with `status: "interrupted"`.
+
+### 1) Start or resume a thread
+
+Start a fresh thread when you need a new Codex conversation.
+
+```json
+{ "method": "thread/start", "id": 10, "params": {
+    // Optionally set config settings. If not specified, will use the user's
+    // current config settings.
+    "model": "gpt-5-codex",
+    "cwd": "/Users/me/project",
+    "approvalPolicy": "never",
+    "sandbox": "workspaceWrite",
+} }
+{ "id": 10, "result": {
+    "thread": {
+        "id": "thr_123",
+        "preview": "",
+        "modelProvider": "openai",
+        "createdAt": 1730910000
+    }
+} }
+{ "method": "thread/started", "params": { "thread": { … } } }
+```
+
+To continue a stored session, call `thread/resume` with the `thread.id` you previously recorded. The response shape matches `thread/start`, and no additional notifications are emitted:
+
+```json
+{ "method": "thread/resume", "id": 11, "params": { "threadId": "thr_123" } }
+{ "id": 11, "result": { "thread": { "id": "thr_123", … } } }
+```
+
+### 2) List threads (pagination & filters)
+
+`thread/list` lets you render a history UI. Pass any combination of:
+- `cursor` — opaque string from a prior response; omit for the first page.
+- `limit` — server defaults to a reasonable page size if unset.
+- `modelProviders` — restrict results to specific providers; unset, null, or an empty array will include all providers.
+
+Example:
+
+```json
+{ "method": "thread/list", "id": 20, "params": {
+    "cursor": null,
+    "limit": 25,
+} }
+{ "id": 20, "result": {
+    "data": [
+        { "id": "thr_a", "preview": "Create a TUI", "modelProvider": "openai", "createdAt": 1730831111 },
+        { "id": "thr_b", "preview": "Fix tests", "modelProvider": "openai", "createdAt": 1730750000 }
+    ],
+    "nextCursor": "opaque-token-or-null"
+} }
+```
+
+When `nextCursor` is `null`, you’ve reached the final page.
+
+### 3) Archive a thread
+
+Use `thread/archive` to move the persisted rollout (stored as a JSONL file on disk) into the archived sessions directory.
+
+```json
+{ "method": "thread/archive", "id": 21, "params": { "threadId": "thr_b" } }
+{ "id": 21, "result": {} }
+```
+
+An archived thread will not appear in future calls to `thread/list`.
+
+### 4) Start a turn (send user input)
+
+Turns attach user input (text or images) to a thread and trigger Codex generation. The `input` field is a list of discriminated unions:
+
+- `{"type":"text","text":"Explain this diff"}`
+- `{"type":"image","url":"https://…png"}`
+- `{"type":"localImage","path":"/tmp/screenshot.png"}`
+
+You can optionally specify config overrides on the new turn. If specified, these settings become the default for subsequent turns on the same thread.
+
+```json
+{ "method": "turn/start", "id": 30, "params": {
+    "threadId": "thr_123",
+    "input": [ { "type": "text", "text": "Run tests" } ],
+    // Below are optional config overrides
+    "cwd": "/Users/me/project",
+    "approvalPolicy": "unlessTrusted",
+    "sandboxPolicy": {
+        "mode": "workspaceWrite",
+        "writableRoots": ["/Users/me/project"],
+        "networkAccess": true
+    },
+    "model": "gpt-5-codex",
+    "effort": "medium",
+    "summary": "concise"
+} }
+{ "id": 30, "result": { "turn": {
+    "id": "turn_456",
+    "status": "inProgress",
+    "items": [],
+    "error": null
+} } }
+```
+
+### 5) Interrupt an active turn
+
+You can cancel a running Turn with `turn/interrupt`.
+
+```json
+{ "method": "turn/interrupt", "id": 31, "params": {
+    "threadId": "thr_123",
+    "turnId": "turn_456"
+} }
+{ "id": 31, "result": {} }
+```
+
+The server requests cancellations for running subprocesses, then emits a `turn/completed` event with `status: "interrupted"`. Rely on the `turn/completed` to know when Codex-side cleanup is done.
+
+## Auth endpoints
+
+The JSON-RPC auth/account surface exposes request/response methods plus server-initiated notifications (no `id`). Use these to determine auth state, start or cancel logins, logout, and inspect ChatGPT rate limits.
+
+### Quick reference
+- `account/read` — fetch current account info; optionally refresh tokens.
+- `account/login/start` — begin login (`apiKey` or `chatgpt`).
+- `account/login/completed` (notify) — emitted when a login attempt finishes (success or error).
+- `account/login/cancel` — cancel a pending ChatGPT login by `loginId`.
+- `account/logout` — sign out; triggers `account/updated`.
+- `account/updated` (notify) — emitted whenever auth mode changes (`authMode`: `apikey`, `chatgpt`, or `null`).
+- `account/rateLimits/read` — fetch ChatGPT rate limits; updates arrive via `account/rateLimits/updated` (notify).
+
+### 1) Check auth state
+
+Request:
+```json
+{ "method": "account/read", "id": 1, "params": { "refreshToken": false } }
+```
+
+Response examples:
+```json
+{ "id": 1, "result": { "account": null, "requiresOpenaiAuth": false } } // No OpenAI auth needed (e.g., OSS/local models)
+{ "id": 1, "result": { "account": null, "requiresOpenaiAuth": true } }  // OpenAI auth required (typical for OpenAI-hosted models)
+{ "id": 1, "result": { "account": { "type": "apiKey" }, "requiresOpenaiAuth": true } }
+{ "id": 1, "result": { "account": { "type": "chatgpt", "email": "user@example.com", "planType": "pro" }, "requiresOpenaiAuth": true } }
+```
+
+Field notes:
+- `refreshToken` (bool): set `true` to force a token refresh.
+- `requiresOpenaiAuth` reflects the active provider; when `false`, Codex can run without OpenAI credentials.
+
+### 2) Log in with an API key
+
+1. Send:
+   ```json
+   { "method": "account/login/start", "id": 2, "params": { "type": "apiKey", "apiKey": "sk-…" } }
+   ```
+2. Expect:
+   ```json
+   { "id": 2, "result": { "type": "apiKey" } }
+   ```
+3. Notifications:
+   ```json
+   { "method": "account/login/completed", "params": { "loginId": null, "success": true, "error": null } }
+   { "method": "account/updated", "params": { "authMode": "apikey" } }
+   ```
+
+### 3) Log in with ChatGPT (browser flow)
+
+1. Start:
+   ```json
+   { "method": "account/login/start", "id": 3, "params": { "type": "chatgpt" } }
+   { "id": 3, "result": { "type": "chatgpt", "loginId": "<uuid>", "authUrl": "https://chatgpt.com/…&redirect_uri=http%3A%2F%2Flocalhost%3A<port>%2Fauth%2Fcallback" } }
+   ```
+2. Open `authUrl` in a browser; the app-server hosts the local callback.
+3. Wait for notifications:
+   ```json
+   { "method": "account/login/completed", "params": { "loginId": "<uuid>", "success": true, "error": null } }
+   { "method": "account/updated", "params": { "authMode": "chatgpt" } }
+   ```
+
+### 4) Cancel a ChatGPT login
+
+```json
+{ "method": "account/login/cancel", "id": 4, "params": { "loginId": "<uuid>" } }
+{ "method": "account/login/completed", "params": { "loginId": "<uuid>", "success": false, "error": "…" } }
+```
+
+### 5) Logout
+
+```json
+{ "method": "account/logout", "id": 5 }
+{ "id": 5, "result": {} }
+{ "method": "account/updated", "params": { "authMode": null } }
+```
+
+### 6) Rate limits (ChatGPT)
+
+```json
+{ "method": "account/rateLimits/read", "id": 6 }
+{ "id": 6, "result": { "rateLimits": { "primary": { "usedPercent": 25, "windowDurationMins": 15, "resetsAt": 1730947200 }, "secondary": null } } }
+{ "method": "account/rateLimits/updated", "params": { "rateLimits": { … } } }
+```
+
+Field notes:
+- `usedPercent` is current usage within the OpenAI quota window.
+- `windowDurationMins` is the quota window length.
+- `resetsAt` is a Unix timestamp (seconds) for the next reset.
+
+### Dev notes
+
+- `codex app-server generate-ts --out <dir>` emits v2 types under `v2/`.
+- `codex app-server generate-json-schema --out <dir>` outputs `codex_app_server_protocol.schemas.json`.
+- See [“Authentication and authorization” in the config docs](../../docs/config.md#authentication-and-authorization) for configuration knobs.

codex-rs/app-server/src/bespoke_event_handling.rs

@@ -0,0 +1,483 @@
+use crate::codex_message_processor::ApiVersion;
+use crate::codex_message_processor::PendingInterrupts;
+use crate::outgoing_message::OutgoingMessageSender;
+use codex_app_server_protocol::AccountRateLimitsUpdatedNotification;
+use codex_app_server_protocol::AgentMessageDeltaNotification;
+use codex_app_server_protocol::ApplyPatchApprovalParams;
+use codex_app_server_protocol::ApplyPatchApprovalResponse;
+use codex_app_server_protocol::ExecCommandApprovalParams;
+use codex_app_server_protocol::ExecCommandApprovalResponse;
+use codex_app_server_protocol::InterruptConversationResponse;
+use codex_app_server_protocol::ItemCompletedNotification;
+use codex_app_server_protocol::ItemStartedNotification;
+use codex_app_server_protocol::McpToolCallError;
+use codex_app_server_protocol::McpToolCallResult;
+use codex_app_server_protocol::McpToolCallStatus;
+use codex_app_server_protocol::ReasoningSummaryPartAddedNotification;
+use codex_app_server_protocol::ReasoningSummaryTextDeltaNotification;
+use codex_app_server_protocol::ReasoningTextDeltaNotification;
+use codex_app_server_protocol::ServerNotification;
+use codex_app_server_protocol::ServerRequestPayload;
+use codex_app_server_protocol::ThreadItem;
+use codex_app_server_protocol::TurnInterruptResponse;
+use codex_core::CodexConversation;
+use codex_core::protocol::ApplyPatchApprovalRequestEvent;
+use codex_core::protocol::Event;
+use codex_core::protocol::EventMsg;
+use codex_core::protocol::ExecApprovalRequestEvent;
+use codex_core::protocol::McpToolCallBeginEvent;
+use codex_core::protocol::McpToolCallEndEvent;
+use codex_core::protocol::Op;
+use codex_core::protocol::ReviewDecision;
+use codex_protocol::ConversationId;
+use std::sync::Arc;
+use tokio::sync::oneshot;
+use tracing::error;
+
+type JsonRpcResult = serde_json::Value;
+
+pub(crate) async fn apply_bespoke_event_handling(
+    event: Event,
+    conversation_id: ConversationId,
+    conversation: Arc<CodexConversation>,
+    outgoing: Arc<OutgoingMessageSender>,
+    pending_interrupts: PendingInterrupts,
+) {
+    let Event { id: event_id, msg } = event;
+    match msg {
+        EventMsg::ApplyPatchApprovalRequest(ApplyPatchApprovalRequestEvent {
+            call_id,
+            changes,
+            reason,
+            grant_root,
+        }) => {
+            let params = ApplyPatchApprovalParams {
+                conversation_id,
+                call_id,
+                file_changes: changes,
+                reason,
+                grant_root,
+            };
+            let rx = outgoing
+                .send_request(ServerRequestPayload::ApplyPatchApproval(params))
+                .await;
+            // TODO(mbolin): Enforce a timeout so this task does not live indefinitely?
+            tokio::spawn(async move {
+                on_patch_approval_response(event_id, rx, conversation).await;
+            });
+        }
+        // TODO(celia): properly construct McpToolCall TurnItem in core.
+        EventMsg::McpToolCallBegin(begin_event) => {
+            let notification = construct_mcp_tool_call_notification(begin_event).await;
+            outgoing
+                .send_server_notification(ServerNotification::ItemStarted(notification))
+                .await;
+        }
+        EventMsg::McpToolCallEnd(end_event) => {
+            let notification = construct_mcp_tool_call_end_notification(end_event).await;
+            outgoing
+                .send_server_notification(ServerNotification::ItemCompleted(notification))
+                .await;
+        }
+        EventMsg::AgentMessageContentDelta(event) => {
+            let notification = AgentMessageDeltaNotification {
+                item_id: event.item_id,
+                delta: event.delta,
+            };
+            outgoing
+                .send_server_notification(ServerNotification::AgentMessageDelta(notification))
+                .await;
+        }
+        EventMsg::ReasoningContentDelta(event) => {
+            let notification = ReasoningSummaryTextDeltaNotification {
+                item_id: event.item_id,
+                delta: event.delta,
+                summary_index: event.summary_index,
+            };
+            outgoing
+                .send_server_notification(ServerNotification::ReasoningSummaryTextDelta(
+                    notification,
+                ))
+                .await;
+        }
+        EventMsg::ReasoningRawContentDelta(event) => {
+            let notification = ReasoningTextDeltaNotification {
+                item_id: event.item_id,
+                delta: event.delta,
+                content_index: event.content_index,
+            };
+            outgoing
+                .send_server_notification(ServerNotification::ReasoningTextDelta(notification))
+                .await;
+        }
+        EventMsg::AgentReasoningSectionBreak(event) => {
+            let notification = ReasoningSummaryPartAddedNotification {
+                item_id: event.item_id,
+                summary_index: event.summary_index,
+            };
+            outgoing
+                .send_server_notification(ServerNotification::ReasoningSummaryPartAdded(
+                    notification,
+                ))
+                .await;
+        }
+        EventMsg::ExecApprovalRequest(ExecApprovalRequestEvent {
+            call_id,
+            command,
+            cwd,
+            reason,
+            risk,
+            parsed_cmd,
+        }) => {
+            let params = ExecCommandApprovalParams {
+                conversation_id,
+                call_id,
+                command,
+                cwd,
+                reason,
+                risk,
+                parsed_cmd,
+            };
+            let rx = outgoing
+                .send_request(ServerRequestPayload::ExecCommandApproval(params))
+                .await;
+
+            // TODO(mbolin): Enforce a timeout so this task does not live indefinitely?
+            tokio::spawn(async move {
+                on_exec_approval_response(event_id, rx, conversation).await;
+            });
+        }
+        EventMsg::TokenCount(token_count_event) => {
+            if let Some(rate_limits) = token_count_event.rate_limits {
+                outgoing
+                    .send_server_notification(ServerNotification::AccountRateLimitsUpdated(
+                        AccountRateLimitsUpdatedNotification {
+                            rate_limits: rate_limits.into(),
+                        },
+                    ))
+                    .await;
+            }
+        }
+        EventMsg::ItemStarted(item_started_event) => {
+            let item: ThreadItem = item_started_event.item.clone().into();
+            let notification = ItemStartedNotification { item };
+            outgoing
+                .send_server_notification(ServerNotification::ItemStarted(notification))
+                .await;
+        }
+        EventMsg::ItemCompleted(item_completed_event) => {
+            let item: ThreadItem = item_completed_event.item.clone().into();
+            let notification = ItemCompletedNotification { item };
+            outgoing
+                .send_server_notification(ServerNotification::ItemCompleted(notification))
+                .await;
+        }
+        // If this is a TurnAborted, reply to any pending interrupt requests.
+        EventMsg::TurnAborted(turn_aborted_event) => {
+            let pending = {
+                let mut map = pending_interrupts.lock().await;
+                map.remove(&conversation_id).unwrap_or_default()
+            };
+            if !pending.is_empty() {
+                for (rid, ver) in pending {
+                    match ver {
+                        ApiVersion::V1 => {
+                            let response = InterruptConversationResponse {
+                                abort_reason: turn_aborted_event.reason.clone(),
+                            };
+                            outgoing.send_response(rid, response).await;
+                        }
+                        ApiVersion::V2 => {
+                            let response = TurnInterruptResponse {};
+                            outgoing.send_response(rid, response).await;
+                        }
+                    }
+                }
+            }
+        }
+
+        _ => {}
+    }
+}
+
+async fn on_patch_approval_response(
+    event_id: String,
+    receiver: oneshot::Receiver<JsonRpcResult>,
+    codex: Arc<CodexConversation>,
+) {
+    let response = receiver.await;
+    let value = match response {
+        Ok(value) => value,
+        Err(err) => {
+            error!("request failed: {err:?}");
+            if let Err(submit_err) = codex
+                .submit(Op::PatchApproval {
+                    id: event_id.clone(),
+                    decision: ReviewDecision::Denied,
+                })
+                .await
+            {
+                error!("failed to submit denied PatchApproval after request failure: {submit_err}");
+            }
+            return;
+        }
+    };
+
+    let response =
+        serde_json::from_value::<ApplyPatchApprovalResponse>(value).unwrap_or_else(|err| {
+            error!("failed to deserialize ApplyPatchApprovalResponse: {err}");
+            ApplyPatchApprovalResponse {
+                decision: ReviewDecision::Denied,
+            }
+        });
+
+    if let Err(err) = codex
+        .submit(Op::PatchApproval {
+            id: event_id,
+            decision: response.decision,
+        })
+        .await
+    {
+        error!("failed to submit PatchApproval: {err}");
+    }
+}
+
+async fn on_exec_approval_response(
+    event_id: String,
+    receiver: oneshot::Receiver<JsonRpcResult>,
+    conversation: Arc<CodexConversation>,
+) {
+    let response = receiver.await;
+    let value = match response {
+        Ok(value) => value,
+        Err(err) => {
+            error!("request failed: {err:?}");
+            return;
+        }
+    };
+
+    // Try to deserialize `value` and then make the appropriate call to `codex`.
+    let response =
+        serde_json::from_value::<ExecCommandApprovalResponse>(value).unwrap_or_else(|err| {
+            error!("failed to deserialize ExecCommandApprovalResponse: {err}");
+            // If we cannot deserialize the response, we deny the request to be
+            // conservative.
+            ExecCommandApprovalResponse {
+                decision: ReviewDecision::Denied,
+            }
+        });
+
+    if let Err(err) = conversation
+        .submit(Op::ExecApproval {
+            id: event_id,
+            decision: response.decision,
+        })
+        .await
+    {
+        error!("failed to submit ExecApproval: {err}");
+    }
+}
+
+/// similar to handle_mcp_tool_call_begin in exec
+async fn construct_mcp_tool_call_notification(
+    begin_event: McpToolCallBeginEvent,
+) -> ItemStartedNotification {
+    let item = ThreadItem::McpToolCall {
+        id: begin_event.call_id,
+        server: begin_event.invocation.server,
+        tool: begin_event.invocation.tool,
+        status: McpToolCallStatus::InProgress,
+        arguments: begin_event
+            .invocation
+            .arguments
+            .unwrap_or(JsonRpcResult::Null),
+        result: None,
+        error: None,
+    };
+    ItemStartedNotification { item }
+}
+
+/// simiilar to handle_mcp_tool_call_end in exec
+async fn construct_mcp_tool_call_end_notification(
+    end_event: McpToolCallEndEvent,
+) -> ItemCompletedNotification {
+    let status = if end_event.is_success() {
+        McpToolCallStatus::Completed
+    } else {
+        McpToolCallStatus::Failed
+    };
+
+    let (result, error) = match &end_event.result {
+        Ok(value) => (
+            Some(McpToolCallResult {
+                content: value.content.clone(),
+                structured_content: value.structured_content.clone(),
+            }),
+            None,
+        ),
+        Err(message) => (
+            None,
+            Some(McpToolCallError {
+                message: message.clone(),
+            }),
+        ),
+    };
+
+    let item = ThreadItem::McpToolCall {
+        id: end_event.call_id,
+        server: end_event.invocation.server,
+        tool: end_event.invocation.tool,
+        status,
+        arguments: end_event
+            .invocation
+            .arguments
+            .unwrap_or(JsonRpcResult::Null),
+        result,
+        error,
+    };
+    ItemCompletedNotification { item }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use codex_core::protocol::McpInvocation;
+    use mcp_types::CallToolResult;
+    use mcp_types::ContentBlock;
+    use mcp_types::TextContent;
+    use pretty_assertions::assert_eq;
+    use serde_json::Value as JsonValue;
+    use std::time::Duration;
+
+    #[tokio::test]
+    async fn test_construct_mcp_tool_call_begin_notification_with_args() {
+        let begin_event = McpToolCallBeginEvent {
+            call_id: "call_123".to_string(),
+            invocation: McpInvocation {
+                server: "codex".to_string(),
+                tool: "list_mcp_resources".to_string(),
+                arguments: Some(serde_json::json!({"server": ""})),
+            },
+        };
+
+        let notification = construct_mcp_tool_call_notification(begin_event.clone()).await;
+
+        let expected = ItemStartedNotification {
+            item: ThreadItem::McpToolCall {
+                id: begin_event.call_id,
+                server: begin_event.invocation.server,
+                tool: begin_event.invocation.tool,
+                status: McpToolCallStatus::InProgress,
+                arguments: serde_json::json!({"server": ""}),
+                result: None,
+                error: None,
+            },
+        };
+
+        assert_eq!(notification, expected);
+    }
+
+    #[tokio::test]
+    async fn test_construct_mcp_tool_call_begin_notification_without_args() {
+        let begin_event = McpToolCallBeginEvent {
+            call_id: "call_456".to_string(),
+            invocation: McpInvocation {
+                server: "codex".to_string(),
+                tool: "list_mcp_resources".to_string(),
+                arguments: None,
+            },
+        };
+
+        let notification = construct_mcp_tool_call_notification(begin_event.clone()).await;
+
+        let expected = ItemStartedNotification {
+            item: ThreadItem::McpToolCall {
+                id: begin_event.call_id,
+                server: begin_event.invocation.server,
+                tool: begin_event.invocation.tool,
+                status: McpToolCallStatus::InProgress,
+                arguments: JsonValue::Null,
+                result: None,
+                error: None,
+            },
+        };
+
+        assert_eq!(notification, expected);
+    }
+
+    #[tokio::test]
+    async fn test_construct_mcp_tool_call_end_notification_success() {
+        let content = vec![ContentBlock::TextContent(TextContent {
+            annotations: None,
+            text: "{\"resources\":[]}".to_string(),
+            r#type: "text".to_string(),
+        })];
+        let result = CallToolResult {
+            content: content.clone(),
+            is_error: Some(false),
+            structured_content: None,
+        };
+
+        let end_event = McpToolCallEndEvent {
+            call_id: "call_789".to_string(),
+            invocation: McpInvocation {
+                server: "codex".to_string(),
+                tool: "list_mcp_resources".to_string(),
+                arguments: Some(serde_json::json!({"server": ""})),
+            },
+            duration: Duration::from_nanos(92708),
+            result: Ok(result),
+        };
+
+        let notification = construct_mcp_tool_call_end_notification(end_event.clone()).await;
+
+        let expected = ItemCompletedNotification {
+            item: ThreadItem::McpToolCall {
+                id: end_event.call_id,
+                server: end_event.invocation.server,
+                tool: end_event.invocation.tool,
+                status: McpToolCallStatus::Completed,
+                arguments: serde_json::json!({"server": ""}),
+                result: Some(McpToolCallResult {
+                    content,
+                    structured_content: None,
+                }),
+                error: None,
+            },
+        };
+
+        assert_eq!(notification, expected);
+    }
+
+    #[tokio::test]
+    async fn test_construct_mcp_tool_call_end_notification_error() {
+        let end_event = McpToolCallEndEvent {
+            call_id: "call_err".to_string(),
+            invocation: McpInvocation {
+                server: "codex".to_string(),
+                tool: "list_mcp_resources".to_string(),
+                arguments: None,
+            },
+            duration: Duration::from_millis(1),
+            result: Err("boom".to_string()),
+        };
+
+        let notification = construct_mcp_tool_call_end_notification(end_event.clone()).await;
+
+        let expected = ItemCompletedNotification {
+            item: ThreadItem::McpToolCall {
+                id: end_event.call_id,
+                server: end_event.invocation.server,
+                tool: end_event.invocation.tool,
+                status: McpToolCallStatus::Failed,
+                arguments: JsonValue::Null,
+                result: None,
+                error: Some(McpToolCallError {
+                    message: "boom".to_string(),
+                }),
+            },
+        };
+
+        assert_eq!(notification, expected);
+    }
+}

codex-rs/app-server/src/lib.rs

@@ -28,6 +28,7 @@ use tracing_subscriber::filter::Targets;
 use tracing_subscriber::layer::SubscriberExt;
 use tracing_subscriber::util::SubscriberInitExt;
 
+mod bespoke_event_handling;
 mod codex_message_processor;
 mod error_code;
 mod fuzzy_file_search;

codex-rs/app-server/src/message_processor.rs

@@ -64,64 +64,79 @@ impl MessageProcessor {
 
     pub(crate) async fn process_request(&mut self, request: JSONRPCRequest) {
         let request_id = request.id.clone();
-        if let Ok(request_json) = serde_json::to_value(request)
-            && let Ok(codex_request) = serde_json::from_value::<ClientRequest>(request_json)
-        {
-            match codex_request {
-                // Handle Initialize internally so CodexMessageProcessor does not have to concern
-                // itself with the `initialized` bool.
-                ClientRequest::Initialize { request_id, params } => {
-                    if self.initialized {
-                        let error = JSONRPCErrorError {
-                            code: INVALID_REQUEST_ERROR_CODE,
-                            message: "Already initialized".to_string(),
-                            data: None,
-                        };
-                        self.outgoing.send_error(request_id, error).await;
-                        return;
-                    } else {
-                        let ClientInfo {
-                            name,
-                            title: _title,
-                            version,
-                        } = params.client_info;
-                        let user_agent_suffix = format!("{name}; {version}");
-                        if let Ok(mut suffix) = USER_AGENT_SUFFIX.lock() {
-                            *suffix = Some(user_agent_suffix);
-                        }
+        let request_json = match serde_json::to_value(&request) {
+            Ok(request_json) => request_json,
+            Err(err) => {
+                let error = JSONRPCErrorError {
+                    code: INVALID_REQUEST_ERROR_CODE,
+                    message: format!("Invalid request: {err}"),
+                    data: None,
+                };
+                self.outgoing.send_error(request_id, error).await;
+                return;
+            }
+        };
 
-                        let user_agent = get_codex_user_agent();
-                        let response = InitializeResponse { user_agent };
-                        self.outgoing.send_response(request_id, response).await;
+        let codex_request = match serde_json::from_value::<ClientRequest>(request_json) {
+            Ok(codex_request) => codex_request,
+            Err(err) => {
+                let error = JSONRPCErrorError {
+                    code: INVALID_REQUEST_ERROR_CODE,
+                    message: format!("Invalid request: {err}"),
+                    data: None,
+                };
+                self.outgoing.send_error(request_id, error).await;
+                return;
+            }
+        };
 
-                        self.initialized = true;
-                        return;
-                    }
-                }
-                _ => {
-                    if !self.initialized {
-                        let error = JSONRPCErrorError {
-                            code: INVALID_REQUEST_ERROR_CODE,
-                            message: "Not initialized".to_string(),
-                            data: None,
-                        };
-                        self.outgoing.send_error(request_id, error).await;
-                        return;
+        match codex_request {
+            // Handle Initialize internally so CodexMessageProcessor does not have to concern
+            // itself with the `initialized` bool.
+            ClientRequest::Initialize { request_id, params } => {
+                if self.initialized {
+                    let error = JSONRPCErrorError {
+                        code: INVALID_REQUEST_ERROR_CODE,
+                        message: "Already initialized".to_string(),
+                        data: None,
+                    };
+                    self.outgoing.send_error(request_id, error).await;
+                    return;
+                } else {
+                    let ClientInfo {
+                        name,
+                        title: _title,
+                        version,
+                    } = params.client_info;
+                    let user_agent_suffix = format!("{name}; {version}");
+                    if let Ok(mut suffix) = USER_AGENT_SUFFIX.lock() {
+                        *suffix = Some(user_agent_suffix);
                     }
+
+                    let user_agent = get_codex_user_agent();
+                    let response = InitializeResponse { user_agent };
+                    self.outgoing.send_response(request_id, response).await;
+
+                    self.initialized = true;
+                    return;
+                }
+            }
+            _ => {
+                if !self.initialized {
+                    let error = JSONRPCErrorError {
+                        code: INVALID_REQUEST_ERROR_CODE,
+                        message: "Not initialized".to_string(),
+                        data: None,
+                    };
+                    self.outgoing.send_error(request_id, error).await;
+                    return;
                 }
             }
-
-            self.codex_message_processor
-                .process_request(codex_request)
-                .await;
-        } else {
-            let error = JSONRPCErrorError {
-                code: INVALID_REQUEST_ERROR_CODE,
-                message: "Invalid request".to_string(),
-                data: None,
-            };
-            self.outgoing.send_error(request_id, error).await;
         }
+
+        self.codex_message_processor
+            .process_request(codex_request)
+            .await;
     }
 
     pub(crate) async fn process_notification(&self, notification: JSONRPCNotification) {

codex-rs/app-server/src/models.rs

@@ -1,11 +1,12 @@
+use codex_app_server_protocol::AuthMode;
 use codex_app_server_protocol::Model;
 use codex_app_server_protocol::ReasoningEffortOption;
 use codex_common::model_presets::ModelPreset;
 use codex_common::model_presets::ReasoningEffortPreset;
 use codex_common::model_presets::builtin_model_presets;
 
-pub fn supported_models() -> Vec<Model> {
-    builtin_model_presets(None)
+pub fn supported_models(auth_mode: Option<AuthMode>) -> Vec<Model> {
+    builtin_model_presets(auth_mode)
         .into_iter()
         .map(model_from_preset)
         .collect()

codex-rs/app-server/src/outgoing_message.rs

@@ -141,9 +141,13 @@ pub(crate) struct OutgoingError {
 
 #[cfg(test)]
 mod tests {
+    use codex_app_server_protocol::AccountLoginCompletedNotification;
+    use codex_app_server_protocol::AccountRateLimitsUpdatedNotification;
+    use codex_app_server_protocol::AccountUpdatedNotification;
+    use codex_app_server_protocol::AuthMode;
     use codex_app_server_protocol::LoginChatGptCompleteNotification;
-    use codex_protocol::protocol::RateLimitSnapshot;
-    use codex_protocol::protocol::RateLimitWindow;
+    use codex_app_server_protocol::RateLimitSnapshot;
+    use codex_app_server_protocol::RateLimitWindow;
     use pretty_assertions::assert_eq;
     use serde_json::json;
     use uuid::Uuid;
@@ -176,27 +180,77 @@ mod tests {
     }
 
     #[test]
-    fn verify_account_rate_limits_notification_serialization() {
-        let notification = ServerNotification::AccountRateLimitsUpdated(RateLimitSnapshot {
-            primary: Some(RateLimitWindow {
-                used_percent: 25.0,
-                window_minutes: Some(15),
-                resets_at: Some(123),
+    fn verify_account_login_completed_notification_serialization() {
+        let notification =
+            ServerNotification::AccountLoginCompleted(AccountLoginCompletedNotification {
+                login_id: Some(Uuid::nil().to_string()),
+                success: true,
+                error: None,
+            });
+
+        let jsonrpc_notification = OutgoingMessage::AppServerNotification(notification);
+        assert_eq!(
+            json!({
+                "method": "account/login/completed",
+                "params": {
+                    "loginId": Uuid::nil().to_string(),
+                    "success": true,
+                    "error": null,
+                },
             }),
-            secondary: None,
-        });
+            serde_json::to_value(jsonrpc_notification)
+                .expect("ensure the notification serializes correctly"),
+            "ensure the notification serializes correctly"
+        );
+    }
+
+    #[test]
+    fn verify_account_rate_limits_notification_serialization() {
+        let notification =
+            ServerNotification::AccountRateLimitsUpdated(AccountRateLimitsUpdatedNotification {
+                rate_limits: RateLimitSnapshot {
+                    primary: Some(RateLimitWindow {
+                        used_percent: 25,
+                        window_duration_mins: Some(15),
+                        resets_at: Some(123),
+                    }),
+                    secondary: None,
+                },
+            });
 
         let jsonrpc_notification = OutgoingMessage::AppServerNotification(notification);
         assert_eq!(
             json!({
                 "method": "account/rateLimits/updated",
                 "params": {
-                    "primary": {
-                        "used_percent": 25.0,
-                        "window_minutes": 15,
-                        "resets_at": 123,
-                    },
-                    "secondary": null,
+                    "rateLimits": {
+                        "primary": {
+                            "usedPercent": 25,
+                            "windowDurationMins": 15,
+                            "resetsAt": 123
+                        },
+                        "secondary": null
+                    }
+                },
+            }),
+            serde_json::to_value(jsonrpc_notification)
+                .expect("ensure the notification serializes correctly"),
+            "ensure the notification serializes correctly"
+        );
+    }
+
+    #[test]
+    fn verify_account_updated_notification_serialization() {
+        let notification = ServerNotification::AccountUpdated(AccountUpdatedNotification {
+            auth_mode: Some(AuthMode::ApiKey),
+        });
+
+        let jsonrpc_notification = OutgoingMessage::AppServerNotification(notification);
+        assert_eq!(
+            json!({
+                "method": "account/updated",
+                "params": {
+                    "authMode": "apikey"
                 },
             }),
             serde_json::to_value(jsonrpc_notification)

codex-rs/app-server/tests/common/Cargo.toml

@@ -13,6 +13,7 @@ base64 = { workspace = true }
 chrono = { workspace = true }
 codex-app-server-protocol = { workspace = true }
 codex-core = { workspace = true }
+codex-protocol = { workspace = true }
 serde = { workspace = true }
 serde_json = { workspace = true }
 tokio = { workspace = true, features = [
@@ -21,4 +22,5 @@ tokio = { workspace = true, features = [
     "process",
     "rt-multi-thread",
 ] }
+uuid = { workspace = true }
 wiremock = { workspace = true }

codex-rs/app-server/tests/common/lib.rs

@@ -2,6 +2,7 @@ mod auth_fixtures;
 mod mcp_process;
 mod mock_model_server;
 mod responses;
+mod rollout;
 
 pub use auth_fixtures::ChatGptAuthFixture;
 pub use auth_fixtures::ChatGptIdTokenClaims;
@@ -10,9 +11,11 @@ pub use auth_fixtures::write_chatgpt_auth;
 use codex_app_server_protocol::JSONRPCResponse;
 pub use mcp_process::McpProcess;
 pub use mock_model_server::create_mock_chat_completions_server;
+pub use mock_model_server::create_mock_chat_completions_server_unchecked;
 pub use responses::create_apply_patch_sse_response;
 pub use responses::create_final_assistant_message_sse_response;
 pub use responses::create_shell_sse_response;
+pub use rollout::create_fake_rollout;
 use serde::de::DeserializeOwned;
 
 pub fn to_response<T: DeserializeOwned>(response: JSONRPCResponse) -> anyhow::Result<T> {

codex-rs/app-server/tests/common/mcp_process.rs

@@ -14,30 +14,37 @@ use anyhow::Context;
 use assert_cmd::prelude::*;
 use codex_app_server_protocol::AddConversationListenerParams;
 use codex_app_server_protocol::ArchiveConversationParams;
+use codex_app_server_protocol::CancelLoginAccountParams;
 use codex_app_server_protocol::CancelLoginChatGptParams;
 use codex_app_server_protocol::ClientInfo;
 use codex_app_server_protocol::ClientNotification;
+use codex_app_server_protocol::FeedbackUploadParams;
+use codex_app_server_protocol::GetAccountParams;
 use codex_app_server_protocol::GetAuthStatusParams;
 use codex_app_server_protocol::InitializeParams;
 use codex_app_server_protocol::InterruptConversationParams;
-use codex_app_server_protocol::ListConversationsParams;
-use codex_app_server_protocol::ListModelsParams;
-use codex_app_server_protocol::LoginApiKeyParams;
-use codex_app_server_protocol::NewConversationParams;
-use codex_app_server_protocol::RemoveConversationListenerParams;
-use codex_app_server_protocol::ResumeConversationParams;
-use codex_app_server_protocol::SendUserMessageParams;
-use codex_app_server_protocol::SendUserTurnParams;
-use codex_app_server_protocol::ServerRequest;
-use codex_app_server_protocol::SetDefaultModelParams;
-use codex_app_server_protocol::UploadFeedbackParams;
-
 use codex_app_server_protocol::JSONRPCError;
 use codex_app_server_protocol::JSONRPCMessage;
 use codex_app_server_protocol::JSONRPCNotification;
 use codex_app_server_protocol::JSONRPCRequest;
 use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::ListConversationsParams;
+use codex_app_server_protocol::LoginApiKeyParams;
+use codex_app_server_protocol::ModelListParams;
+use codex_app_server_protocol::NewConversationParams;
+use codex_app_server_protocol::RemoveConversationListenerParams;
 use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::ResumeConversationParams;
+use codex_app_server_protocol::SendUserMessageParams;
+use codex_app_server_protocol::SendUserTurnParams;
+use codex_app_server_protocol::ServerRequest;
+use codex_app_server_protocol::SetDefaultModelParams;
+use codex_app_server_protocol::ThreadArchiveParams;
+use codex_app_server_protocol::ThreadListParams;
+use codex_app_server_protocol::ThreadResumeParams;
+use codex_app_server_protocol::ThreadStartParams;
+use codex_app_server_protocol::TurnInterruptParams;
+use codex_app_server_protocol::TurnStartParams;
 use std::process::Command as StdCommand;
 use tokio::process::Command;
 
@@ -243,10 +250,19 @@ impl McpProcess {
         self.send_request("account/rateLimits/read", None).await
     }
 
-    /// Send a `feedback/upload` JSON-RPC request.
-    pub async fn send_upload_feedback_request(
+    /// Send an `account/read` JSON-RPC request.
+    pub async fn send_get_account_request(
         &mut self,
-        params: UploadFeedbackParams,
+        params: GetAccountParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("account/read", params).await
+    }
+
+    /// Send a `feedback/upload` JSON-RPC request.
+    pub async fn send_feedback_upload_request(
+        &mut self,
+        params: FeedbackUploadParams,
     ) -> anyhow::Result<i64> {
         let params = Some(serde_json::to_value(params)?);
         self.send_request("feedback/upload", params).await
@@ -275,10 +291,46 @@ impl McpProcess {
         self.send_request("listConversations", params).await
     }
 
+    /// Send a `thread/start` JSON-RPC request.
+    pub async fn send_thread_start_request(
+        &mut self,
+        params: ThreadStartParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("thread/start", params).await
+    }
+
+    /// Send a `thread/resume` JSON-RPC request.
+    pub async fn send_thread_resume_request(
+        &mut self,
+        params: ThreadResumeParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("thread/resume", params).await
+    }
+
+    /// Send a `thread/archive` JSON-RPC request.
+    pub async fn send_thread_archive_request(
+        &mut self,
+        params: ThreadArchiveParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("thread/archive", params).await
+    }
+
+    /// Send a `thread/list` JSON-RPC request.
+    pub async fn send_thread_list_request(
+        &mut self,
+        params: ThreadListParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("thread/list", params).await
+    }
+
     /// Send a `model/list` JSON-RPC request.
     pub async fn send_list_models_request(
         &mut self,
-        params: ListModelsParams,
+        params: ModelListParams,
     ) -> anyhow::Result<i64> {
         let params = Some(serde_json::to_value(params)?);
         self.send_request("model/list", params).await
@@ -307,6 +359,24 @@ impl McpProcess {
         self.send_request("loginChatGpt", None).await
     }
 
+    /// Send a `turn/start` JSON-RPC request (v2).
+    pub async fn send_turn_start_request(
+        &mut self,
+        params: TurnStartParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("turn/start", params).await
+    }
+
+    /// Send a `turn/interrupt` JSON-RPC request (v2).
+    pub async fn send_turn_interrupt_request(
+        &mut self,
+        params: TurnInterruptParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("turn/interrupt", params).await
+    }
+
     /// Send a `cancelLoginChatGpt` JSON-RPC request.
     pub async fn send_cancel_login_chat_gpt_request(
         &mut self,
@@ -321,6 +391,40 @@ impl McpProcess {
         self.send_request("logoutChatGpt", None).await
     }
 
+    /// Send an `account/logout` JSON-RPC request.
+    pub async fn send_logout_account_request(&mut self) -> anyhow::Result<i64> {
+        self.send_request("account/logout", None).await
+    }
+
+    /// Send an `account/login/start` JSON-RPC request for API key login.
+    pub async fn send_login_account_api_key_request(
+        &mut self,
+        api_key: &str,
+    ) -> anyhow::Result<i64> {
+        let params = serde_json::json!({
+            "type": "apiKey",
+            "apiKey": api_key,
+        });
+        self.send_request("account/login/start", Some(params)).await
+    }
+
+    /// Send an `account/login/start` JSON-RPC request for ChatGPT login.
+    pub async fn send_login_account_chatgpt_request(&mut self) -> anyhow::Result<i64> {
+        let params = serde_json::json!({
+            "type": "chatgpt"
+        });
+        self.send_request("account/login/start", Some(params)).await
+    }
+
+    /// Send an `account/login/cancel` JSON-RPC request.
+    pub async fn send_cancel_login_account_request(
+        &mut self,
+        params: CancelLoginAccountParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("account/login/cancel", params).await
+    }
+
     /// Send a `fuzzyFileSearch` JSON-RPC request.
     pub async fn send_fuzzy_file_search_request(
         &mut self,

codex-rs/app-server/tests/common/mock_model_server.rs

@@ -29,6 +29,25 @@ pub async fn create_mock_chat_completions_server(responses: Vec<String>) -> Mock
     server
 }
 
+/// Same as `create_mock_chat_completions_server` but does not enforce an
+/// expectation on the number of calls.
+pub async fn create_mock_chat_completions_server_unchecked(responses: Vec<String>) -> MockServer {
+    let server = MockServer::start().await;
+
+    let seq_responder = SeqResponder {
+        num_calls: AtomicUsize::new(0),
+        responses,
+    };
+
+    Mock::given(method("POST"))
+        .and(path("/v1/chat/completions"))
+        .respond_with(seq_responder)
+        .mount(&server)
+        .await;
+
+    server
+}
+
 struct SeqResponder {
     num_calls: AtomicUsize,
     responses: Vec<String>,

codex-rs/app-server/tests/common/rollout.rs

@@ -0,0 +1,82 @@
+use anyhow::Result;
+use codex_protocol::ConversationId;
+use codex_protocol::protocol::SessionMeta;
+use codex_protocol::protocol::SessionSource;
+use serde_json::json;
+use std::fs;
+use std::path::Path;
+use std::path::PathBuf;
+use uuid::Uuid;
+
+/// Create a minimal rollout file under `CODEX_HOME/sessions/YYYY/MM/DD/`.
+///
+/// - `filename_ts` is the filename timestamp component in `YYYY-MM-DDThh-mm-ss` format.
+/// - `meta_rfc3339` is the envelope timestamp used in JSON lines.
+/// - `preview` is the user message preview text.
+/// - `model_provider` optionally sets the provider in the session meta payload.
+///
+/// Returns the generated conversation/session UUID as a string.
+pub fn create_fake_rollout(
+    codex_home: &Path,
+    filename_ts: &str,
+    meta_rfc3339: &str,
+    preview: &str,
+    model_provider: Option<&str>,
+) -> Result<String> {
+    let uuid = Uuid::new_v4();
+    let uuid_str = uuid.to_string();
+    let conversation_id = ConversationId::from_string(&uuid_str)?;
+
+    // sessions/YYYY/MM/DD derived from filename_ts (YYYY-MM-DDThh-mm-ss)
+    let year = &filename_ts[0..4];
+    let month = &filename_ts[5..7];
+    let day = &filename_ts[8..10];
+    let dir = codex_home.join("sessions").join(year).join(month).join(day);
+    fs::create_dir_all(&dir)?;
+
+    let file_path = dir.join(format!("rollout-{filename_ts}-{uuid}.jsonl"));
+
+    // Build JSONL lines
+    let payload = serde_json::to_value(SessionMeta {
+        id: conversation_id,
+        timestamp: meta_rfc3339.to_string(),
+        cwd: PathBuf::from("/"),
+        originator: "codex".to_string(),
+        cli_version: "0.0.0".to_string(),
+        instructions: None,
+        source: SessionSource::Cli,
+        model_provider: model_provider.map(str::to_string),
+    })?;
+
+    let lines = [
+        json!({
+            "timestamp": meta_rfc3339,
+            "type": "session_meta",
+            "payload": payload
+        })
+        .to_string(),
+        json!({
+            "timestamp": meta_rfc3339,
+            "type":"response_item",
+            "payload": {
+                "type":"message",
+                "role":"user",
+                "content":[{"type":"input_text","text": preview}]
+            }
+        })
+        .to_string(),
+        json!({
+            "timestamp": meta_rfc3339,
+            "type":"event_msg",
+            "payload": {
+                "type":"user_message",
+                "message": preview,
+                "kind": "plain"
+            }
+        })
+        .to_string(),
+    ];
+
+    fs::write(file_path, lines.join("\n") + "\n")?;
+    Ok(uuid_str)
+}

codex-rs/app-server/tests/suite/archive_conversation.rs

@@ -12,7 +12,7 @@ use std::path::Path;
 use tempfile::TempDir;
 use tokio::time::timeout;
 
-const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
+const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(20);
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn archive_conversation_moves_rollout_into_archived_directory() -> Result<()> {

codex-rs/app-server/tests/suite/interrupt.rs

@@ -146,7 +146,7 @@ fn create_config_toml(codex_home: &Path, server_uri: String) -> std::io::Result<
             r#"
 model = "mock-model"
 approval_policy = "never"
-sandbox_mode = "danger-full-access"
+sandbox_mode = "read-only"
 
 model_provider = "mock_provider"
 

codex-rs/app-server/tests/suite/list_resume.rs

@@ -1,5 +1,6 @@
 use anyhow::Result;
 use app_test_support::McpProcess;
+use app_test_support::create_fake_rollout;
 use app_test_support::to_response;
 use codex_app_server_protocol::JSONRPCNotification;
 use codex_app_server_protocol::JSONRPCResponse;
@@ -15,12 +16,8 @@ use codex_core::protocol::EventMsg;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ResponseItem;
 use pretty_assertions::assert_eq;
-use serde_json::json;
-use std::fs;
-use std::path::Path;
 use tempfile::TempDir;
 use tokio::time::timeout;
-use uuid::Uuid;
 
 const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
 
@@ -357,70 +354,3 @@ async fn test_list_and_resume_conversations() -> Result<()> {
 
     Ok(())
 }
-
-fn create_fake_rollout(
-    codex_home: &Path,
-    filename_ts: &str,
-    meta_rfc3339: &str,
-    preview: &str,
-    model_provider: Option<&str>,
-) -> Result<()> {
-    let uuid = Uuid::new_v4();
-    // sessions/YYYY/MM/DD/ derived from filename_ts (YYYY-MM-DDThh-mm-ss)
-    let year = &filename_ts[0..4];
-    let month = &filename_ts[5..7];
-    let day = &filename_ts[8..10];
-    let dir = codex_home.join("sessions").join(year).join(month).join(day);
-    fs::create_dir_all(&dir)?;
-
-    let file_path = dir.join(format!("rollout-{filename_ts}-{uuid}.jsonl"));
-    let mut lines = Vec::new();
-    // Meta line with timestamp (flattened meta in payload for new schema)
-    let mut payload = json!({
-        "id": uuid,
-        "timestamp": meta_rfc3339,
-        "cwd": "/",
-        "originator": "codex",
-        "cli_version": "0.0.0",
-        "instructions": null,
-    });
-    if let Some(provider) = model_provider {
-        payload["model_provider"] = json!(provider);
-    }
-    lines.push(
-        json!({
-            "timestamp": meta_rfc3339,
-            "type": "session_meta",
-            "payload": payload
-        })
-        .to_string(),
-    );
-    // Minimal user message entry as a persisted response item (with envelope timestamp)
-    lines.push(
-        json!({
-            "timestamp": meta_rfc3339,
-            "type":"response_item",
-            "payload": {
-                "type":"message",
-                "role":"user",
-                "content":[{"type":"input_text","text": preview}]
-            }
-        })
-        .to_string(),
-    );
-    // Add a matching user message event line to satisfy filters
-    lines.push(
-        json!({
-            "timestamp": meta_rfc3339,
-            "type":"event_msg",
-            "payload": {
-                "type":"user_message",
-                "message": preview,
-                "kind": "plain"
-            }
-        })
-        .to_string(),
-    );
-    fs::write(file_path, lines.join("\n") + "\n")?;
-    Ok(())
-}

codex-rs/app-server/tests/suite/mod.rs

@@ -7,9 +7,8 @@ mod fuzzy_file_search;
 mod interrupt;
 mod list_resume;
 mod login;
-mod model_list;
-mod rate_limits;
 mod send_message;
 mod set_default_model;
 mod user_agent;
 mod user_info;
+mod v2;

codex-rs/app-server/tests/suite/v2/account.rs

@@ -0,0 +1,492 @@
+use anyhow::Result;
+use anyhow::bail;
+use app_test_support::McpProcess;
+use app_test_support::to_response;
+
+use app_test_support::ChatGptAuthFixture;
+use app_test_support::write_chatgpt_auth;
+use codex_app_server_protocol::Account;
+use codex_app_server_protocol::AuthMode;
+use codex_app_server_protocol::CancelLoginAccountParams;
+use codex_app_server_protocol::CancelLoginAccountResponse;
+use codex_app_server_protocol::GetAccountParams;
+use codex_app_server_protocol::GetAccountResponse;
+use codex_app_server_protocol::JSONRPCError;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::LoginAccountResponse;
+use codex_app_server_protocol::LogoutAccountResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::ServerNotification;
+use codex_core::auth::AuthCredentialsStoreMode;
+use codex_login::login_with_api_key;
+use codex_protocol::account::PlanType as AccountPlanType;
+use pretty_assertions::assert_eq;
+use serial_test::serial;
+use std::path::Path;
+use std::time::Duration;
+use tempfile::TempDir;
+use tokio::time::timeout;
+
+const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
+
+// Helper to create a minimal config.toml for the app server
+#[derive(Default)]
+struct CreateConfigTomlParams {
+    forced_method: Option<String>,
+    forced_workspace_id: Option<String>,
+    requires_openai_auth: Option<bool>,
+}
+
+fn create_config_toml(codex_home: &Path, params: CreateConfigTomlParams) -> std::io::Result<()> {
+    let config_toml = codex_home.join("config.toml");
+    let forced_line = if let Some(method) = params.forced_method {
+        format!("forced_login_method = \"{method}\"\n")
+    } else {
+        String::new()
+    };
+    let forced_workspace_line = if let Some(ws) = params.forced_workspace_id {
+        format!("forced_chatgpt_workspace_id = \"{ws}\"\n")
+    } else {
+        String::new()
+    };
+    let requires_line = match params.requires_openai_auth {
+        Some(true) => "requires_openai_auth = true\n".to_string(),
+        Some(false) => String::new(),
+        None => String::new(),
+    };
+    let contents = format!(
+        r#"
+model = "mock-model"
+approval_policy = "never"
+sandbox_mode = "danger-full-access"
+{forced_line}
+{forced_workspace_line}
+
+model_provider = "mock_provider"
+
+[model_providers.mock_provider]
+name = "Mock provider for test"
+base_url = "http://127.0.0.1:0/v1"
+wire_api = "chat"
+request_max_retries = 0
+stream_max_retries = 0
+{requires_line}
+"#
+    );
+    std::fs::write(config_toml, contents)
+}
+
+#[tokio::test]
+async fn logout_account_removes_auth_and_notifies() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), CreateConfigTomlParams::default())?;
+
+    login_with_api_key(
+        codex_home.path(),
+        "sk-test-key",
+        AuthCredentialsStoreMode::File,
+    )?;
+    assert!(codex_home.path().join("auth.json").exists());
+
+    let mut mcp = McpProcess::new_with_env(codex_home.path(), &[("OPENAI_API_KEY", None)]).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let id = mcp.send_logout_account_request().await?;
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(id)),
+    )
+    .await??;
+    let _ok: LogoutAccountResponse = to_response(resp)?;
+
+    let note = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("account/updated"),
+    )
+    .await??;
+    let parsed: ServerNotification = note.try_into()?;
+    let ServerNotification::AccountUpdated(payload) = parsed else {
+        bail!("unexpected notification: {parsed:?}");
+    };
+    assert!(
+        payload.auth_mode.is_none(),
+        "auth_method should be None after logout"
+    );
+
+    assert!(
+        !codex_home.path().join("auth.json").exists(),
+        "auth.json should be deleted"
+    );
+
+    let get_id = mcp
+        .send_get_account_request(GetAccountParams {
+            refresh_token: false,
+        })
+        .await?;
+    let get_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(get_id)),
+    )
+    .await??;
+    let account: GetAccountResponse = to_response(get_resp)?;
+    assert_eq!(account.account, None);
+    Ok(())
+}
+
+#[tokio::test]
+async fn login_account_api_key_succeeds_and_notifies() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), CreateConfigTomlParams::default())?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let req_id = mcp
+        .send_login_account_api_key_request("sk-test-key")
+        .await?;
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(req_id)),
+    )
+    .await??;
+    let login: LoginAccountResponse = to_response(resp)?;
+    assert_eq!(login, LoginAccountResponse::ApiKey {});
+
+    let note = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("account/login/completed"),
+    )
+    .await??;
+    let parsed: ServerNotification = note.try_into()?;
+    let ServerNotification::AccountLoginCompleted(payload) = parsed else {
+        bail!("unexpected notification: {parsed:?}");
+    };
+    pretty_assertions::assert_eq!(payload.login_id, None);
+    pretty_assertions::assert_eq!(payload.success, true);
+    pretty_assertions::assert_eq!(payload.error, None);
+
+    let note = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("account/updated"),
+    )
+    .await??;
+    let parsed: ServerNotification = note.try_into()?;
+    let ServerNotification::AccountUpdated(payload) = parsed else {
+        bail!("unexpected notification: {parsed:?}");
+    };
+    pretty_assertions::assert_eq!(payload.auth_mode, Some(AuthMode::ApiKey));
+
+    assert!(codex_home.path().join("auth.json").exists());
+    Ok(())
+}
+
+#[tokio::test]
+async fn login_account_api_key_rejected_when_forced_chatgpt() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(
+        codex_home.path(),
+        CreateConfigTomlParams {
+            forced_method: Some("chatgpt".to_string()),
+            ..Default::default()
+        },
+    )?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let request_id = mcp
+        .send_login_account_api_key_request("sk-test-key")
+        .await?;
+    let err: JSONRPCError = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_error_message(RequestId::Integer(request_id)),
+    )
+    .await??;
+
+    assert_eq!(
+        err.error.message,
+        "API key login is disabled. Use ChatGPT login instead."
+    );
+    Ok(())
+}
+
+#[tokio::test]
+async fn login_account_chatgpt_rejected_when_forced_api() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(
+        codex_home.path(),
+        CreateConfigTomlParams {
+            forced_method: Some("api".to_string()),
+            ..Default::default()
+        },
+    )?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let request_id = mcp.send_login_account_chatgpt_request().await?;
+    let err: JSONRPCError = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_error_message(RequestId::Integer(request_id)),
+    )
+    .await??;
+
+    assert_eq!(
+        err.error.message,
+        "ChatGPT login is disabled. Use API key login instead."
+    );
+    Ok(())
+}
+
+#[tokio::test]
+// Serialize tests that launch the login server since it binds to a fixed port.
+#[serial(login_port)]
+async fn login_account_chatgpt_start() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), CreateConfigTomlParams::default())?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let request_id = mcp.send_login_account_chatgpt_request().await?;
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
+    )
+    .await??;
+
+    let login: LoginAccountResponse = to_response(resp)?;
+    let LoginAccountResponse::Chatgpt { login_id, auth_url } = login else {
+        bail!("unexpected login response: {login:?}");
+    };
+    assert!(
+        auth_url.contains("redirect_uri=http%3A%2F%2Flocalhost"),
+        "auth_url should contain a redirect_uri to localhost"
+    );
+
+    let cancel_id = mcp
+        .send_cancel_login_account_request(CancelLoginAccountParams {
+            login_id: login_id.clone(),
+        })
+        .await?;
+    let cancel_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(cancel_id)),
+    )
+    .await??;
+    let _ok: CancelLoginAccountResponse = to_response(cancel_resp)?;
+
+    let note = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("account/login/completed"),
+    )
+    .await??;
+    let parsed: ServerNotification = note.try_into()?;
+    let ServerNotification::AccountLoginCompleted(payload) = parsed else {
+        bail!("unexpected notification: {parsed:?}");
+    };
+    pretty_assertions::assert_eq!(payload.login_id, Some(login_id));
+    pretty_assertions::assert_eq!(payload.success, false);
+    assert!(
+        payload.error.is_some(),
+        "expected a non-empty error on cancel"
+    );
+
+    let maybe_updated = timeout(
+        Duration::from_millis(500),
+        mcp.read_stream_until_notification_message("account/updated"),
+    )
+    .await;
+    assert!(
+        maybe_updated.is_err(),
+        "account/updated should not be emitted when login is cancelled"
+    );
+    Ok(())
+}
+
+#[tokio::test]
+// Serialize tests that launch the login server since it binds to a fixed port.
+#[serial(login_port)]
+async fn login_account_chatgpt_includes_forced_workspace_query_param() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(
+        codex_home.path(),
+        CreateConfigTomlParams {
+            forced_workspace_id: Some("ws-forced".to_string()),
+            ..Default::default()
+        },
+    )?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let request_id = mcp.send_login_account_chatgpt_request().await?;
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
+    )
+    .await??;
+
+    let login: LoginAccountResponse = to_response(resp)?;
+    let LoginAccountResponse::Chatgpt { auth_url, .. } = login else {
+        bail!("unexpected login response: {login:?}");
+    };
+    assert!(
+        auth_url.contains("allowed_workspace_id=ws-forced"),
+        "auth URL should include forced workspace"
+    );
+    Ok(())
+}
+
+#[tokio::test]
+async fn get_account_no_auth() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(
+        codex_home.path(),
+        CreateConfigTomlParams {
+            requires_openai_auth: Some(true),
+            ..Default::default()
+        },
+    )?;
+
+    let mut mcp = McpProcess::new_with_env(codex_home.path(), &[("OPENAI_API_KEY", None)]).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let params = GetAccountParams {
+        refresh_token: false,
+    };
+    let request_id = mcp.send_get_account_request(params).await?;
+
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
+    )
+    .await??;
+    let account: GetAccountResponse = to_response(resp)?;
+
+    assert_eq!(account.account, None, "expected no account");
+    assert_eq!(account.requires_openai_auth, true);
+    Ok(())
+}
+
+#[tokio::test]
+async fn get_account_with_api_key() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(
+        codex_home.path(),
+        CreateConfigTomlParams {
+            requires_openai_auth: Some(true),
+            ..Default::default()
+        },
+    )?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let req_id = mcp
+        .send_login_account_api_key_request("sk-test-key")
+        .await?;
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(req_id)),
+    )
+    .await??;
+    let _login_ok = to_response::<LoginAccountResponse>(resp)?;
+
+    let params = GetAccountParams {
+        refresh_token: false,
+    };
+    let request_id = mcp.send_get_account_request(params).await?;
+
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
+    )
+    .await??;
+    let received: GetAccountResponse = to_response(resp)?;
+
+    let expected = GetAccountResponse {
+        account: Some(Account::ApiKey {}),
+        requires_openai_auth: true,
+    };
+    assert_eq!(received, expected);
+    Ok(())
+}
+
+#[tokio::test]
+async fn get_account_when_auth_not_required() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(
+        codex_home.path(),
+        CreateConfigTomlParams {
+            requires_openai_auth: Some(false),
+            ..Default::default()
+        },
+    )?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let params = GetAccountParams {
+        refresh_token: false,
+    };
+    let request_id = mcp.send_get_account_request(params).await?;
+
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
+    )
+    .await??;
+    let received: GetAccountResponse = to_response(resp)?;
+
+    let expected = GetAccountResponse {
+        account: None,
+        requires_openai_auth: false,
+    };
+    assert_eq!(received, expected);
+    Ok(())
+}
+
+#[tokio::test]
+async fn get_account_with_chatgpt() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(
+        codex_home.path(),
+        CreateConfigTomlParams {
+            requires_openai_auth: Some(true),
+            ..Default::default()
+        },
+    )?;
+    write_chatgpt_auth(
+        codex_home.path(),
+        ChatGptAuthFixture::new("access-chatgpt")
+            .email("user@example.com")
+            .plan_type("pro"),
+        AuthCredentialsStoreMode::File,
+    )?;
+
+    let mut mcp = McpProcess::new_with_env(codex_home.path(), &[("OPENAI_API_KEY", None)]).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let params = GetAccountParams {
+        refresh_token: false,
+    };
+    let request_id = mcp.send_get_account_request(params).await?;
+
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
+    )
+    .await??;
+    let received: GetAccountResponse = to_response(resp)?;
+
+    let expected = GetAccountResponse {
+        account: Some(Account::Chatgpt {
+            email: "user@example.com".to_string(),
+            plan_type: AccountPlanType::Pro,
+        }),
+        requires_openai_auth: true,
+    };
+    assert_eq!(received, expected);
+    Ok(())
+}

codex-rs/app-server/tests/suite/v2/mod.rs

@@ -0,0 +1,9 @@
+mod account;
+mod model_list;
+mod rate_limits;
+mod thread_archive;
+mod thread_list;
+mod thread_resume;
+mod thread_start;
+mod turn_interrupt;
+mod turn_start;

codex-rs/app-server/tests/suite/v2/model_list.rs

@@ -6,9 +6,9 @@ use app_test_support::McpProcess;
 use app_test_support::to_response;
 use codex_app_server_protocol::JSONRPCError;
 use codex_app_server_protocol::JSONRPCResponse;
-use codex_app_server_protocol::ListModelsParams;
-use codex_app_server_protocol::ListModelsResponse;
 use codex_app_server_protocol::Model;
+use codex_app_server_protocol::ModelListParams;
+use codex_app_server_protocol::ModelListResponse;
 use codex_app_server_protocol::ReasoningEffortOption;
 use codex_app_server_protocol::RequestId;
 use codex_protocol::config_types::ReasoningEffort;
@@ -19,7 +19,7 @@ use tokio::time::timeout;
 const DEFAULT_TIMEOUT: Duration = Duration::from_secs(10);
 const INVALID_REQUEST_ERROR_CODE: i64 = -32600;
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[tokio::test]
 async fn list_models_returns_all_models_with_large_limit() -> Result<()> {
     let codex_home = TempDir::new()?;
     let mut mcp = McpProcess::new(codex_home.path()).await?;
@@ -27,8 +27,8 @@ async fn list_models_returns_all_models_with_large_limit() -> Result<()> {
     timeout(DEFAULT_TIMEOUT, mcp.initialize()).await??;
 
     let request_id = mcp
-        .send_list_models_request(ListModelsParams {
-            page_size: Some(100),
+        .send_list_models_request(ModelListParams {
+            limit: Some(100),
             cursor: None,
         })
         .await?;
@@ -39,14 +39,17 @@ async fn list_models_returns_all_models_with_large_limit() -> Result<()> {
     )
     .await??;
 
-    let ListModelsResponse { items, next_cursor } = to_response::<ListModelsResponse>(response)?;
+    let ModelListResponse {
+        data: items,
+        next_cursor,
+    } = to_response::<ModelListResponse>(response)?;
 
     let expected_models = vec![
         Model {
-            id: "gpt-5-codex".to_string(),
-            model: "gpt-5-codex".to_string(),
-            display_name: "gpt-5-codex".to_string(),
-            description: "Optimized for coding tasks with many tools.".to_string(),
+            id: "gpt-5.1-codex".to_string(),
+            model: "gpt-5.1-codex".to_string(),
+            display_name: "gpt-5.1-codex".to_string(),
+            description: "Optimized for codex.".to_string(),
             supported_reasoning_efforts: vec![
                 ReasoningEffortOption {
                     reasoning_effort: ReasoningEffort::Low,
@@ -66,15 +69,30 @@ async fn list_models_returns_all_models_with_large_limit() -> Result<()> {
             is_default: true,
         },
         Model {
-            id: "gpt-5".to_string(),
-            model: "gpt-5".to_string(),
-            display_name: "gpt-5".to_string(),
-            description: "Broad world knowledge with strong general reasoning.".to_string(),
+            id: "gpt-5.1-codex-mini".to_string(),
+            model: "gpt-5.1-codex-mini".to_string(),
+            display_name: "gpt-5.1-codex-mini".to_string(),
+            description: "Optimized for codex. Cheaper, faster, but less capable.".to_string(),
             supported_reasoning_efforts: vec![
                 ReasoningEffortOption {
-                    reasoning_effort: ReasoningEffort::Minimal,
-                    description: "Fastest responses with little reasoning".to_string(),
+                    reasoning_effort: ReasoningEffort::Medium,
+                    description: "Dynamically adjusts reasoning based on the task".to_string(),
                 },
+                ReasoningEffortOption {
+                    reasoning_effort: ReasoningEffort::High,
+                    description: "Maximizes reasoning depth for complex or ambiguous problems"
+                        .to_string(),
+                },
+            ],
+            default_reasoning_effort: ReasoningEffort::Medium,
+            is_default: false,
+        },
+        Model {
+            id: "gpt-5.1".to_string(),
+            model: "gpt-5.1".to_string(),
+            display_name: "gpt-5.1".to_string(),
+            description: "Broad world knowledge with strong general reasoning.".to_string(),
+            supported_reasoning_efforts: vec![
                 ReasoningEffortOption {
                     reasoning_effort: ReasoningEffort::Low,
                     description: "Balances speed with some reasoning; useful for straightforward \
@@ -103,7 +121,7 @@ async fn list_models_returns_all_models_with_large_limit() -> Result<()> {
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[tokio::test]
 async fn list_models_pagination_works() -> Result<()> {
     let codex_home = TempDir::new()?;
     let mut mcp = McpProcess::new(codex_home.path()).await?;
@@ -111,8 +129,8 @@ async fn list_models_pagination_works() -> Result<()> {
     timeout(DEFAULT_TIMEOUT, mcp.initialize()).await??;
 
     let first_request = mcp
-        .send_list_models_request(ListModelsParams {
-            page_size: Some(1),
+        .send_list_models_request(ModelListParams {
+            limit: Some(1),
             cursor: None,
         })
         .await?;
@@ -123,18 +141,18 @@ async fn list_models_pagination_works() -> Result<()> {
     )
     .await??;
 
-    let ListModelsResponse {
-        items: first_items,
+    let ModelListResponse {
+        data: first_items,
         next_cursor: first_cursor,
-    } = to_response::<ListModelsResponse>(first_response)?;
+    } = to_response::<ModelListResponse>(first_response)?;
 
     assert_eq!(first_items.len(), 1);
-    assert_eq!(first_items[0].id, "gpt-5-codex");
+    assert_eq!(first_items[0].id, "gpt-5.1-codex");
     let next_cursor = first_cursor.ok_or_else(|| anyhow!("cursor for second page"))?;
 
     let second_request = mcp
-        .send_list_models_request(ListModelsParams {
-            page_size: Some(1),
+        .send_list_models_request(ModelListParams {
+            limit: Some(1),
             cursor: Some(next_cursor.clone()),
         })
         .await?;
@@ -145,18 +163,40 @@ async fn list_models_pagination_works() -> Result<()> {
     )
     .await??;
 
-    let ListModelsResponse {
-        items: second_items,
+    let ModelListResponse {
+        data: second_items,
         next_cursor: second_cursor,
-    } = to_response::<ListModelsResponse>(second_response)?;
+    } = to_response::<ModelListResponse>(second_response)?;
 
     assert_eq!(second_items.len(), 1);
-    assert_eq!(second_items[0].id, "gpt-5");
-    assert!(second_cursor.is_none());
+    assert_eq!(second_items[0].id, "gpt-5.1-codex-mini");
+    let third_cursor = second_cursor.ok_or_else(|| anyhow!("cursor for third page"))?;
+
+    let third_request = mcp
+        .send_list_models_request(ModelListParams {
+            limit: Some(1),
+            cursor: Some(third_cursor.clone()),
+        })
+        .await?;
+
+    let third_response: JSONRPCResponse = timeout(
+        DEFAULT_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(third_request)),
+    )
+    .await??;
+
+    let ModelListResponse {
+        data: third_items,
+        next_cursor: third_cursor,
+    } = to_response::<ModelListResponse>(third_response)?;
+
+    assert_eq!(third_items.len(), 1);
+    assert_eq!(third_items[0].id, "gpt-5.1");
+    assert!(third_cursor.is_none());
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[tokio::test]
 async fn list_models_rejects_invalid_cursor() -> Result<()> {
     let codex_home = TempDir::new()?;
     let mut mcp = McpProcess::new(codex_home.path()).await?;
@@ -164,8 +204,8 @@ async fn list_models_rejects_invalid_cursor() -> Result<()> {
     timeout(DEFAULT_TIMEOUT, mcp.initialize()).await??;
 
     let request_id = mcp
-        .send_list_models_request(ListModelsParams {
-            page_size: None,
+        .send_list_models_request(ModelListParams {
+            limit: None,
             cursor: Some("invalid".to_string()),
         })
         .await?;

codex-rs/app-server/tests/suite/v2/rate_limits.rs

@@ -7,10 +7,10 @@ use codex_app_server_protocol::GetAccountRateLimitsResponse;
 use codex_app_server_protocol::JSONRPCError;
 use codex_app_server_protocol::JSONRPCResponse;
 use codex_app_server_protocol::LoginApiKeyParams;
+use codex_app_server_protocol::RateLimitSnapshot;
+use codex_app_server_protocol::RateLimitWindow;
 use codex_app_server_protocol::RequestId;
 use codex_core::auth::AuthCredentialsStoreMode;
-use codex_protocol::protocol::RateLimitSnapshot;
-use codex_protocol::protocol::RateLimitWindow;
 use pretty_assertions::assert_eq;
 use serde_json::json;
 use std::path::Path;
@@ -26,7 +26,7 @@ use wiremock::matchers::path;
 const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
 const INVALID_REQUEST_ERROR_CODE: i64 = -32600;
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[tokio::test]
 async fn get_account_rate_limits_requires_auth() -> Result<()> {
     let codex_home = TempDir::new()?;
 
@@ -51,7 +51,7 @@ async fn get_account_rate_limits_requires_auth() -> Result<()> {
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[tokio::test]
 async fn get_account_rate_limits_requires_chatgpt_auth() -> Result<()> {
     let codex_home = TempDir::new()?;
 
@@ -78,7 +78,7 @@ async fn get_account_rate_limits_requires_chatgpt_auth() -> Result<()> {
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+#[tokio::test]
 async fn get_account_rate_limits_returns_snapshot() -> Result<()> {
     let codex_home = TempDir::new()?;
     write_chatgpt_auth(
@@ -143,13 +143,13 @@ async fn get_account_rate_limits_returns_snapshot() -> Result<()> {
     let expected = GetAccountRateLimitsResponse {
         rate_limits: RateLimitSnapshot {
             primary: Some(RateLimitWindow {
-                used_percent: 42.0,
-                window_minutes: Some(60),
+                used_percent: 42,
+                window_duration_mins: Some(60),
                 resets_at: Some(primary_reset_timestamp),
             }),
             secondary: Some(RateLimitWindow {
-                used_percent: 5.0,
-                window_minutes: Some(1440),
+                used_percent: 5,
+                window_duration_mins: Some(1440),
                 resets_at: Some(secondary_reset_timestamp),
             }),
         },

codex-rs/app-server/tests/suite/v2/thread_archive.rs

@@ -0,0 +1,93 @@
+use anyhow::Result;
+use app_test_support::McpProcess;
+use app_test_support::to_response;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::ThreadArchiveParams;
+use codex_app_server_protocol::ThreadArchiveResponse;
+use codex_app_server_protocol::ThreadStartParams;
+use codex_app_server_protocol::ThreadStartResponse;
+use codex_core::ARCHIVED_SESSIONS_SUBDIR;
+use codex_core::find_conversation_path_by_id_str;
+use std::path::Path;
+use tempfile::TempDir;
+use tokio::time::timeout;
+
+const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
+
+#[tokio::test]
+async fn thread_archive_moves_rollout_into_archived_directory() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path())?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    // Start a thread.
+    let start_id = mcp
+        .send_thread_start_request(ThreadStartParams {
+            model: Some("mock-model".to_string()),
+            ..Default::default()
+        })
+        .await?;
+    let start_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
+    )
+    .await??;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(start_resp)?;
+    assert!(!thread.id.is_empty());
+
+    // Locate the rollout path recorded for this thread id.
+    let rollout_path = find_conversation_path_by_id_str(codex_home.path(), &thread.id)
+        .await?
+        .expect("expected rollout path for thread id to exist");
+    assert!(
+        rollout_path.exists(),
+        "expected {} to exist",
+        rollout_path.display()
+    );
+
+    // Archive the thread.
+    let archive_id = mcp
+        .send_thread_archive_request(ThreadArchiveParams {
+            thread_id: thread.id.clone(),
+        })
+        .await?;
+    let archive_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(archive_id)),
+    )
+    .await??;
+    let _: ThreadArchiveResponse = to_response::<ThreadArchiveResponse>(archive_resp)?;
+
+    // Verify file moved.
+    let archived_directory = codex_home.path().join(ARCHIVED_SESSIONS_SUBDIR);
+    // The archived file keeps the original filename (rollout-...-<id>.jsonl).
+    let archived_rollout_path =
+        archived_directory.join(rollout_path.file_name().expect("rollout file name"));
+    assert!(
+        !rollout_path.exists(),
+        "expected rollout path {} to be moved",
+        rollout_path.display()
+    );
+    assert!(
+        archived_rollout_path.exists(),
+        "expected archived rollout path {} to exist",
+        archived_rollout_path.display()
+    );
+
+    Ok(())
+}
+
+fn create_config_toml(codex_home: &Path) -> std::io::Result<()> {
+    let config_toml = codex_home.join("config.toml");
+    std::fs::write(config_toml, config_contents())
+}
+
+fn config_contents() -> &'static str {
+    r#"model = "mock-model"
+approval_policy = "never"
+sandbox_mode = "read-only"
+"#
+}

codex-rs/app-server/tests/suite/v2/thread_list.rs

@@ -0,0 +1,185 @@
+use anyhow::Result;
+use app_test_support::McpProcess;
+use app_test_support::create_fake_rollout;
+use app_test_support::to_response;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::ThreadListParams;
+use codex_app_server_protocol::ThreadListResponse;
+use tempfile::TempDir;
+use tokio::time::timeout;
+
+const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
+
+#[tokio::test]
+async fn thread_list_basic_empty() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_minimal_config(codex_home.path())?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    // List threads in an empty CODEX_HOME; should return an empty page with nextCursor: null.
+    let list_id = mcp
+        .send_thread_list_request(ThreadListParams {
+            cursor: None,
+            limit: Some(10),
+            model_providers: None,
+        })
+        .await?;
+    let list_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(list_id)),
+    )
+    .await??;
+    let ThreadListResponse { data, next_cursor } = to_response::<ThreadListResponse>(list_resp)?;
+    assert!(data.is_empty());
+    assert_eq!(next_cursor, None);
+
+    Ok(())
+}
+
+// Minimal config.toml for listing.
+fn create_minimal_config(codex_home: &std::path::Path) -> std::io::Result<()> {
+    let config_toml = codex_home.join("config.toml");
+    std::fs::write(
+        config_toml,
+        r#"
+model = "mock-model"
+approval_policy = "never"
+"#,
+    )
+}
+
+#[tokio::test]
+async fn thread_list_pagination_next_cursor_none_on_last_page() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_minimal_config(codex_home.path())?;
+
+    // Create three rollouts so we can paginate with limit=2.
+    let _a = create_fake_rollout(
+        codex_home.path(),
+        "2025-01-02T12-00-00",
+        "2025-01-02T12:00:00Z",
+        "Hello",
+        Some("mock_provider"),
+    )?;
+    let _b = create_fake_rollout(
+        codex_home.path(),
+        "2025-01-01T13-00-00",
+        "2025-01-01T13:00:00Z",
+        "Hello",
+        Some("mock_provider"),
+    )?;
+    let _c = create_fake_rollout(
+        codex_home.path(),
+        "2025-01-01T12-00-00",
+        "2025-01-01T12:00:00Z",
+        "Hello",
+        Some("mock_provider"),
+    )?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    // Page 1: limit 2 → expect next_cursor Some.
+    let page1_id = mcp
+        .send_thread_list_request(ThreadListParams {
+            cursor: None,
+            limit: Some(2),
+            model_providers: Some(vec!["mock_provider".to_string()]),
+        })
+        .await?;
+    let page1_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(page1_id)),
+    )
+    .await??;
+    let ThreadListResponse {
+        data: data1,
+        next_cursor: cursor1,
+    } = to_response::<ThreadListResponse>(page1_resp)?;
+    assert_eq!(data1.len(), 2);
+    for thread in &data1 {
+        assert_eq!(thread.preview, "Hello");
+        assert_eq!(thread.model_provider, "mock_provider");
+        assert!(thread.created_at > 0);
+    }
+    let cursor1 = cursor1.expect("expected nextCursor on first page");
+
+    // Page 2: with cursor → expect next_cursor None when no more results.
+    let page2_id = mcp
+        .send_thread_list_request(ThreadListParams {
+            cursor: Some(cursor1),
+            limit: Some(2),
+            model_providers: Some(vec!["mock_provider".to_string()]),
+        })
+        .await?;
+    let page2_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(page2_id)),
+    )
+    .await??;
+    let ThreadListResponse {
+        data: data2,
+        next_cursor: cursor2,
+    } = to_response::<ThreadListResponse>(page2_resp)?;
+    assert!(data2.len() <= 2);
+    for thread in &data2 {
+        assert_eq!(thread.preview, "Hello");
+        assert_eq!(thread.model_provider, "mock_provider");
+        assert!(thread.created_at > 0);
+    }
+    assert_eq!(cursor2, None, "expected nextCursor to be null on last page");
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn thread_list_respects_provider_filter() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_minimal_config(codex_home.path())?;
+
+    // Create rollouts under two providers.
+    let _a = create_fake_rollout(
+        codex_home.path(),
+        "2025-01-02T10-00-00",
+        "2025-01-02T10:00:00Z",
+        "X",
+        Some("mock_provider"),
+    )?; // mock_provider
+    let _b = create_fake_rollout(
+        codex_home.path(),
+        "2025-01-02T11-00-00",
+        "2025-01-02T11:00:00Z",
+        "X",
+        Some("other_provider"),
+    )?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    // Filter to only other_provider; expect 1 item, nextCursor None.
+    let list_id = mcp
+        .send_thread_list_request(ThreadListParams {
+            cursor: None,
+            limit: Some(10),
+            model_providers: Some(vec!["other_provider".to_string()]),
+        })
+        .await?;
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(list_id)),
+    )
+    .await??;
+    let ThreadListResponse { data, next_cursor } = to_response::<ThreadListResponse>(resp)?;
+    assert_eq!(data.len(), 1);
+    assert_eq!(next_cursor, None);
+    let thread = &data[0];
+    assert_eq!(thread.preview, "X");
+    assert_eq!(thread.model_provider, "other_provider");
+    let expected_ts = chrono::DateTime::parse_from_rfc3339("2025-01-02T11:00:00Z")?.timestamp();
+    assert_eq!(thread.created_at, expected_ts);
+
+    Ok(())
+}

codex-rs/app-server/tests/suite/v2/thread_resume.rs

@@ -0,0 +1,181 @@
+use anyhow::Result;
+use app_test_support::McpProcess;
+use app_test_support::create_mock_chat_completions_server;
+use app_test_support::to_response;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::ThreadResumeParams;
+use codex_app_server_protocol::ThreadResumeResponse;
+use codex_app_server_protocol::ThreadStartParams;
+use codex_app_server_protocol::ThreadStartResponse;
+use codex_protocol::models::ContentItem;
+use codex_protocol::models::ResponseItem;
+use tempfile::TempDir;
+use tokio::time::timeout;
+
+const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
+
+#[tokio::test]
+async fn thread_resume_returns_original_thread() -> Result<()> {
+    let server = create_mock_chat_completions_server(vec![]).await;
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), &server.uri())?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    // Start a thread.
+    let start_id = mcp
+        .send_thread_start_request(ThreadStartParams {
+            model: Some("gpt-5-codex".to_string()),
+            ..Default::default()
+        })
+        .await?;
+    let start_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
+    )
+    .await??;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(start_resp)?;
+
+    // Resume it via v2 API.
+    let resume_id = mcp
+        .send_thread_resume_request(ThreadResumeParams {
+            thread_id: thread.id.clone(),
+            ..Default::default()
+        })
+        .await?;
+    let resume_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(resume_id)),
+    )
+    .await??;
+    let ThreadResumeResponse { thread: resumed } =
+        to_response::<ThreadResumeResponse>(resume_resp)?;
+    assert_eq!(resumed, thread);
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn thread_resume_prefers_path_over_thread_id() -> Result<()> {
+    let server = create_mock_chat_completions_server(vec![]).await;
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), &server.uri())?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let start_id = mcp
+        .send_thread_start_request(ThreadStartParams {
+            model: Some("gpt-5-codex".to_string()),
+            ..Default::default()
+        })
+        .await?;
+    let start_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
+    )
+    .await??;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(start_resp)?;
+
+    let thread_path = thread.path.clone();
+    let resume_id = mcp
+        .send_thread_resume_request(ThreadResumeParams {
+            thread_id: "not-a-valid-thread-id".to_string(),
+            path: Some(thread_path),
+            ..Default::default()
+        })
+        .await?;
+
+    let resume_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(resume_id)),
+    )
+    .await??;
+    let ThreadResumeResponse { thread: resumed } =
+        to_response::<ThreadResumeResponse>(resume_resp)?;
+    assert_eq!(resumed, thread);
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn thread_resume_supports_history_and_overrides() -> Result<()> {
+    let server = create_mock_chat_completions_server(vec![]).await;
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), &server.uri())?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    // Start a thread.
+    let start_id = mcp
+        .send_thread_start_request(ThreadStartParams {
+            model: Some("gpt-5-codex".to_string()),
+            ..Default::default()
+        })
+        .await?;
+    let start_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
+    )
+    .await??;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(start_resp)?;
+
+    let history_text = "Hello from history";
+    let history = vec![ResponseItem::Message {
+        id: None,
+        role: "user".to_string(),
+        content: vec![ContentItem::InputText {
+            text: history_text.to_string(),
+        }],
+    }];
+
+    // Resume with explicit history and override the model.
+    let resume_id = mcp
+        .send_thread_resume_request(ThreadResumeParams {
+            thread_id: thread.id,
+            history: Some(history),
+            model: Some("mock-model".to_string()),
+            model_provider: Some("mock_provider".to_string()),
+            ..Default::default()
+        })
+        .await?;
+    let resume_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(resume_id)),
+    )
+    .await??;
+    let ThreadResumeResponse { thread: resumed } =
+        to_response::<ThreadResumeResponse>(resume_resp)?;
+    assert!(!resumed.id.is_empty());
+    assert_eq!(resumed.model_provider, "mock_provider");
+    assert_eq!(resumed.preview, history_text);
+
+    Ok(())
+}
+
+// Helper to create a config.toml pointing at the mock model server.
+fn create_config_toml(codex_home: &std::path::Path, server_uri: &str) -> std::io::Result<()> {
+    let config_toml = codex_home.join("config.toml");
+    std::fs::write(
+        config_toml,
+        format!(
+            r#"
+model = "mock-model"
+approval_policy = "never"
+sandbox_mode = "read-only"
+
+model_provider = "mock_provider"
+
+[model_providers.mock_provider]
+name = "Mock provider for test"
+base_url = "{server_uri}/v1"
+wire_api = "chat"
+request_max_retries = 0
+stream_max_retries = 0
+"#
+        ),
+    )
+}

codex-rs/app-server/tests/suite/v2/thread_start.rs

@@ -0,0 +1,90 @@
+use anyhow::Result;
+use app_test_support::McpProcess;
+use app_test_support::create_mock_chat_completions_server;
+use app_test_support::to_response;
+use codex_app_server_protocol::JSONRPCNotification;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::ThreadStartParams;
+use codex_app_server_protocol::ThreadStartResponse;
+use codex_app_server_protocol::ThreadStartedNotification;
+use std::path::Path;
+use tempfile::TempDir;
+use tokio::time::timeout;
+
+const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
+
+#[tokio::test]
+async fn thread_start_creates_thread_and_emits_started() -> Result<()> {
+    // Provide a mock server and config so model wiring is valid.
+    let server = create_mock_chat_completions_server(vec![]).await;
+
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), &server.uri())?;
+
+    // Start server and initialize.
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    // Start a v2 thread with an explicit model override.
+    let req_id = mcp
+        .send_thread_start_request(ThreadStartParams {
+            model: Some("gpt-5".to_string()),
+            ..Default::default()
+        })
+        .await?;
+
+    // Expect a proper JSON-RPC response with a thread id.
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(req_id)),
+    )
+    .await??;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(resp)?;
+    assert!(!thread.id.is_empty(), "thread id should not be empty");
+    assert!(
+        thread.preview.is_empty(),
+        "new threads should start with an empty preview"
+    );
+    assert_eq!(thread.model_provider, "mock_provider");
+    assert!(
+        thread.created_at > 0,
+        "created_at should be a positive UNIX timestamp"
+    );
+
+    // A corresponding thread/started notification should arrive.
+    let notif: JSONRPCNotification = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("thread/started"),
+    )
+    .await??;
+    let started: ThreadStartedNotification =
+        serde_json::from_value(notif.params.expect("params must be present"))?;
+    assert_eq!(started.thread, thread);
+
+    Ok(())
+}
+
+// Helper to create a config.toml pointing at the mock model server.
+fn create_config_toml(codex_home: &Path, server_uri: &str) -> std::io::Result<()> {
+    let config_toml = codex_home.join("config.toml");
+    std::fs::write(
+        config_toml,
+        format!(
+            r#"
+model = "mock-model"
+approval_policy = "never"
+sandbox_mode = "read-only"
+
+model_provider = "mock_provider"
+
+[model_providers.mock_provider]
+name = "Mock provider for test"
+base_url = "{server_uri}/v1"
+wire_api = "chat"
+request_max_retries = 0
+stream_max_retries = 0
+"#
+        ),
+    )
+}

codex-rs/app-server/tests/suite/v2/turn_interrupt.rs

@@ -0,0 +1,128 @@
+#![cfg(unix)]
+
+use anyhow::Result;
+use app_test_support::McpProcess;
+use app_test_support::create_mock_chat_completions_server;
+use app_test_support::create_shell_sse_response;
+use app_test_support::to_response;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::ThreadStartParams;
+use codex_app_server_protocol::ThreadStartResponse;
+use codex_app_server_protocol::TurnInterruptParams;
+use codex_app_server_protocol::TurnInterruptResponse;
+use codex_app_server_protocol::TurnStartParams;
+use codex_app_server_protocol::TurnStartResponse;
+use codex_app_server_protocol::UserInput as V2UserInput;
+use tempfile::TempDir;
+use tokio::time::timeout;
+
+const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
+
+#[tokio::test]
+async fn turn_interrupt_aborts_running_turn() -> Result<()> {
+    // Use a portable sleep command to keep the turn running.
+    #[cfg(target_os = "windows")]
+    let shell_command = vec![
+        "powershell".to_string(),
+        "-Command".to_string(),
+        "Start-Sleep -Seconds 10".to_string(),
+    ];
+    #[cfg(not(target_os = "windows"))]
+    let shell_command = vec!["sleep".to_string(), "10".to_string()];
+
+    let tmp = TempDir::new()?;
+    let codex_home = tmp.path().join("codex_home");
+    std::fs::create_dir(&codex_home)?;
+    let working_directory = tmp.path().join("workdir");
+    std::fs::create_dir(&working_directory)?;
+
+    // Mock server: long-running shell command then (after abort) nothing else needed.
+    let server = create_mock_chat_completions_server(vec![create_shell_sse_response(
+        shell_command.clone(),
+        Some(&working_directory),
+        Some(10_000),
+        "call_sleep",
+    )?])
+    .await;
+    create_config_toml(&codex_home, &server.uri())?;
+
+    let mut mcp = McpProcess::new(&codex_home).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    // Start a v2 thread and capture its id.
+    let thread_req = mcp
+        .send_thread_start_request(ThreadStartParams {
+            model: Some("mock-model".to_string()),
+            ..Default::default()
+        })
+        .await?;
+    let thread_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(thread_req)),
+    )
+    .await??;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(thread_resp)?;
+
+    // Start a turn that triggers a long-running command.
+    let turn_req = mcp
+        .send_turn_start_request(TurnStartParams {
+            thread_id: thread.id.clone(),
+            input: vec![V2UserInput::Text {
+                text: "run sleep".to_string(),
+            }],
+            cwd: Some(working_directory.clone()),
+            ..Default::default()
+        })
+        .await?;
+    let turn_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(turn_req)),
+    )
+    .await??;
+    let TurnStartResponse { turn } = to_response::<TurnStartResponse>(turn_resp)?;
+
+    // Give the command a brief moment to start.
+    tokio::time::sleep(std::time::Duration::from_secs(1)).await;
+
+    // Interrupt the in-progress turn by id (v2 API).
+    let interrupt_id = mcp
+        .send_turn_interrupt_request(TurnInterruptParams {
+            thread_id: thread.id,
+            turn_id: turn.id,
+        })
+        .await?;
+    let interrupt_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(interrupt_id)),
+    )
+    .await??;
+    let _resp: TurnInterruptResponse = to_response::<TurnInterruptResponse>(interrupt_resp)?;
+
+    // No fields to assert on; successful deserialization confirms proper response shape.
+    Ok(())
+}
+
+// Helper to create a config.toml pointing at the mock model server.
+fn create_config_toml(codex_home: &std::path::Path, server_uri: &str) -> std::io::Result<()> {
+    let config_toml = codex_home.join("config.toml");
+    std::fs::write(
+        config_toml,
+        format!(
+            r#"
+model = "mock-model"
+approval_policy = "never"
+sandbox_mode = "workspace-write"
+
+model_provider = "mock_provider"
+
+[model_providers.mock_provider]
+name = "Mock provider for test"
+base_url = "{server_uri}/v1"
+wire_api = "chat"
+request_max_retries = 0
+stream_max_retries = 0
+"#
+        ),
+    )
+}

codex-rs/app-server/tests/suite/v2/turn_start.rs

@@ -0,0 +1,486 @@
+use anyhow::Result;
+use app_test_support::McpProcess;
+use app_test_support::create_final_assistant_message_sse_response;
+use app_test_support::create_mock_chat_completions_server;
+use app_test_support::create_mock_chat_completions_server_unchecked;
+use app_test_support::create_shell_sse_response;
+use app_test_support::to_response;
+use codex_app_server_protocol::JSONRPCNotification;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::ServerRequest;
+use codex_app_server_protocol::ThreadStartParams;
+use codex_app_server_protocol::ThreadStartResponse;
+use codex_app_server_protocol::TurnStartParams;
+use codex_app_server_protocol::TurnStartResponse;
+use codex_app_server_protocol::TurnStartedNotification;
+use codex_app_server_protocol::UserInput as V2UserInput;
+use codex_core::protocol_config_types::ReasoningEffort;
+use codex_core::protocol_config_types::ReasoningSummary;
+use codex_protocol::parse_command::ParsedCommand;
+use codex_protocol::protocol::Event;
+use codex_protocol::protocol::EventMsg;
+use core_test_support::skip_if_no_network;
+use pretty_assertions::assert_eq;
+use std::path::Path;
+use tempfile::TempDir;
+use tokio::time::timeout;
+
+const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
+
+#[tokio::test]
+async fn turn_start_emits_notifications_and_accepts_model_override() -> Result<()> {
+    // Provide a mock server and config so model wiring is valid.
+    // Three Codex turns hit the mock model (session start + two turn/start calls).
+    let responses = vec![
+        create_final_assistant_message_sse_response("Done")?,
+        create_final_assistant_message_sse_response("Done")?,
+        create_final_assistant_message_sse_response("Done")?,
+    ];
+    let server = create_mock_chat_completions_server_unchecked(responses).await;
+
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), &server.uri(), "never")?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    // Start a thread (v2) and capture its id.
+    let thread_req = mcp
+        .send_thread_start_request(ThreadStartParams {
+            model: Some("mock-model".to_string()),
+            ..Default::default()
+        })
+        .await?;
+    let thread_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(thread_req)),
+    )
+    .await??;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(thread_resp)?;
+
+    // Start a turn with only input and thread_id set (no overrides).
+    let turn_req = mcp
+        .send_turn_start_request(TurnStartParams {
+            thread_id: thread.id.clone(),
+            input: vec![V2UserInput::Text {
+                text: "Hello".to_string(),
+            }],
+            ..Default::default()
+        })
+        .await?;
+    let turn_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(turn_req)),
+    )
+    .await??;
+    let TurnStartResponse { turn } = to_response::<TurnStartResponse>(turn_resp)?;
+    assert!(!turn.id.is_empty());
+
+    // Expect a turn/started notification.
+    let notif: JSONRPCNotification = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("turn/started"),
+    )
+    .await??;
+    let started: TurnStartedNotification =
+        serde_json::from_value(notif.params.expect("params must be present"))?;
+    assert_eq!(
+        started.turn.status,
+        codex_app_server_protocol::TurnStatus::InProgress
+    );
+
+    // Send a second turn that exercises the overrides path: change the model.
+    let turn_req2 = mcp
+        .send_turn_start_request(TurnStartParams {
+            thread_id: thread.id.clone(),
+            input: vec![V2UserInput::Text {
+                text: "Second".to_string(),
+            }],
+            model: Some("mock-model-override".to_string()),
+            ..Default::default()
+        })
+        .await?;
+    let turn_resp2: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(turn_req2)),
+    )
+    .await??;
+    let TurnStartResponse { turn: turn2 } = to_response::<TurnStartResponse>(turn_resp2)?;
+    assert!(!turn2.id.is_empty());
+    // Ensure the second turn has a different id than the first.
+    assert_ne!(turn.id, turn2.id);
+
+    // Expect a second turn/started notification as well.
+    let _notif2: JSONRPCNotification = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("turn/started"),
+    )
+    .await??;
+
+    // And we should ultimately get a task_complete without having to add a
+    // legacy conversation listener explicitly (auto-attached by thread/start).
+    let _task_complete: JSONRPCNotification = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("codex/event/task_complete"),
+    )
+    .await??;
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn turn_start_accepts_local_image_input() -> Result<()> {
+    // Two Codex turns hit the mock model (session start + turn/start).
+    let responses = vec![
+        create_final_assistant_message_sse_response("Done")?,
+        create_final_assistant_message_sse_response("Done")?,
+    ];
+    // Use the unchecked variant because the request payload includes a LocalImage
+    // which the strict matcher does not currently cover.
+    let server = create_mock_chat_completions_server_unchecked(responses).await;
+
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), &server.uri(), "never")?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let thread_req = mcp
+        .send_thread_start_request(ThreadStartParams {
+            model: Some("mock-model".to_string()),
+            ..Default::default()
+        })
+        .await?;
+    let thread_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(thread_req)),
+    )
+    .await??;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(thread_resp)?;
+
+    let image_path = codex_home.path().join("image.png");
+    // No need to actually write the file; we just exercise the input path.
+
+    let turn_req = mcp
+        .send_turn_start_request(TurnStartParams {
+            thread_id: thread.id.clone(),
+            input: vec![V2UserInput::LocalImage { path: image_path }],
+            ..Default::default()
+        })
+        .await?;
+    let turn_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(turn_req)),
+    )
+    .await??;
+    let TurnStartResponse { turn } = to_response::<TurnStartResponse>(turn_resp)?;
+    assert!(!turn.id.is_empty());
+
+    // This test only validates that turn/start responds and returns a turn.
+    Ok(())
+}
+
+#[tokio::test]
+async fn turn_start_exec_approval_toggle_v2() -> Result<()> {
+    skip_if_no_network!(Ok(()));
+
+    let tmp = TempDir::new()?;
+    let codex_home = tmp.path().to_path_buf();
+
+    // Mock server: first turn requests a shell call (elicitation), then completes.
+    // Second turn same, but we'll set approval_policy=never to avoid elicitation.
+    let responses = vec![
+        create_shell_sse_response(
+            vec![
+                "python3".to_string(),
+                "-c".to_string(),
+                "print(42)".to_string(),
+            ],
+            None,
+            Some(5000),
+            "call1",
+        )?,
+        create_final_assistant_message_sse_response("done 1")?,
+        create_shell_sse_response(
+            vec![
+                "python3".to_string(),
+                "-c".to_string(),
+                "print(42)".to_string(),
+            ],
+            None,
+            Some(5000),
+            "call2",
+        )?,
+        create_final_assistant_message_sse_response("done 2")?,
+    ];
+    let server = create_mock_chat_completions_server(responses).await;
+    // Default approval is untrusted to force elicitation on first turn.
+    create_config_toml(codex_home.as_path(), &server.uri(), "untrusted")?;
+
+    let mut mcp = McpProcess::new(codex_home.as_path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    // thread/start
+    let start_id = mcp
+        .send_thread_start_request(ThreadStartParams {
+            model: Some("mock-model".to_string()),
+            ..Default::default()
+        })
+        .await?;
+    let start_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
+    )
+    .await??;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(start_resp)?;
+
+    // turn/start — expect ExecCommandApproval request from server
+    let first_turn_id = mcp
+        .send_turn_start_request(TurnStartParams {
+            thread_id: thread.id.clone(),
+            input: vec![V2UserInput::Text {
+                text: "run python".to_string(),
+            }],
+            ..Default::default()
+        })
+        .await?;
+    // Acknowledge RPC
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(first_turn_id)),
+    )
+    .await??;
+
+    // Receive elicitation
+    let server_req = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_request_message(),
+    )
+    .await??;
+    let ServerRequest::ExecCommandApproval { request_id, params } = server_req else {
+        panic!("expected ExecCommandApproval request");
+    };
+    assert_eq!(params.call_id, "call1");
+    assert_eq!(
+        params.parsed_cmd,
+        vec![ParsedCommand::Unknown {
+            cmd: "python3 -c 'print(42)'".to_string()
+        }]
+    );
+
+    // Approve and wait for task completion
+    mcp.send_response(
+        request_id,
+        serde_json::json!({ "decision": codex_core::protocol::ReviewDecision::Approved }),
+    )
+    .await?;
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("codex/event/task_complete"),
+    )
+    .await??;
+
+    // Second turn with approval_policy=never should not elicit approval
+    let second_turn_id = mcp
+        .send_turn_start_request(TurnStartParams {
+            thread_id: thread.id.clone(),
+            input: vec![V2UserInput::Text {
+                text: "run python again".to_string(),
+            }],
+            approval_policy: Some(codex_app_server_protocol::AskForApproval::Never),
+            sandbox_policy: Some(codex_app_server_protocol::SandboxPolicy::DangerFullAccess),
+            model: Some("mock-model".to_string()),
+            effort: Some(ReasoningEffort::Medium),
+            summary: Some(ReasoningSummary::Auto),
+            ..Default::default()
+        })
+        .await?;
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(second_turn_id)),
+    )
+    .await??;
+
+    // Ensure we do NOT receive an ExecCommandApproval request before task completes
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("codex/event/task_complete"),
+    )
+    .await??;
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn turn_start_updates_sandbox_and_cwd_between_turns_v2() -> Result<()> {
+    // When returning Result from a test, pass an Ok(()) to the skip macro
+    // so the early return type matches. The no-arg form returns unit.
+    skip_if_no_network!(Ok(()));
+
+    let tmp = TempDir::new()?;
+    let codex_home = tmp.path().join("codex_home");
+    std::fs::create_dir(&codex_home)?;
+    let workspace_root = tmp.path().join("workspace");
+    std::fs::create_dir(&workspace_root)?;
+    let first_cwd = workspace_root.join("turn1");
+    let second_cwd = workspace_root.join("turn2");
+    std::fs::create_dir(&first_cwd)?;
+    std::fs::create_dir(&second_cwd)?;
+
+    let responses = vec![
+        create_shell_sse_response(
+            vec![
+                "bash".to_string(),
+                "-lc".to_string(),
+                "echo first turn".to_string(),
+            ],
+            None,
+            Some(5000),
+            "call-first",
+        )?,
+        create_final_assistant_message_sse_response("done first")?,
+        create_shell_sse_response(
+            vec![
+                "bash".to_string(),
+                "-lc".to_string(),
+                "echo second turn".to_string(),
+            ],
+            None,
+            Some(5000),
+            "call-second",
+        )?,
+        create_final_assistant_message_sse_response("done second")?,
+    ];
+    let server = create_mock_chat_completions_server(responses).await;
+    create_config_toml(&codex_home, &server.uri(), "untrusted")?;
+
+    let mut mcp = McpProcess::new(&codex_home).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    // thread/start
+    let start_id = mcp
+        .send_thread_start_request(ThreadStartParams {
+            model: Some("mock-model".to_string()),
+            ..Default::default()
+        })
+        .await?;
+    let start_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
+    )
+    .await??;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(start_resp)?;
+
+    // first turn with workspace-write sandbox and first_cwd
+    let first_turn = mcp
+        .send_turn_start_request(TurnStartParams {
+            thread_id: thread.id.clone(),
+            input: vec![V2UserInput::Text {
+                text: "first turn".to_string(),
+            }],
+            cwd: Some(first_cwd.clone()),
+            approval_policy: Some(codex_app_server_protocol::AskForApproval::Never),
+            sandbox_policy: Some(codex_app_server_protocol::SandboxPolicy::WorkspaceWrite {
+                writable_roots: vec![first_cwd.clone()],
+                network_access: false,
+                exclude_tmpdir_env_var: false,
+                exclude_slash_tmp: false,
+            }),
+            model: Some("mock-model".to_string()),
+            effort: Some(ReasoningEffort::Medium),
+            summary: Some(ReasoningSummary::Auto),
+        })
+        .await?;
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(first_turn)),
+    )
+    .await??;
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("codex/event/task_complete"),
+    )
+    .await??;
+
+    // second turn with workspace-write and second_cwd, ensure exec begins in second_cwd
+    let second_turn = mcp
+        .send_turn_start_request(TurnStartParams {
+            thread_id: thread.id.clone(),
+            input: vec![V2UserInput::Text {
+                text: "second turn".to_string(),
+            }],
+            cwd: Some(second_cwd.clone()),
+            approval_policy: Some(codex_app_server_protocol::AskForApproval::Never),
+            sandbox_policy: Some(codex_app_server_protocol::SandboxPolicy::DangerFullAccess),
+            model: Some("mock-model".to_string()),
+            effort: Some(ReasoningEffort::Medium),
+            summary: Some(ReasoningSummary::Auto),
+        })
+        .await?;
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(second_turn)),
+    )
+    .await??;
+
+    let exec_begin_notification = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("codex/event/exec_command_begin"),
+    )
+    .await??;
+    let params = exec_begin_notification
+        .params
+        .clone()
+        .expect("exec_command_begin params");
+    let event: Event = serde_json::from_value(params).expect("deserialize exec begin event");
+    let exec_begin = match event.msg {
+        EventMsg::ExecCommandBegin(exec_begin) => exec_begin,
+        other => panic!("expected ExecCommandBegin event, got {other:?}"),
+    };
+    assert_eq!(exec_begin.cwd, second_cwd);
+    assert_eq!(
+        exec_begin.command,
+        vec![
+            "bash".to_string(),
+            "-lc".to_string(),
+            "echo second turn".to_string()
+        ]
+    );
+
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("codex/event/task_complete"),
+    )
+    .await??;
+
+    Ok(())
+}
+
+// Helper to create a config.toml pointing at the mock model server.
+fn create_config_toml(
+    codex_home: &Path,
+    server_uri: &str,
+    approval_policy: &str,
+) -> std::io::Result<()> {
+    let config_toml = codex_home.join("config.toml");
+    std::fs::write(
+        config_toml,
+        format!(
+            r#"
+model = "mock-model"
+approval_policy = "{approval_policy}"
+sandbox_mode = "read-only"
+
+model_provider = "mock_provider"
+
+[model_providers.mock_provider]
+name = "Mock provider for test"
+base_url = "{server_uri}/v1"
+wire_api = "chat"
+request_max_retries = 0
+stream_max_retries = 0
+"#
+        ),
+    )
+}

codex-rs/apply-patch/src/lib.rs

@@ -288,7 +288,7 @@ pub fn maybe_parse_apply_patch_verified(argv: &[String], cwd: &Path) -> MaybeApp
                             path,
                             ApplyPatchFileChange::Update {
                                 unified_diff,
-                                move_path: move_path.map(|p| cwd.join(p)),
+                                move_path: move_path.map(|p| effective_cwd.join(p)),
                                 new_content: contents,
                             },
                         );
@@ -1603,6 +1603,53 @@ g
         );
     }
 
+    #[test]
+    fn test_apply_patch_resolves_move_path_with_effective_cwd() {
+        let session_dir = tempdir().unwrap();
+        let worktree_rel = "alt";
+        let worktree_dir = session_dir.path().join(worktree_rel);
+        fs::create_dir_all(&worktree_dir).unwrap();
+
+        let source_name = "old.txt";
+        let dest_name = "renamed.txt";
+        let source_path = worktree_dir.join(source_name);
+        fs::write(&source_path, "before\n").unwrap();
+
+        let patch = wrap_patch(&format!(
+            r#"*** Update File: {source_name}
+*** Move to: {dest_name}
+@@
+-before
++after"#
+        ));
+
+        let shell_script = format!("cd {worktree_rel} && apply_patch <<'PATCH'\n{patch}\nPATCH");
+        let argv = vec!["bash".into(), "-lc".into(), shell_script];
+
+        let result = maybe_parse_apply_patch_verified(&argv, session_dir.path());
+        let action = match result {
+            MaybeApplyPatchVerified::Body(action) => action,
+            other => panic!("expected verified body, got {other:?}"),
+        };
+
+        assert_eq!(action.cwd, worktree_dir);
+
+        let change = action
+            .changes()
+            .get(&worktree_dir.join(source_name))
+            .expect("source file change present");
+
+        match change {
+            ApplyPatchFileChange::Update { move_path, .. } => {
+                assert_eq!(
+                    move_path.as_deref(),
+                    Some(worktree_dir.join(dest_name).as_path())
+                );
+            }
+            other => panic!("expected update change, got {other:?}"),
+        }
+    }
+
     #[test]
     fn test_apply_patch_fails_on_write_error() {
         let dir = tempdir().unwrap();

codex-rs/arg0/src/lib.rs

@@ -11,32 +11,7 @@ const LINUX_SANDBOX_ARG0: &str = "codex-linux-sandbox";
 const APPLY_PATCH_ARG0: &str = "apply_patch";
 const MISSPELLED_APPLY_PATCH_ARG0: &str = "applypatch";
 
-/// While we want to deploy the Codex CLI as a single executable for simplicity,
-/// we also want to expose some of its functionality as distinct CLIs, so we use
-/// the "arg0 trick" to determine which CLI to dispatch. This effectively allows
-/// us to simulate deploying multiple executables as a single binary on Mac and
-/// Linux (but not Windows).
-///
-/// When the current executable is invoked through the hard-link or alias named
-/// `codex-linux-sandbox` we *directly* execute
-/// [`codex_linux_sandbox::run_main`] (which never returns). Otherwise we:
-///
-/// 1.  Load `.env` values from `~/.codex/.env` before creating any threads.
-/// 2.  Construct a Tokio multi-thread runtime.
-/// 3.  Derive the path to the current executable (so children can re-invoke the
-///     sandbox) when running on Linux.
-/// 4.  Execute the provided async `main_fn` inside that runtime, forwarding any
-///     error. Note that `main_fn` receives `codex_linux_sandbox_exe:
-///     Option<PathBuf>`, as an argument, which is generally needed as part of
-///     constructing [`codex_core::config::Config`].
-///
-/// This function should be used to wrap any `main()` function in binary crates
-/// in this workspace that depends on these helper CLIs.
-pub fn arg0_dispatch_or_else<F, Fut>(main_fn: F) -> anyhow::Result<()>
-where
-    F: FnOnce(Option<PathBuf>) -> Fut,
-    Fut: Future<Output = anyhow::Result<()>>,
-{
+pub fn arg0_dispatch() -> Option<TempDir> {
     // Determine if we were invoked via the special alias.
     let mut args = std::env::args_os();
     let argv0 = args.next().unwrap_or_default();
@@ -76,10 +51,7 @@ where
     // before creating any threads/the Tokio runtime.
     load_dotenv();
 
-    // Retain the TempDir so it exists for the lifetime of the invocation of
-    // this executable. Admittedly, we could invoke `keep()` on it, but it
-    // would be nice to avoid leaving temporary directories behind, if possible.
-    let _path_entry = match prepend_path_entry_for_apply_patch() {
+    match prepend_path_entry_for_codex_aliases() {
         Ok(path_entry) => Some(path_entry),
         Err(err) => {
             // It is possible that Codex will proceed successfully even if
@@ -87,7 +59,39 @@ where
             eprintln!("WARNING: proceeding, even though we could not update PATH: {err}");
             None
         }
-    };
+    }
+}
+
+/// While we want to deploy the Codex CLI as a single executable for simplicity,
+/// we also want to expose some of its functionality as distinct CLIs, so we use
+/// the "arg0 trick" to determine which CLI to dispatch. This effectively allows
+/// us to simulate deploying multiple executables as a single binary on Mac and
+/// Linux (but not Windows).
+///
+/// When the current executable is invoked through the hard-link or alias named
+/// `codex-linux-sandbox` we *directly* execute
+/// [`codex_linux_sandbox::run_main`] (which never returns). Otherwise we:
+///
+/// 1.  Load `.env` values from `~/.codex/.env` before creating any threads.
+/// 2.  Construct a Tokio multi-thread runtime.
+/// 3.  Derive the path to the current executable (so children can re-invoke the
+///     sandbox) when running on Linux.
+/// 4.  Execute the provided async `main_fn` inside that runtime, forwarding any
+///     error. Note that `main_fn` receives `codex_linux_sandbox_exe:
+///     Option<PathBuf>`, as an argument, which is generally needed as part of
+///     constructing [`codex_core::config::Config`].
+///
+/// This function should be used to wrap any `main()` function in binary crates
+/// in this workspace that depends on these helper CLIs.
+pub fn arg0_dispatch_or_else<F, Fut>(main_fn: F) -> anyhow::Result<()>
+where
+    F: FnOnce(Option<PathBuf>) -> Fut,
+    Fut: Future<Output = anyhow::Result<()>>,
+{
+    // Retain the TempDir so it exists for the lifetime of the invocation of
+    // this executable. Admittedly, we could invoke `keep()` on it, but it
+    // would be nice to avoid leaving temporary directories behind, if possible.
+    let _path_entry = arg0_dispatch();
 
     // Regular invocation – create a Tokio runtime and execute the provided
     // async entry-point.
@@ -144,11 +148,16 @@ where
 ///
 /// IMPORTANT: This function modifies the PATH environment variable, so it MUST
 /// be called before multiple threads are spawned.
-fn prepend_path_entry_for_apply_patch() -> std::io::Result<TempDir> {
+pub fn prepend_path_entry_for_codex_aliases() -> std::io::Result<TempDir> {
     let temp_dir = TempDir::new()?;
     let path = temp_dir.path();
 
-    for filename in &[APPLY_PATCH_ARG0, MISSPELLED_APPLY_PATCH_ARG0] {
+    for filename in &[
+        APPLY_PATCH_ARG0,
+        MISSPELLED_APPLY_PATCH_ARG0,
+        #[cfg(target_os = "linux")]
+        LINUX_SANDBOX_ARG0,
+    ] {
         let exe = std::env::current_exe()?;
 
         #[cfg(unix)]

codex-rs/cli/Cargo.toml

@@ -30,15 +30,17 @@ codex-login = { workspace = true }
 codex-mcp-server = { workspace = true }
 codex-process-hardening = { workspace = true }
 codex-protocol = { workspace = true }
-codex-protocol-ts = { workspace = true }
 codex-responses-api-proxy = { workspace = true }
 codex-rmcp-client = { workspace = true }
 codex-stdio-to-uds = { workspace = true }
 codex-tui = { workspace = true }
 ctor = { workspace = true }
+libc = { workspace = true }
 owo-colors = { workspace = true }
+regex-lite = { workspace = true}
 serde_json = { workspace = true }
 supports-color = { workspace = true }
+toml = { workspace = true }
 tokio = { workspace = true, features = [
     "io-std",
     "macros",
@@ -46,6 +48,7 @@ tokio = { workspace = true, features = [
     "rt-multi-thread",
     "signal",
 ] }
+tracing = { workspace = true }
 
 [target.'cfg(target_os = "windows")'.dependencies]
 codex_windows_sandbox = { package = "codex-windows-sandbox", path = "../windows-sandbox-rs" }

codex-rs/cli/src/debug_sandbox.rs

@@ -1,3 +1,8 @@
+#[cfg(target_os = "macos")]
+mod pid_tracker;
+#[cfg(target_os = "macos")]
+mod seatbelt;
+
 use std::path::PathBuf;
 
 use codex_common::CliConfigOverrides;
@@ -5,6 +10,7 @@ use codex_core::config::Config;
 use codex_core::config::ConfigOverrides;
 use codex_core::exec_env::create_env;
 use codex_core::landlock::spawn_command_under_linux_sandbox;
+#[cfg(target_os = "macos")]
 use codex_core::seatbelt::spawn_command_under_seatbelt;
 use codex_core::spawn::StdioPolicy;
 use codex_protocol::config_types::SandboxMode;
@@ -14,12 +20,17 @@ use crate::SeatbeltCommand;
 use crate::WindowsCommand;
 use crate::exit_status::handle_exit_status;
 
+#[cfg(target_os = "macos")]
+use seatbelt::DenialLogger;
+
+#[cfg(target_os = "macos")]
 pub async fn run_command_under_seatbelt(
     command: SeatbeltCommand,
     codex_linux_sandbox_exe: Option<PathBuf>,
 ) -> anyhow::Result<()> {
     let SeatbeltCommand {
         full_auto,
+        log_denials,
         config_overrides,
         command,
     } = command;
@@ -29,10 +40,19 @@ pub async fn run_command_under_seatbelt(
         config_overrides,
         codex_linux_sandbox_exe,
         SandboxType::Seatbelt,
+        log_denials,
     )
     .await
 }
 
+#[cfg(not(target_os = "macos"))]
+pub async fn run_command_under_seatbelt(
+    _command: SeatbeltCommand,
+    _codex_linux_sandbox_exe: Option<PathBuf>,
+) -> anyhow::Result<()> {
+    anyhow::bail!("Seatbelt sandbox is only available on macOS");
+}
+
 pub async fn run_command_under_landlock(
     command: LandlockCommand,
     codex_linux_sandbox_exe: Option<PathBuf>,
@@ -48,6 +68,7 @@ pub async fn run_command_under_landlock(
         config_overrides,
         codex_linux_sandbox_exe,
         SandboxType::Landlock,
+        false,
     )
     .await
 }
@@ -67,11 +88,13 @@ pub async fn run_command_under_windows(
         config_overrides,
         codex_linux_sandbox_exe,
         SandboxType::Windows,
+        false,
     )
     .await
 }
 
 enum SandboxType {
+    #[cfg(target_os = "macos")]
     Seatbelt,
     Landlock,
     Windows,
@@ -83,6 +106,7 @@ async fn run_command_under_sandbox(
     config_overrides: CliConfigOverrides,
     codex_linux_sandbox_exe: Option<PathBuf>,
     sandbox_type: SandboxType,
+    log_denials: bool,
 ) -> anyhow::Result<()> {
     let sandbox_mode = create_sandbox_mode(full_auto);
     let config = Config::load_with_cli_overrides(
@@ -124,6 +148,9 @@ async fn run_command_under_sandbox(
             let cwd_clone = cwd.clone();
             let env_map = env.clone();
             let command_vec = command.clone();
+            let base_dir = config.codex_home.clone();
+
+            // Preflight audit is invoked elsewhere at the appropriate times.
             let res = tokio::task::spawn_blocking(move || {
                 run_windows_sandbox_capture(
                     policy_str,
@@ -132,6 +159,7 @@ async fn run_command_under_sandbox(
                     &cwd_clone,
                     env_map,
                     None,
+                    Some(base_dir.as_path()),
                 )
             })
             .await;
@@ -165,7 +193,13 @@ async fn run_command_under_sandbox(
         }
     }
 
+    #[cfg(target_os = "macos")]
+    let mut denial_logger = log_denials.then(DenialLogger::new).flatten();
+    #[cfg(not(target_os = "macos"))]
+    let _ = log_denials;
+
     let mut child = match sandbox_type {
+        #[cfg(target_os = "macos")]
         SandboxType::Seatbelt => {
             spawn_command_under_seatbelt(
                 command,
@@ -197,8 +231,27 @@ async fn run_command_under_sandbox(
             unreachable!("Windows sandbox should have been handled above");
         }
     };
+
+    #[cfg(target_os = "macos")]
+    if let Some(denial_logger) = &mut denial_logger {
+        denial_logger.on_child_spawn(&child);
+    }
+
     let status = child.wait().await?;
 
+    #[cfg(target_os = "macos")]
+    if let Some(denial_logger) = denial_logger {
+        let denials = denial_logger.finish().await;
+        eprintln!("\n=== Sandbox denials ===");
+        if denials.is_empty() {
+            eprintln!("None found.");
+        } else {
+            for seatbelt::SandboxDenial { name, capability } in denials {
+                eprintln!("({name}) {capability}");
+            }
+        }
+    }
+
     handle_exit_status(status);
 }
 

codex-rs/cli/src/debug_sandbox/pid_tracker.rs

@@ -0,0 +1,372 @@
+use std::collections::HashSet;
+use tokio::task::JoinHandle;
+use tracing::warn;
+
+/// Tracks the (recursive) descendants of a process by using `kqueue` to watch for fork events, and
+/// `proc_listchildpids` to list the children of a process.
+pub(crate) struct PidTracker {
+    kq: libc::c_int,
+    handle: JoinHandle<HashSet<i32>>,
+}
+
+impl PidTracker {
+    pub(crate) fn new(root_pid: i32) -> Option<Self> {
+        if root_pid <= 0 {
+            return None;
+        }
+
+        let kq = unsafe { libc::kqueue() };
+        let handle = tokio::task::spawn_blocking(move || track_descendants(kq, root_pid));
+
+        Some(Self { kq, handle })
+    }
+
+    pub(crate) async fn stop(self) -> HashSet<i32> {
+        trigger_stop_event(self.kq);
+        self.handle.await.unwrap_or_default()
+    }
+}
+
+unsafe extern "C" {
+    fn proc_listchildpids(
+        ppid: libc::c_int,
+        buffer: *mut libc::c_void,
+        buffersize: libc::c_int,
+    ) -> libc::c_int;
+}
+
+/// Wrap proc_listchildpids.
+fn list_child_pids(parent: i32) -> Vec<i32> {
+    unsafe {
+        let mut capacity: usize = 16;
+        loop {
+            let mut buf: Vec<i32> = vec![0; capacity];
+            let count = proc_listchildpids(
+                parent as libc::c_int,
+                buf.as_mut_ptr() as *mut libc::c_void,
+                (buf.len() * std::mem::size_of::<i32>()) as libc::c_int,
+            );
+            if count <= 0 {
+                return Vec::new();
+            }
+            let returned = count as usize;
+            if returned < capacity {
+                buf.truncate(returned);
+                return buf;
+            }
+            capacity = capacity.saturating_mul(2).max(returned + 16);
+        }
+    }
+}
+
+fn pid_is_alive(pid: i32) -> bool {
+    if pid <= 0 {
+        return false;
+    }
+    let res = unsafe { libc::kill(pid as libc::pid_t, 0) };
+    if res == 0 {
+        true
+    } else {
+        matches!(
+            std::io::Error::last_os_error().raw_os_error(),
+            Some(libc::EPERM)
+        )
+    }
+}
+
+enum WatchPidError {
+    ProcessGone,
+    Other(std::io::Error),
+}
+
+/// Add `pid` to the watch list in `kq`.
+fn watch_pid(kq: libc::c_int, pid: i32) -> Result<(), WatchPidError> {
+    if pid <= 0 {
+        return Err(WatchPidError::ProcessGone);
+    }
+
+    let kev = libc::kevent {
+        ident: pid as libc::uintptr_t,
+        filter: libc::EVFILT_PROC,
+        flags: libc::EV_ADD | libc::EV_CLEAR,
+        fflags: libc::NOTE_FORK | libc::NOTE_EXEC | libc::NOTE_EXIT,
+        data: 0,
+        udata: std::ptr::null_mut(),
+    };
+
+    let res = unsafe { libc::kevent(kq, &kev, 1, std::ptr::null_mut(), 0, std::ptr::null()) };
+    if res < 0 {
+        let err = std::io::Error::last_os_error();
+        if err.raw_os_error() == Some(libc::ESRCH) {
+            Err(WatchPidError::ProcessGone)
+        } else {
+            Err(WatchPidError::Other(err))
+        }
+    } else {
+        Ok(())
+    }
+}
+
+fn watch_children(
+    kq: libc::c_int,
+    parent: i32,
+    seen: &mut HashSet<i32>,
+    active: &mut HashSet<i32>,
+) {
+    for child_pid in list_child_pids(parent) {
+        add_pid_watch(kq, child_pid, seen, active);
+    }
+}
+
+/// Watch `pid` and its children, updating `seen` and `active` sets.
+fn add_pid_watch(kq: libc::c_int, pid: i32, seen: &mut HashSet<i32>, active: &mut HashSet<i32>) {
+    if pid <= 0 {
+        return;
+    }
+
+    let newly_seen = seen.insert(pid);
+    let mut should_recurse = newly_seen;
+
+    if active.insert(pid) {
+        match watch_pid(kq, pid) {
+            Ok(()) => {
+                should_recurse = true;
+            }
+            Err(WatchPidError::ProcessGone) => {
+                active.remove(&pid);
+                return;
+            }
+            Err(WatchPidError::Other(err)) => {
+                warn!("failed to watch pid {pid}: {err}");
+                active.remove(&pid);
+                return;
+            }
+        }
+    }
+
+    if should_recurse {
+        watch_children(kq, pid, seen, active);
+    }
+}
+const STOP_IDENT: libc::uintptr_t = 1;
+
+fn register_stop_event(kq: libc::c_int) -> bool {
+    let kev = libc::kevent {
+        ident: STOP_IDENT,
+        filter: libc::EVFILT_USER,
+        flags: libc::EV_ADD | libc::EV_CLEAR,
+        fflags: 0,
+        data: 0,
+        udata: std::ptr::null_mut(),
+    };
+
+    let res = unsafe { libc::kevent(kq, &kev, 1, std::ptr::null_mut(), 0, std::ptr::null()) };
+    res >= 0
+}
+
+fn trigger_stop_event(kq: libc::c_int) {
+    if kq < 0 {
+        return;
+    }
+
+    let kev = libc::kevent {
+        ident: STOP_IDENT,
+        filter: libc::EVFILT_USER,
+        flags: 0,
+        fflags: libc::NOTE_TRIGGER,
+        data: 0,
+        udata: std::ptr::null_mut(),
+    };
+
+    let _ = unsafe { libc::kevent(kq, &kev, 1, std::ptr::null_mut(), 0, std::ptr::null()) };
+}
+
+/// Put all of the above together to track all the descendants of `root_pid`.
+fn track_descendants(kq: libc::c_int, root_pid: i32) -> HashSet<i32> {
+    if kq < 0 {
+        let mut seen = HashSet::new();
+        seen.insert(root_pid);
+        return seen;
+    }
+
+    if !register_stop_event(kq) {
+        let mut seen = HashSet::new();
+        seen.insert(root_pid);
+        let _ = unsafe { libc::close(kq) };
+        return seen;
+    }
+
+    let mut seen: HashSet<i32> = HashSet::new();
+    let mut active: HashSet<i32> = HashSet::new();
+
+    add_pid_watch(kq, root_pid, &mut seen, &mut active);
+
+    const EVENTS_CAP: usize = 32;
+    let mut events: [libc::kevent; EVENTS_CAP] =
+        unsafe { std::mem::MaybeUninit::zeroed().assume_init() };
+
+    let mut stop_requested = false;
+    loop {
+        if active.is_empty() {
+            if !pid_is_alive(root_pid) {
+                break;
+            }
+            add_pid_watch(kq, root_pid, &mut seen, &mut active);
+            if active.is_empty() {
+                continue;
+            }
+        }
+
+        let nev = unsafe {
+            libc::kevent(
+                kq,
+                std::ptr::null::<libc::kevent>(),
+                0,
+                events.as_mut_ptr(),
+                EVENTS_CAP as libc::c_int,
+                std::ptr::null(),
+            )
+        };
+
+        if nev < 0 {
+            let err = std::io::Error::last_os_error();
+            if err.kind() == std::io::ErrorKind::Interrupted {
+                continue;
+            }
+            break;
+        }
+
+        if nev == 0 {
+            continue;
+        }
+
+        for ev in events.iter().take(nev as usize) {
+            let pid = ev.ident as i32;
+
+            if ev.filter == libc::EVFILT_USER && ev.ident == STOP_IDENT {
+                stop_requested = true;
+                break;
+            }
+
+            if (ev.flags & libc::EV_ERROR) != 0 {
+                if ev.data == libc::ESRCH as isize {
+                    active.remove(&pid);
+                }
+                continue;
+            }
+
+            if (ev.fflags & libc::NOTE_FORK) != 0 {
+                watch_children(kq, pid, &mut seen, &mut active);
+            }
+
+            if (ev.fflags & libc::NOTE_EXIT) != 0 {
+                active.remove(&pid);
+            }
+        }
+
+        if stop_requested {
+            break;
+        }
+    }
+
+    let _ = unsafe { libc::close(kq) };
+
+    seen
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::process::Command;
+    use std::process::Stdio;
+    use std::time::Duration;
+
+    #[test]
+    fn pid_is_alive_detects_current_process() {
+        let pid = std::process::id() as i32;
+        assert!(pid_is_alive(pid));
+    }
+
+    #[cfg(target_os = "macos")]
+    #[test]
+    fn list_child_pids_includes_spawned_child() {
+        let mut child = Command::new("/bin/sleep")
+            .arg("5")
+            .stdin(Stdio::null())
+            .spawn()
+            .expect("failed to spawn child process");
+
+        let child_pid = child.id() as i32;
+        let parent_pid = std::process::id() as i32;
+
+        let mut found = false;
+        for _ in 0..100 {
+            if list_child_pids(parent_pid).contains(&child_pid) {
+                found = true;
+                break;
+            }
+            std::thread::sleep(Duration::from_millis(10));
+        }
+
+        let _ = child.kill();
+        let _ = child.wait();
+
+        assert!(found, "expected to find child pid {child_pid} in list");
+    }
+
+    #[cfg(target_os = "macos")]
+    #[tokio::test]
+    async fn pid_tracker_collects_spawned_children() {
+        let tracker = PidTracker::new(std::process::id() as i32).expect("failed to create tracker");
+
+        let mut child = Command::new("/bin/sleep")
+            .arg("0.1")
+            .stdin(Stdio::null())
+            .spawn()
+            .expect("failed to spawn child process");
+
+        let child_pid = child.id() as i32;
+        let parent_pid = std::process::id() as i32;
+
+        let _ = child.wait();
+
+        let seen = tracker.stop().await;
+
+        assert!(
+            seen.contains(&parent_pid),
+            "expected tracker to include parent pid {parent_pid}"
+        );
+        assert!(
+            seen.contains(&child_pid),
+            "expected tracker to include child pid {child_pid}"
+        );
+    }
+
+    #[cfg(target_os = "macos")]
+    #[tokio::test]
+    async fn pid_tracker_collects_bash_subshell_descendants() {
+        let tracker = PidTracker::new(std::process::id() as i32).expect("failed to create tracker");
+
+        let child = Command::new("/bin/bash")
+            .arg("-c")
+            .arg("(sleep 0.1 & echo $!; wait)")
+            .stdin(Stdio::null())
+            .stdout(Stdio::piped())
+            .stderr(Stdio::null())
+            .spawn()
+            .expect("failed to spawn bash");
+
+        let output = child.wait_with_output().unwrap().stdout;
+        let subshell_pid = String::from_utf8_lossy(&output)
+            .trim()
+            .parse::<i32>()
+            .expect("failed to parse subshell pid");
+
+        let seen = tracker.stop().await;
+
+        assert!(
+            seen.contains(&subshell_pid),
+            "expected tracker to include subshell pid {subshell_pid}"
+        );
+    }
+}

codex-rs/cli/src/debug_sandbox/seatbelt.rs

@@ -0,0 +1,114 @@
+use std::collections::HashSet;
+use tokio::io::AsyncBufReadExt;
+use tokio::process::Child;
+use tokio::task::JoinHandle;
+
+use super::pid_tracker::PidTracker;
+
+pub struct SandboxDenial {
+    pub name: String,
+    pub capability: String,
+}
+
+pub struct DenialLogger {
+    log_stream: Child,
+    pid_tracker: Option<PidTracker>,
+    log_reader: Option<JoinHandle<Vec<u8>>>,
+}
+
+impl DenialLogger {
+    pub(crate) fn new() -> Option<Self> {
+        let mut log_stream = start_log_stream()?;
+        let stdout = log_stream.stdout.take()?;
+        let log_reader = tokio::spawn(async move {
+            let mut reader = tokio::io::BufReader::new(stdout);
+            let mut logs = Vec::new();
+            let mut chunk = Vec::new();
+            loop {
+                match reader.read_until(b'\n', &mut chunk).await {
+                    Ok(0) | Err(_) => break,
+                    Ok(_) => {
+                        logs.extend_from_slice(&chunk);
+                        chunk.clear();
+                    }
+                }
+            }
+            logs
+        });
+
+        Some(Self {
+            log_stream,
+            pid_tracker: None,
+            log_reader: Some(log_reader),
+        })
+    }
+
+    pub(crate) fn on_child_spawn(&mut self, child: &Child) {
+        if let Some(root_pid) = child.id() {
+            self.pid_tracker = PidTracker::new(root_pid as i32);
+        }
+    }
+
+    pub(crate) async fn finish(mut self) -> Vec<SandboxDenial> {
+        let pid_set = match self.pid_tracker {
+            Some(tracker) => tracker.stop().await,
+            None => Default::default(),
+        };
+
+        if pid_set.is_empty() {
+            return Vec::new();
+        }
+
+        let _ = self.log_stream.kill().await;
+        let _ = self.log_stream.wait().await;
+
+        let logs_bytes = match self.log_reader.take() {
+            Some(handle) => handle.await.unwrap_or_default(),
+            None => Vec::new(),
+        };
+        let logs = String::from_utf8_lossy(&logs_bytes);
+
+        let mut seen: HashSet<(String, String)> = HashSet::new();
+        let mut denials: Vec<SandboxDenial> = Vec::new();
+        for line in logs.lines() {
+            if let Ok(json) = serde_json::from_str::<serde_json::Value>(line)
+                && let Some(msg) = json.get("eventMessage").and_then(|v| v.as_str())
+                && let Some((pid, name, capability)) = parse_message(msg)
+                && pid_set.contains(&pid)
+                && seen.insert((name.clone(), capability.clone()))
+            {
+                denials.push(SandboxDenial { name, capability });
+            }
+        }
+        denials
+    }
+}
+
+fn start_log_stream() -> Option<Child> {
+    use std::process::Stdio;
+
+    const PREDICATE: &str = r#"(((processID == 0) AND (senderImagePath CONTAINS "/Sandbox")) OR (subsystem == "com.apple.sandbox.reporting"))"#;
+
+    tokio::process::Command::new("log")
+        .args(["stream", "--style", "ndjson", "--predicate", PREDICATE])
+        .stdin(Stdio::null())
+        .stdout(Stdio::piped())
+        .stderr(Stdio::null())
+        .kill_on_drop(true)
+        .spawn()
+        .ok()
+}
+
+fn parse_message(msg: &str) -> Option<(i32, String, String)> {
+    // Example message:
+    // Sandbox: processname(1234) deny(1) capability-name args...
+    static RE: std::sync::OnceLock<regex_lite::Regex> = std::sync::OnceLock::new();
+    let re = RE.get_or_init(|| {
+        #[expect(clippy::unwrap_used)]
+        regex_lite::Regex::new(r"^Sandbox:\s*(.+?)\((\d+)\)\s+deny\(.*?\)\s*(.+)$").unwrap()
+    });
+
+    let (_, [name, pid_str, capability]) = re.captures(msg)?.extract();
+    let pid = pid_str.trim().parse::<i32>().ok()?;
+    Some((pid, name.to_string(), capability.to_string()))
+}

codex-rs/cli/src/lib.rs

@@ -11,6 +11,10 @@ pub struct SeatbeltCommand {
     #[arg(long = "full-auto", default_value_t = false)]
     pub full_auto: bool,
 
+    /// While the command runs, capture macOS sandbox denials via `log stream` and print them after exit
+    #[arg(long = "log-denials", default_value_t = false)]
+    pub log_denials: bool,
+
     #[clap(skip)]
     pub config_overrides: CliConfigOverrides,
 

codex-rs/cli/src/main.rs

@@ -1,3 +1,4 @@
+use clap::Args;
 use clap::CommandFactory;
 use clap::Parser;
 use clap_complete::Shell;
@@ -20,14 +21,17 @@ use codex_exec::Cli as ExecCli;
 use codex_responses_api_proxy::Args as ResponsesApiProxyArgs;
 use codex_tui::AppExitInfo;
 use codex_tui::Cli as TuiCli;
-use codex_tui::updates::UpdateAction;
+use codex_tui::update_action::UpdateAction;
 use owo_colors::OwoColorize;
 use std::path::PathBuf;
 use supports_color::Stream;
 
 mod mcp_cmd;
+#[cfg(not(windows))]
+mod wsl_paths;
 
 use crate::mcp_cmd::McpCli;
+
 use codex_core::config::Config;
 use codex_core::config::ConfigOverrides;
 use codex_core::features::is_known_feature_key;
@@ -79,8 +83,8 @@ enum Subcommand {
     /// [experimental] Run the Codex MCP server (stdio transport).
     McpServer,
 
-    /// [experimental] Run the app server.
-    AppServer,
+    /// [experimental] Run the app server or related tooling.
+    AppServer(AppServerCommand),
 
     /// Generate shell completion scripts.
     Completion(CompletionCommand),
@@ -96,9 +100,6 @@ enum Subcommand {
     /// Resume a previous interactive session (picker by default; use --last to continue the most recent).
     Resume(ResumeCommand),
 
-    /// Internal: generate TypeScript protocol bindings.
-    #[clap(hide = true)]
-    GenerateTs(GenerateTsCommand),
     /// [EXPERIMENTAL] Browse tasks from Codex Cloud and apply changes locally.
     #[clap(name = "cloud", alias = "cloud-tasks")]
     Cloud(CloudTasksCli),
@@ -205,6 +206,22 @@ struct LogoutCommand {
 }
 
 #[derive(Debug, Parser)]
+struct AppServerCommand {
+    /// Omit to run the app server; specify a subcommand for tooling.
+    #[command(subcommand)]
+    subcommand: Option<AppServerSubcommand>,
+}
+
+#[derive(Debug, clap::Subcommand)]
+enum AppServerSubcommand {
+    /// [experimental] Generate TypeScript bindings for the app server protocol.
+    GenerateTs(GenerateTsCommand),
+
+    /// [experimental] Generate JSON Schema for the app server protocol.
+    GenerateJsonSchema(GenerateJsonSchemaCommand),
+}
+
+#[derive(Debug, Args)]
 struct GenerateTsCommand {
     /// Output directory where .ts files will be written
     #[arg(short = 'o', long = "out", value_name = "DIR")]
@@ -215,6 +232,13 @@ struct GenerateTsCommand {
     prettier: Option<PathBuf>,
 }
 
+#[derive(Debug, Args)]
+struct GenerateJsonSchemaCommand {
+    /// Output directory where the schema bundle will be written
+    #[arg(short = 'o', long = "out", value_name = "DIR")]
+    out_dir: PathBuf,
+}
+
 #[derive(Debug, Parser)]
 struct StdioToUdsCommand {
     /// Path to the Unix domain socket to connect to.
@@ -267,10 +291,30 @@ fn handle_app_exit(exit_info: AppExitInfo) -> anyhow::Result<()> {
 /// Run the update action and print the result.
 fn run_update_action(action: UpdateAction) -> anyhow::Result<()> {
     println!();
-    let (cmd, args) = action.command_args();
     let cmd_str = action.command_str();
     println!("Updating Codex via `{cmd_str}`...");
-    let status = std::process::Command::new(cmd).args(args).status()?;
+
+    let status = {
+        #[cfg(windows)]
+        {
+            // On Windows, run via cmd.exe so .CMD/.BAT are correctly resolved (PATHEXT semantics).
+            std::process::Command::new("cmd")
+                .args(["/C", &cmd_str])
+                .status()?
+        }
+        #[cfg(not(windows))]
+        {
+            let (cmd, args) = action.command_args();
+            let command_path = crate::wsl_paths::normalize_for_wsl(cmd);
+            let normalized_args: Vec<String> = args
+                .iter()
+                .map(crate::wsl_paths::normalize_for_wsl)
+                .collect();
+            std::process::Command::new(&command_path)
+                .args(&normalized_args)
+                .status()?
+        }
+    };
     if !status.success() {
         anyhow::bail!("`{cmd_str}` failed with status {status}");
     }
@@ -387,9 +431,20 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
             prepend_config_flags(&mut mcp_cli.config_overrides, root_config_overrides.clone());
             mcp_cli.run().await?;
         }
-        Some(Subcommand::AppServer) => {
-            codex_app_server::run_main(codex_linux_sandbox_exe, root_config_overrides).await?;
-        }
+        Some(Subcommand::AppServer(app_server_cli)) => match app_server_cli.subcommand {
+            None => {
+                codex_app_server::run_main(codex_linux_sandbox_exe, root_config_overrides).await?;
+            }
+            Some(AppServerSubcommand::GenerateTs(gen_cli)) => {
+                codex_app_server_protocol::generate_ts(
+                    &gen_cli.out_dir,
+                    gen_cli.prettier.as_deref(),
+                )?;
+            }
+            Some(AppServerSubcommand::GenerateJsonSchema(gen_cli)) => {
+                codex_app_server_protocol::generate_json(&gen_cli.out_dir)?;
+            }
+        },
         Some(Subcommand::Resume(ResumeCommand {
             session_id,
             last,
@@ -504,21 +559,24 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
             tokio::task::spawn_blocking(move || codex_stdio_to_uds::run(socket_path.as_path()))
                 .await??;
         }
-        Some(Subcommand::GenerateTs(gen_cli)) => {
-            codex_protocol_ts::generate_ts(&gen_cli.out_dir, gen_cli.prettier.as_deref())?;
-        }
         Some(Subcommand::Features(FeaturesCli { sub })) => match sub {
             FeaturesSubcommand::List => {
                 // Respect root-level `-c` overrides plus top-level flags like `--profile`.
-                let cli_kv_overrides = root_config_overrides
+                let mut cli_kv_overrides = root_config_overrides
                     .parse_overrides()
                     .map_err(anyhow::Error::msg)?;
 
+                // Honor `--search` via the new feature toggle.
+                if interactive.web_search {
+                    cli_kv_overrides.push((
+                        "features.web_search_request".to_string(),
+                        toml::Value::Boolean(true),
+                    ));
+                }
+
                 // Thread through relevant top-level flags (at minimum, `--profile`).
-                // Also honor `--search` since it maps to a feature toggle.
                 let overrides = ConfigOverrides {
                     config_profile: interactive.config_profile.clone(),
-                    tools_web_search_request: interactive.web_search.then_some(true),
                     ..Default::default()
                 };
 

codex-rs/cli/src/mcp_cmd.rs

@@ -353,7 +353,9 @@ async fn run_login(config_overrides: &CliConfigOverrides, login_args: LoginArgs)
         .context("failed to load configuration")?;
 
     if !config.features.enabled(Feature::RmcpClient) {
-        bail!("OAuth login is only supported when [feature].rmcp_client is true in config.toml.");
+        bail!(
+            "OAuth login is only supported when [features].rmcp_client is true in config.toml. See https://github.com/openai/codex/blob/main/docs/config.md#feature-flags for details."
+        );
     }
 
     let LoginArgs { name, scopes } = login_args;

codex-rs/cli/src/wsl_paths.rs

@@ -0,0 +1,76 @@
+use std::ffi::OsStr;
+
+/// WSL-specific path helpers used by the updater logic.
+///
+/// See https://github.com/openai/codex/issues/6086.
+pub fn is_wsl() -> bool {
+    #[cfg(target_os = "linux")]
+    {
+        if std::env::var_os("WSL_DISTRO_NAME").is_some() {
+            return true;
+        }
+        match std::fs::read_to_string("/proc/version") {
+            Ok(version) => version.to_lowercase().contains("microsoft"),
+            Err(_) => false,
+        }
+    }
+    #[cfg(not(target_os = "linux"))]
+    {
+        false
+    }
+}
+
+/// Convert a Windows absolute path (`C:\foo\bar` or `C:/foo/bar`) to a WSL mount path (`/mnt/c/foo/bar`).
+/// Returns `None` if the input does not look like a Windows drive path.
+pub fn win_path_to_wsl(path: &str) -> Option<String> {
+    let bytes = path.as_bytes();
+    if bytes.len() < 3
+        || bytes[1] != b':'
+        || !(bytes[2] == b'\\' || bytes[2] == b'/')
+        || !bytes[0].is_ascii_alphabetic()
+    {
+        return None;
+    }
+    let drive = (bytes[0] as char).to_ascii_lowercase();
+    let tail = path[3..].replace('\\', "/");
+    if tail.is_empty() {
+        return Some(format!("/mnt/{drive}"));
+    }
+    Some(format!("/mnt/{drive}/{tail}"))
+}
+
+/// If under WSL and given a Windows-style path, return the equivalent `/mnt/<drive>/…` path.
+/// Otherwise returns the input unchanged.
+pub fn normalize_for_wsl<P: AsRef<OsStr>>(path: P) -> String {
+    let value = path.as_ref().to_string_lossy().to_string();
+    if !is_wsl() {
+        return value;
+    }
+    if let Some(mapped) = win_path_to_wsl(&value) {
+        return mapped;
+    }
+    value
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn win_to_wsl_basic() {
+        assert_eq!(
+            win_path_to_wsl(r"C:\Temp\codex.zip").as_deref(),
+            Some("/mnt/c/Temp/codex.zip")
+        );
+        assert_eq!(
+            win_path_to_wsl("D:/Work/codex.tgz").as_deref(),
+            Some("/mnt/d/Work/codex.tgz")
+        );
+        assert!(win_path_to_wsl("/home/user/codex").is_none());
+    }
+
+    #[test]
+    fn normalize_is_noop_on_unix_paths() {
+        assert_eq!(normalize_for_wsl("/home/u/x"), "/home/u/x");
+    }
+}

codex-rs/cloud-tasks/src/lib.rs

@@ -8,6 +8,7 @@ pub mod util;
 pub use cli::Cli;
 
 use anyhow::anyhow;
+use codex_login::AuthManager;
 use std::io::IsTerminal;
 use std::io::Read;
 use std::path::PathBuf;
@@ -56,20 +57,8 @@ async fn init_backend(user_agent_suffix: &str) -> anyhow::Result<BackendContext>
     };
     append_error_log(format!("startup: base_url={base_url} path_style={style}"));
 
-    let auth = match codex_core::config::find_codex_home()
-        .ok()
-        .map(|home| {
-            let store_mode = codex_core::config::Config::load_from_base_config_with_overrides(
-                codex_core::config::ConfigToml::default(),
-                codex_core::config::ConfigOverrides::default(),
-                home.clone(),
-            )
-            .map(|cfg| cfg.cli_auth_credentials_store_mode)
-            .unwrap_or_default();
-            codex_login::AuthManager::new(home, false, store_mode)
-        })
-        .and_then(|am| am.auth())
-    {
+    let auth_manager = util::load_auth_manager().await;
+    let auth = match auth_manager.as_ref().and_then(AuthManager::auth) {
         Some(auth) => auth,
         None => {
             eprintln!(
@@ -1044,7 +1033,7 @@ pub async fn run_main(cli: Cli, _codex_linux_sandbox_exe: Option<PathBuf>) -> an
                             // Close task modal/pending apply if present before opening env modal
                             app.diff_overlay = None;
                             app.env_modal = Some(app::EnvModalState { query: String::new(), selected: 0 });
-                            // Cache environments until user explicitly refreshes with 'r' inside the modal.
+                            // Cache environments while the modal is open to avoid repeated fetches.
                             let should_fetch = app.environments.is_empty();
                             if should_fetch {
                                 app.env_loading = true;
@@ -1115,7 +1104,7 @@ pub async fn run_main(cli: Cli, _codex_linux_sandbox_exe: Option<PathBuf>) -> an
                                                     let _ = tx.send(evt);
                                                 });
                                             } else {
-                                                app.status = "No environment selected (press 'e' to choose)".to_string();
+                                                app.status = "No environment selected".to_string();
                                             }
                                     }
                                     needs_redraw = true;
@@ -1313,18 +1302,6 @@ pub async fn run_main(cli: Cli, _codex_linux_sandbox_exe: Option<PathBuf>) -> an
                             // Environment modal key handling
                             match key.code {
                                 KeyCode::Esc => { app.env_modal = None; needs_redraw = true; }
-                                KeyCode::Char('r') | KeyCode::Char('R') => {
-                                    // Trigger refresh of environments
-                                    app.env_loading = true; app.env_error = None; needs_redraw = true;
-                                    let _ = frame_tx.send(Instant::now() + Duration::from_millis(100));
-                                    let tx = tx.clone();
-                                    tokio::spawn(async move {
-            let base_url = crate::util::normalize_base_url(&std::env::var("CODEX_CLOUD_TASKS_BASE_URL").unwrap_or_else(|_| "https://chatgpt.com/backend-api".to_string()));
-            let headers = crate::util::build_chatgpt_headers().await;
-                                        let res = crate::env_detect::list_environments(&base_url, &headers).await;
-                                        let _ = tx.send(app::AppEvent::EnvironmentsLoaded(res));
-                                    });
-                                }
                                 KeyCode::Char(ch) if !key.modifiers.contains(KeyModifiers::CONTROL) && !key.modifiers.contains(KeyModifiers::ALT) => {
                                     if let Some(m) = app.env_modal.as_mut() { m.query.push(ch); }
                                     needs_redraw = true;
@@ -1431,7 +1408,7 @@ pub async fn run_main(cli: Cli, _codex_linux_sandbox_exe: Option<PathBuf>) -> an
                                 }
                                 KeyCode::Char('o') | KeyCode::Char('O') => {
                                     app.env_modal = Some(app::EnvModalState { query: String::new(), selected: 0 });
-                                    // Cache environments until user explicitly refreshes with 'r' inside the modal.
+                                    // Cache environments while the modal is open to avoid repeated fetches.
                                     let should_fetch = app.environments.is_empty();
                                     if should_fetch { app.env_loading = true; app.env_error = None; }
                                     needs_redraw = true;
@@ -1744,6 +1721,7 @@ mod tests {
     use ratatui::layout::Rect;
 
     #[test]
+    #[ignore = "very slow"]
     fn composer_input_renders_typed_characters() {
         let mut composer = ComposerInput::new();
         let key = KeyEvent::new(KeyCode::Char('a'), KeyModifiers::NONE);

codex-rs/cloud-tasks/src/ui.rs

@@ -945,9 +945,7 @@ pub fn draw_env_modal(frame: &mut Frame, area: Rect, app: &mut App) {
 
     // Subheader with usage hints (dim cyan)
     let subheader = Paragraph::new(Line::from(
-        "Type to search, Enter select, Esc cancel; r refresh"
-            .cyan()
-            .dim(),
+        "Type to search, Enter select, Esc cancel".cyan().dim(),
     ))
     .wrap(Wrap { trim: true });
     frame.render_widget(subheader, rows[0]);

codex-rs/cloud-tasks/src/util.rs

@@ -2,6 +2,10 @@ use base64::Engine as _;
 use chrono::Utc;
 use reqwest::header::HeaderMap;
 
+use codex_core::config::Config;
+use codex_core::config::ConfigOverrides;
+use codex_login::AuthManager;
+
 pub fn set_user_agent_suffix(suffix: &str) {
     if let Ok(mut guard) = codex_core::default_client::USER_AGENT_SUFFIX.lock() {
         guard.replace(suffix.to_string());
@@ -54,6 +58,18 @@ pub fn extract_chatgpt_account_id(token: &str) -> Option<String> {
         .map(str::to_string)
 }
 
+pub async fn load_auth_manager() -> Option<AuthManager> {
+    // TODO: pass in cli overrides once cloud tasks properly support them.
+    let config = Config::load_with_cli_overrides(Vec::new(), ConfigOverrides::default())
+        .await
+        .ok()?;
+    Some(AuthManager::new(
+        config.codex_home,
+        false,
+        config.cli_auth_credentials_store_mode,
+    ))
+}
+
 /// Build headers for ChatGPT-backed requests: `User-Agent`, optional `Authorization`,
 /// and optional `ChatGPT-Account-Id`.
 pub async fn build_chatgpt_headers() -> HeaderMap {
@@ -69,31 +85,22 @@ pub async fn build_chatgpt_headers() -> HeaderMap {
         USER_AGENT,
         HeaderValue::from_str(&ua).unwrap_or(HeaderValue::from_static("codex-cli")),
     );
-    if let Ok(home) = codex_core::config::find_codex_home() {
-        let store_mode = codex_core::config::Config::load_from_base_config_with_overrides(
-            codex_core::config::ConfigToml::default(),
-            codex_core::config::ConfigOverrides::default(),
-            home.clone(),
-        )
-        .map(|cfg| cfg.cli_auth_credentials_store_mode)
-        .unwrap_or_default();
-        let am = codex_login::AuthManager::new(home, false, store_mode);
-        if let Some(auth) = am.auth()
-            && let Ok(tok) = auth.get_token().await
-            && !tok.is_empty()
+    if let Some(am) = load_auth_manager().await
+        && let Some(auth) = am.auth()
+        && let Ok(tok) = auth.get_token().await
+        && !tok.is_empty()
+    {
+        let v = format!("Bearer {tok}");
+        if let Ok(hv) = HeaderValue::from_str(&v) {
+            headers.insert(AUTHORIZATION, hv);
+        }
+        if let Some(acc) = auth
+            .get_account_id()
+            .or_else(|| extract_chatgpt_account_id(&tok))
+            && let Ok(name) = HeaderName::from_bytes(b"ChatGPT-Account-Id")
+            && let Ok(hv) = HeaderValue::from_str(&acc)
         {
-            let v = format!("Bearer {tok}");
-            if let Ok(hv) = HeaderValue::from_str(&v) {
-                headers.insert(AUTHORIZATION, hv);
-            }
-            if let Some(acc) = auth
-                .get_account_id()
-                .or_else(|| extract_chatgpt_account_id(&tok))
-                && let Ok(name) = HeaderName::from_bytes(b"ChatGPT-Account-Id")
-                && let Ok(hv) = HeaderValue::from_str(&acc)
-            {
-                headers.insert(name, hv);
-            }
+            headers.insert(name, hv);
         }
     }
     headers

codex-rs/common/Cargo.toml

@@ -8,9 +8,10 @@ workspace = true
 
 [dependencies]
 clap = { workspace = true, features = ["derive", "wrap_help"], optional = true }
+codex-app-server-protocol = { workspace = true }
 codex-core = { workspace = true }
 codex-protocol = { workspace = true }
-codex-app-server-protocol = { workspace = true }
+once_cell = { workspace = true }
 serde = { workspace = true, optional = true }
 toml = { workspace = true, optional = true }
 

codex-rs/common/src/config_override.rs

@@ -19,8 +19,8 @@ use toml::Value;
 pub struct CliConfigOverrides {
     /// Override a configuration value that would otherwise be loaded from
     /// `~/.codex/config.toml`. Use a dotted path (`foo.bar.baz`) to override
-    /// nested values. The `value` portion is parsed as JSON. If it fails to
-    /// parse as JSON, the raw string is used as a literal.
+    /// nested values. The `value` portion is parsed as TOML. If it fails to
+    /// parse as TOML, the raw string is used as a literal.
     ///
     /// Examples:
     ///   - `-c model="o3"`
@@ -59,7 +59,7 @@ impl CliConfigOverrides {
                     return Err(format!("Empty key in override: {s}"));
                 }
 
-                // Attempt to parse as JSON. If that fails, treat it as a raw
+                // Attempt to parse as TOML. If that fails, treat it as a raw
                 // string. This allows convenient usage such as
                 // `-c model=o3` without the quotes.
                 let value: Value = match parse_toml_value(value_str) {
@@ -151,6 +151,15 @@ mod tests {
         assert_eq!(v.as_integer(), Some(42));
     }
 
+    #[test]
+    fn parses_bool() {
+        let true_literal = parse_toml_value("true").expect("parse");
+        assert_eq!(true_literal.as_bool(), Some(true));
+
+        let false_literal = parse_toml_value("false").expect("parse");
+        assert_eq!(false_literal.as_bool(), Some(false));
+    }
+
     #[test]
     fn fails_on_unquoted_string() {
         assert!(parse_toml_value("hello").is_err());

codex-rs/common/src/model_presets.rs

@@ -1,5 +1,8 @@
+use std::collections::HashMap;
+
 use codex_app_server_protocol::AuthMode;
 use codex_core::protocol_config_types::ReasoningEffort;
+use once_cell::sync::Lazy;
 
 /// A reasoning effort option that can be surfaced for a model.
 #[derive(Debug, Clone, Copy)]
@@ -10,8 +13,14 @@ pub struct ReasoningEffortPreset {
     pub description: &'static str,
 }
 
+#[derive(Debug, Clone)]
+pub struct ModelUpgrade {
+    pub id: &'static str,
+    pub reasoning_effort_mapping: Option<HashMap<ReasoningEffort, ReasoningEffort>>,
+}
+
 /// Metadata describing a Codex-supported model.
-#[derive(Debug, Clone, Copy)]
+#[derive(Debug, Clone)]
 pub struct ModelPreset {
     /// Stable identifier for the preset.
     pub id: &'static str,
@@ -27,61 +36,173 @@ pub struct ModelPreset {
     pub supported_reasoning_efforts: &'static [ReasoningEffortPreset],
     /// Whether this is the default model for new users.
     pub is_default: bool,
+    /// recommended upgrade model
+    pub upgrade: Option<ModelUpgrade>,
 }
 
-const PRESETS: &[ModelPreset] = &[
-    ModelPreset {
-        id: "gpt-5-codex",
-        model: "gpt-5-codex",
-        display_name: "gpt-5-codex",
-        description: "Optimized for coding tasks with many tools.",
-        default_reasoning_effort: ReasoningEffort::Medium,
-        supported_reasoning_efforts: &[
-            ReasoningEffortPreset {
-                effort: ReasoningEffort::Low,
-                description: "Fastest responses with limited reasoning",
-            },
-            ReasoningEffortPreset {
-                effort: ReasoningEffort::Medium,
-                description: "Dynamically adjusts reasoning based on the task",
-            },
-            ReasoningEffortPreset {
-                effort: ReasoningEffort::High,
-                description: "Maximizes reasoning depth for complex or ambiguous problems",
-            },
-        ],
-        is_default: true,
-    },
-    ModelPreset {
-        id: "gpt-5",
-        model: "gpt-5",
-        display_name: "gpt-5",
-        description: "Broad world knowledge with strong general reasoning.",
-        default_reasoning_effort: ReasoningEffort::Medium,
-        supported_reasoning_efforts: &[
-            ReasoningEffortPreset {
-                effort: ReasoningEffort::Minimal,
-                description: "Fastest responses with little reasoning",
-            },
-            ReasoningEffortPreset {
-                effort: ReasoningEffort::Low,
-                description: "Balances speed with some reasoning; useful for straightforward queries and short explanations",
-            },
-            ReasoningEffortPreset {
-                effort: ReasoningEffort::Medium,
-                description: "Provides a solid balance of reasoning depth and latency for general-purpose tasks",
-            },
-            ReasoningEffortPreset {
-                effort: ReasoningEffort::High,
-                description: "Maximizes reasoning depth for complex or ambiguous problems",
-            },
-        ],
-        is_default: false,
-    },
-];
+static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
+    vec![
+        ModelPreset {
+            id: "gpt-5.1-codex",
+            model: "gpt-5.1-codex",
+            display_name: "gpt-5.1-codex",
+            description: "Optimized for codex.",
+            default_reasoning_effort: ReasoningEffort::Medium,
+            supported_reasoning_efforts: &[
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Low,
+                    description: "Fastest responses with limited reasoning",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Medium,
+                    description: "Dynamically adjusts reasoning based on the task",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::High,
+                    description: "Maximizes reasoning depth for complex or ambiguous problems",
+                },
+            ],
+            is_default: true,
+            upgrade: None,
+        },
+        ModelPreset {
+            id: "gpt-5.1-codex-mini",
+            model: "gpt-5.1-codex-mini",
+            display_name: "gpt-5.1-codex-mini",
+            description: "Optimized for codex. Cheaper, faster, but less capable.",
+            default_reasoning_effort: ReasoningEffort::Medium,
+            supported_reasoning_efforts: &[
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Medium,
+                    description: "Dynamically adjusts reasoning based on the task",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::High,
+                    description: "Maximizes reasoning depth for complex or ambiguous problems",
+                },
+            ],
+            is_default: false,
+            upgrade: None,
+        },
+        ModelPreset {
+            id: "gpt-5.1",
+            model: "gpt-5.1",
+            display_name: "gpt-5.1",
+            description: "Broad world knowledge with strong general reasoning.",
+            default_reasoning_effort: ReasoningEffort::Medium,
+            supported_reasoning_efforts: &[
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Low,
+                    description: "Balances speed with some reasoning; useful for straightforward queries and short explanations",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Medium,
+                    description: "Provides a solid balance of reasoning depth and latency for general-purpose tasks",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::High,
+                    description: "Maximizes reasoning depth for complex or ambiguous problems",
+                },
+            ],
+            is_default: false,
+            upgrade: None,
+        },
+        // Deprecated models.
+        ModelPreset {
+            id: "gpt-5-codex",
+            model: "gpt-5-codex",
+            display_name: "gpt-5-codex",
+            description: "Optimized for codex.",
+            default_reasoning_effort: ReasoningEffort::Medium,
+            supported_reasoning_efforts: &[
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Low,
+                    description: "Fastest responses with limited reasoning",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Medium,
+                    description: "Dynamically adjusts reasoning based on the task",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::High,
+                    description: "Maximizes reasoning depth for complex or ambiguous problems",
+                },
+            ],
+            is_default: false,
+            upgrade: Some(ModelUpgrade {
+                id: "gpt-5.1-codex",
+                reasoning_effort_mapping: None,
+            }),
+        },
+        ModelPreset {
+            id: "gpt-5-codex-mini",
+            model: "gpt-5-codex-mini",
+            display_name: "gpt-5-codex-mini",
+            description: "Optimized for codex. Cheaper, faster, but less capable.",
+            default_reasoning_effort: ReasoningEffort::Medium,
+            supported_reasoning_efforts: &[
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Medium,
+                    description: "Dynamically adjusts reasoning based on the task",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::High,
+                    description: "Maximizes reasoning depth for complex or ambiguous problems",
+                },
+            ],
+            is_default: false,
+            upgrade: Some(ModelUpgrade {
+                id: "gpt-5.1-codex-mini",
+                reasoning_effort_mapping: None,
+            }),
+        },
+        ModelPreset {
+            id: "gpt-5",
+            model: "gpt-5",
+            display_name: "gpt-5",
+            description: "Broad world knowledge with strong general reasoning.",
+            default_reasoning_effort: ReasoningEffort::Medium,
+            supported_reasoning_efforts: &[
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Minimal,
+                    description: "Fastest responses with little reasoning",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Low,
+                    description: "Balances speed with some reasoning; useful for straightforward queries and short explanations",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Medium,
+                    description: "Provides a solid balance of reasoning depth and latency for general-purpose tasks",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::High,
+                    description: "Maximizes reasoning depth for complex or ambiguous problems",
+                },
+            ],
+            is_default: false,
+            upgrade: Some(ModelUpgrade {
+                id: "gpt-5.1",
+                reasoning_effort_mapping: Some(HashMap::from([(
+                    ReasoningEffort::Minimal,
+                    ReasoningEffort::Low,
+                )])),
+            }),
+        },
+    ]
+});
 
 pub fn builtin_model_presets(_auth_mode: Option<AuthMode>) -> Vec<ModelPreset> {
-    PRESETS.to_vec()
+    // leave auth mode for later use
+    PRESETS
+        .iter()
+        .filter(|preset| preset.upgrade.is_none())
+        .cloned()
+        .collect()
+}
+
+pub fn all_model_presets() -> &'static Vec<ModelPreset> {
+    &PRESETS
 }
 
 #[cfg(test)]

codex-rs/core/Cargo.toml

@@ -32,6 +32,7 @@ codex-utils-pty = { workspace = true }
 codex-utils-readiness = { workspace = true }
 codex-utils-string = { workspace = true }
 codex-utils-tokenizer = { workspace = true }
+codex-windows-sandbox = { package = "codex-windows-sandbox", path = "../windows-sandbox-rs" }
 dirs = { workspace = true }
 dunce = { workspace = true }
 env-flags = { workspace = true }
@@ -59,6 +60,7 @@ shlex = { workspace = true }
 similar = { workspace = true }
 strum_macros = { workspace = true }
 tempfile = { workspace = true }
+test-case = "3.3.1"
 test-log = { workspace = true }
 thiserror = { workspace = true }
 time = { workspace = true, features = [
@@ -80,10 +82,9 @@ toml_edit = { workspace = true }
 tracing = { workspace = true, features = ["log"] }
 tree-sitter = { workspace = true }
 tree-sitter-bash = { workspace = true }
-uuid = { workspace = true, features = ["serde", "v4"] }
+uuid = { workspace = true, features = ["serde", "v4", "v5"] }
 which = { workspace = true }
 wildmatch = { workspace = true }
-codex_windows_sandbox = { package = "codex-windows-sandbox", path = "../windows-sandbox-rs" }
 
 
 [target.'cfg(target_os = "linux")'.dependencies]
@@ -104,7 +105,9 @@ openssl-sys = { workspace = true, features = ["vendored"] }
 [dev-dependencies]
 assert_cmd = { workspace = true }
 assert_matches = { workspace = true }
+codex-arg0 = { workspace = true }
 core_test_support = { workspace = true }
+ctor = { workspace = true }
 escargot = { workspace = true }
 image = { workspace = true, features = ["jpeg", "png"] }
 maplit = { workspace = true }

codex-rs/core/gpt_5_1_prompt.md

@@ -0,0 +1,370 @@
+You are GPT-5.1 running in the Codex CLI, a terminal-based coding assistant. Codex CLI is an open source project led by OpenAI. You are expected to be precise, safe, and helpful.
+
+Your capabilities:
+
+- Receive user prompts and other context provided by the harness, such as files in the workspace.
+- Communicate with the user by streaming thinking & responses, and by making & updating plans.
+- Emit function calls to run terminal commands and apply patches. Depending on how this specific run is configured, you can request that these function calls be escalated to the user for approval before running. More on this in the "Sandbox and approvals" section.
+
+Within this context, Codex refers to the open-source agentic coding interface (not the old Codex language model built by OpenAI).
+
+# How you work
+
+## Personality
+
+Your default personality and tone is concise, direct, and friendly. You communicate efficiently, always keeping the user clearly informed about ongoing actions without unnecessary detail. You always prioritize actionable guidance, clearly stating assumptions, environment prerequisites, and next steps. Unless explicitly asked, you avoid excessively verbose explanations about your work.
+
+# AGENTS.md spec
+- Repos often contain AGENTS.md files. These files can appear anywhere within the repository.
+- These files are a way for humans to give you (the agent) instructions or tips for working within the container.
+- Some examples might be: coding conventions, info about how code is organized, or instructions for how to run or test code.
+- Instructions in AGENTS.md files:
+    - The scope of an AGENTS.md file is the entire directory tree rooted at the folder that contains it.
+    - For every file you touch in the final patch, you must obey instructions in any AGENTS.md file whose scope includes that file.
+    - Instructions about code style, structure, naming, etc. apply only to code within the AGENTS.md file's scope, unless the file states otherwise.
+    - More-deeply-nested AGENTS.md files take precedence in the case of conflicting instructions.
+    - Direct system/developer/user instructions (as part of a prompt) take precedence over AGENTS.md instructions.
+- The contents of the AGENTS.md file at the root of the repo and any directories from the CWD up to the root are included with the developer message and don't need to be re-read. When working in a subdirectory of CWD, or a directory outside the CWD, check for any AGENTS.md files that may be applicable.
+
+## Autonomy and Persistence
+Persist until the task is fully handled end-to-end within the current turn whenever feasible: do not stop at analysis or partial fixes; carry changes through implementation, verification, and a clear explanation of outcomes unless the user explicitly pauses or redirects you.
+
+Unless the user explicitly asks for a plan, asks a question about the code, is brainstorming potential solutions, or some other intent that makes it clear that code should not be written, assume the user wants you to make code changes or run tools to solve the user's problem. In these cases, it's bad to output your proposed solution in a message, you should go ahead and actually implement the change. If you encounter challenges or blockers, you should attempt to resolve them yourself.
+
+## Responsiveness
+
+### User Updates Spec
+You'll work for stretches with tool calls — it's critical to keep the user updated as you work.
+
+Frequency & Length:
+- Send short updates (1–2 sentences) whenever there is a meaningful, important insight you need to share with the user to keep them informed.
+- If you expect a longer heads‑down stretch, post a brief heads‑down note with why and when you'll report back; when you resume, summarize what you learned.
+- Only the initial plan, plan updates, and final recap can be longer, with multiple bullets and paragraphs
+
+Tone:
+- Friendly, confident, senior-engineer energy. Positive, collaborative, humble; fix mistakes quickly.
+
+Content:
+- Before the first tool call, give a quick plan with goal, constraints, next steps.
+- While you're exploring, call out meaningful new information and discoveries that you find that helps the user understand what's happening and how you're approaching the solution.
+- If you change the plan (e.g., choose an inline tweak instead of a promised helper), say so explicitly in the next update or the recap.
+
+**Examples:**
+
+- “I’ve explored the repo; now checking the API route definitions.”
+- “Next, I’ll patch the config and update the related tests.”
+- “I’m about to scaffold the CLI commands and helper functions.”
+- “Ok cool, so I’ve wrapped my head around the repo. Now digging into the API routes.”
+- “Config’s looking tidy. Next up is patching helpers to keep things in sync.”
+- “Finished poking at the DB gateway. I will now chase down error handling.”
+- “Alright, build pipeline order is interesting. Checking how it reports failures.”
+- “Spotted a clever caching util; now hunting where it gets used.”
+
+## Planning
+
+You have access to an `update_plan` tool which tracks steps and progress and renders them to the user. Using the tool helps demonstrate that you've understood the task and convey how you're approaching it. Plans can help to make complex, ambiguous, or multi-phase work clearer and more collaborative for the user. A good plan should break the task into meaningful, logically ordered steps that are easy to verify as you go.
+
+Note that plans are not for padding out simple work with filler steps or stating the obvious. The content of your plan should not involve doing anything that you aren't capable of doing (i.e. don't try to test things that you can't test). Do not use plans for simple or single-step queries that you can just do or answer immediately.
+
+Do not repeat the full contents of the plan after an `update_plan` call — the harness already displays it. Instead, summarize the change made and highlight any important context or next step.
+
+Before running a command, consider whether or not you have completed the previous step, and make sure to mark it as completed before moving on to the next step. It may be the case that you complete all steps in your plan after a single pass of implementation. If this is the case, you can simply mark all the planned steps as completed. Sometimes, you may need to change plans in the middle of a task: call `update_plan` with the updated plan and make sure to provide an `explanation` of the rationale when doing so.
+
+Maintain statuses in the tool: exactly one item in_progress at a time; mark items complete when done; post timely status transitions. Do not jump an item from pending to completed: always set it to in_progress first. Do not batch-complete multiple items after the fact. Finish with all items completed or explicitly canceled/deferred before ending the turn. Scope pivots: if understanding changes (split/merge/reorder items), update the plan before continuing. Do not let the plan go stale while coding.
+
+Use a plan when:
+
+- The task is non-trivial and will require multiple actions over a long time horizon.
+- There are logical phases or dependencies where sequencing matters.
+- The work has ambiguity that benefits from outlining high-level goals.
+- You want intermediate checkpoints for feedback and validation.
+- When the user asked you to do more than one thing in a single prompt
+- The user has asked you to use the plan tool (aka "TODOs")
+- You generate additional steps while working, and plan to do them before yielding to the user
+
+### Examples
+
+**High-quality plans**
+
+Example 1:
+
+1. Add CLI entry with file args
+2. Parse Markdown via CommonMark library
+3. Apply semantic HTML template
+4. Handle code blocks, images, links
+5. Add error handling for invalid files
+
+Example 2:
+
+1. Define CSS variables for colors
+2. Add toggle with localStorage state
+3. Refactor components to use variables
+4. Verify all views for readability
+5. Add smooth theme-change transition
+
+Example 3:
+
+1. Set up Node.js + WebSocket server
+2. Add join/leave broadcast events
+3. Implement messaging with timestamps
+4. Add usernames + mention highlighting
+5. Persist messages in lightweight DB
+6. Add typing indicators + unread count
+
+**Low-quality plans**
+
+Example 1:
+
+1. Create CLI tool
+2. Add Markdown parser
+3. Convert to HTML
+
+Example 2:
+
+1. Add dark mode toggle
+2. Save preference
+3. Make styles look good
+
+Example 3:
+
+1. Create single-file HTML game
+2. Run quick sanity check
+3. Summarize usage instructions
+
+If you need to write a plan, only write high quality plans, not low quality ones.
+
+## Task execution
+
+You are a coding agent. You must keep going until the query or task is completely resolved, before ending your turn and yielding back to the user. Persist until the task is fully handled end-to-end within the current turn whenever feasible and persevere even when function calls fail. Only terminate your turn when you are sure that the problem is solved. Autonomously resolve the query to the best of your ability, using the tools available to you, before coming back to the user. Do NOT guess or make up an answer.
+
+You MUST adhere to the following criteria when solving queries:
+
+- Working on the repo(s) in the current environment is allowed, even if they are proprietary.
+- Analyzing code for vulnerabilities is allowed.
+- Showing user code and tool call details is allowed.
+- Use the `apply_patch` tool to edit files (NEVER try `applypatch` or `apply-patch`, only `apply_patch`). This is a FREEFORM tool, so do not wrap the patch in JSON.
+
+If completing the user's task requires writing or modifying files, your code and final answer should follow these coding guidelines, though user instructions (i.e. AGENTS.md) may override these guidelines:
+
+- Fix the problem at the root cause rather than applying surface-level patches, when possible.
+- Avoid unneeded complexity in your solution.
+- Do not attempt to fix unrelated bugs or broken tests. It is not your responsibility to fix them. (You may mention them to the user in your final message though.)
+- Update documentation as necessary.
+- Keep changes consistent with the style of the existing codebase. Changes should be minimal and focused on the task.
+- Use `git log` and `git blame` to search the history of the codebase if additional context is required.
+- NEVER add copyright or license headers unless specifically requested.
+- Do not waste tokens by re-reading files after calling `apply_patch` on them. The tool call will fail if it didn't work. The same goes for making folders, deleting folders, etc.
+- Do not `git commit` your changes or create new git branches unless explicitly requested.
+- Do not add inline comments within code unless explicitly requested.
+- Do not use one-letter variable names unless explicitly requested.
+- NEVER output inline citations like "【F:README.md†L5-L14】" in your outputs. The CLI is not able to render these so they will just be broken in the UI. Instead, if you output valid filepaths, users will be able to click on them to open the files in their editor.
+
+## Codex CLI harness, sandboxing, and approvals
+
+The Codex CLI harness supports several different configurations for sandboxing and escalation approvals that the user can choose from.
+
+Filesystem sandboxing defines which files can be read or written. The options for `sandbox_mode` are:
+- **read-only**: The sandbox only permits reading files.
+- **workspace-write**: The sandbox permits reading files, and editing files in `cwd` and `writable_roots`. Editing files in other directories requires approval.
+- **danger-full-access**: No filesystem sandboxing - all commands are permitted.
+
+Network sandboxing defines whether network can be accessed without approval. Options for `network_access` are:
+- **restricted**: Requires approval
+- **enabled**: No approval needed
+
+Approvals are your mechanism to get user consent to run shell commands without the sandbox. Possible configuration options for `approval_policy` are
+- **untrusted**: The harness will escalate most commands for user approval, apart from a limited allowlist of safe "read" commands.
+- **on-failure**: The harness will allow all commands to run in the sandbox (if enabled), and failures will be escalated to the user for approval to run again without the sandbox.
+- **on-request**: Commands will be run in the sandbox by default, and you can specify in your tool call if you want to escalate a command to run without sandboxing. (Note that this mode is not always available. If it is, you'll see parameters for escalating in the tool definition.)
+- **never**: This is a non-interactive mode where you may NEVER ask the user for approval to run commands. Instead, you must always persist and work around constraints to solve the task for the user. You MUST do your utmost best to finish the task and validate your work before yielding. If this mode is paired with `danger-full-access`, take advantage of it to deliver the best outcome for the user. Further, in this mode, your default testing philosophy is overridden: Even if you don't see local patterns for testing, you may add tests and scripts to validate your work. Just remove them before yielding.
+
+When you are running with `approval_policy == on-request`, and sandboxing enabled, here are scenarios where you'll need to request approval:
+- You need to run a command that writes to a directory that requires it (e.g. running tests that write to /var)
+- You need to run a GUI app (e.g., open/xdg-open/osascript) to open browsers or files.
+- You are running sandboxed and need to run a command that requires network access (e.g. installing packages)
+- If you run a command that is important to solving the user's query, but it fails because of sandboxing, rerun the command with approval. ALWAYS proceed to use the `with_escalated_permissions` and `justification` parameters. Within this harness, prefer requesting approval via the tool over asking in natural language.
+- You are about to take a potentially destructive action such as an `rm` or `git reset` that the user did not explicitly ask for
+- (for all of these, you should weigh alternative paths that do not require approval)
+
+When `sandbox_mode` is set to read-only, you'll need to request approval for any command that isn't a read.
+
+You will be told what filesystem sandboxing, network sandboxing, and approval mode are active in a developer or user message. If you are not told about this, assume that you are running with workspace-write, network sandboxing enabled, and approval on-failure.
+
+Although they introduce friction to the user because your work is paused until the user responds, you should leverage them when necessary to accomplish important work. If the completing the task requires escalated permissions, Do not let these settings or the sandbox deter you from attempting to accomplish the user's task unless it is set to "never", in which case never ask for approvals.
+
+When requesting approval to execute a command that will require escalated privileges:
+  - Provide the `with_escalated_permissions` parameter with the boolean value true
+  - Include a short, 1 sentence explanation for why you need to enable `with_escalated_permissions` in the justification parameter
+
+## Validating your work
+
+If the codebase has tests or the ability to build or run, consider using them to verify changes once your work is complete.
+
+When testing, your philosophy should be to start as specific as possible to the code you changed so that you can catch issues efficiently, then make your way to broader tests as you build confidence. If there's no test for the code you changed, and if the adjacent patterns in the codebases show that there's a logical place for you to add a test, you may do so. However, do not add tests to codebases with no tests.
+
+Similarly, once you're confident in correctness, you can suggest or use formatting commands to ensure that your code is well formatted. If there are issues you can iterate up to 3 times to get formatting right, but if you still can't manage it's better to save the user time and present them a correct solution where you call out the formatting in your final message. If the codebase does not have a formatter configured, do not add one.
+
+For all of testing, running, building, and formatting, do not attempt to fix unrelated bugs. It is not your responsibility to fix them. (You may mention them to the user in your final message though.)
+
+Be mindful of whether to run validation commands proactively. In the absence of behavioral guidance:
+
+- When running in non-interactive approval modes like **never** or **on-failure**, you can proactively run tests, lint and do whatever you need to ensure you've completed the task. If you are unable to run tests, you must still do your utmost best to complete the task.
+- When working in interactive approval modes like **untrusted**, or **on-request**, hold off on running tests or lint commands until the user is ready for you to finalize your output, because these commands take time to run and slow down iteration. Instead suggest what you want to do next, and let the user confirm first.
+- When working on test-related tasks, such as adding tests, fixing tests, or reproducing a bug to verify behavior, you may proactively run tests regardless of approval mode. Use your judgement to decide whether this is a test-related task.
+
+## Ambition vs. precision
+
+For tasks that have no prior context (i.e. the user is starting something brand new), you should feel free to be ambitious and demonstrate creativity with your implementation.
+
+If you're operating in an existing codebase, you should make sure you do exactly what the user asks with surgical precision. Treat the surrounding codebase with respect, and don't overstep (i.e. changing filenames or variables unnecessarily). You should balance being sufficiently ambitious and proactive when completing tasks of this nature.
+
+You should use judicious initiative to decide on the right level of detail and complexity to deliver based on the user's needs. This means showing good judgment that you're capable of doing the right extras without gold-plating. This might be demonstrated by high-value, creative touches when scope of the task is vague; while being surgical and targeted when scope is tightly specified.
+
+## Sharing progress updates
+
+For especially longer tasks that you work on (i.e. requiring many tool calls, or a plan with multiple steps), you should provide progress updates back to the user at reasonable intervals. These updates should be structured as a concise sentence or two (no more than 8-10 words long) recapping progress so far in plain language: this update demonstrates your understanding of what needs to be done, progress so far (i.e. files explores, subtasks complete), and where you're going next.
+
+Before doing large chunks of work that may incur latency as experienced by the user (i.e. writing a new file), you should send a concise message to the user with an update indicating what you're about to do to ensure they know what you're spending time on. Don't start editing or writing large files before informing the user what you are doing and why.
+
+The messages you send before tool calls should describe what is immediately about to be done next in very concise language. If there was previous work done, this preamble message should also include a note about the work done so far to bring the user along.
+
+## Presenting your work and final message
+
+Your final message should read naturally, like an update from a concise teammate. For casual conversation, brainstorming tasks, or quick questions from the user, respond in a friendly, conversational tone. You should ask questions, suggest ideas, and adapt to the user’s style. If you've finished a large amount of work, when describing what you've done to the user, you should follow the final answer formatting guidelines to communicate substantive changes. You don't need to add structured formatting for one-word answers, greetings, or purely conversational exchanges.
+
+You can skip heavy formatting for single, simple actions or confirmations. In these cases, respond in plain sentences with any relevant next step or quick option. Reserve multi-section structured responses for results that need grouping or explanation.
+
+The user is working on the same computer as you, and has access to your work. As such there's no need to show the contents of files you have already written unless the user explicitly asks for them. Similarly, if you've created or modified files using `apply_patch`, there's no need to tell users to "save the file" or "copy the code into a file"—just reference the file path.
+
+If there's something that you think you could help with as a logical next step, concisely ask the user if they want you to do so. Good examples of this are running tests, committing changes, or building out the next logical component. If there’s something that you couldn't do (even with approval) but that the user might want to do (such as verifying changes by running the app), include those instructions succinctly.
+
+Brevity is very important as a default. You should be very concise (i.e. no more than 10 lines), but can relax this requirement for tasks where additional detail and comprehensiveness is important for the user's understanding.
+
+### Final answer structure and style guidelines
+
+You are producing plain text that will later be styled by the CLI. Follow these rules exactly. Formatting should make results easy to scan, but not feel mechanical. Use judgment to decide how much structure adds value.
+
+**Section Headers**
+
+- Use only when they improve clarity — they are not mandatory for every answer.
+- Choose descriptive names that fit the content
+- Keep headers short (1–3 words) and in `**Title Case**`. Always start headers with `**` and end with `**`
+- Leave no blank line before the first bullet under a header.
+- Section headers should only be used where they genuinely improve scanability; avoid fragmenting the answer.
+
+**Bullets**
+
+- Use `-` followed by a space for every bullet.
+- Merge related points when possible; avoid a bullet for every trivial detail.
+- Keep bullets to one line unless breaking for clarity is unavoidable.
+- Group into short lists (4–6 bullets) ordered by importance.
+- Use consistent keyword phrasing and formatting across sections.
+
+**Monospace**
+
+- Wrap all commands, file paths, env vars, code identifiers, and code samples in backticks (`` `...` ``).
+- Apply to inline examples and to bullet keywords if the keyword itself is a literal file/command.
+- Never mix monospace and bold markers; choose one based on whether it’s a keyword (`**`) or inline code/path (`` ` ``).
+
+**File References**
+When referencing files in your response, make sure to include the relevant start line and always follow the below rules:
+  * Use inline code to make file paths clickable.
+  * Each reference should have a stand alone path. Even if it's the same file.
+  * Accepted: absolute, workspace‑relative, a/ or b/ diff prefixes, or bare filename/suffix.
+  * Line/column (1‑based, optional): :line[:column] or #Lline[Ccolumn] (column defaults to 1).
+  * Do not use URIs like file://, vscode://, or https://.
+  * Do not provide range of lines
+  * Examples: src/app.ts, src/app.ts:42, b/server/index.js#L10, C:\repo\project\main.rs:12:5
+
+**Structure**
+
+- Place related bullets together; don’t mix unrelated concepts in the same section.
+- Order sections from general → specific → supporting info.
+- For subsections (e.g., “Binaries” under “Rust Workspace”), introduce with a bolded keyword bullet, then list items under it.
+- Match structure to complexity:
+  - Multi-part or detailed results → use clear headers and grouped bullets.
+  - Simple results → minimal headers, possibly just a short list or paragraph.
+
+**Tone**
+
+- Keep the voice collaborative and natural, like a coding partner handing off work.
+- Be concise and factual — no filler or conversational commentary and avoid unnecessary repetition
+- Use present tense and active voice (e.g., “Runs tests” not “This will run tests”).
+- Keep descriptions self-contained; don’t refer to “above” or “below”.
+- Use parallel structure in lists for consistency.
+
+**Verbosity**
+- Final answer compactness rules (enforced):
+  - Tiny/small single-file change (≤ ~10 lines): 2–5 sentences or ≤3 bullets. No headings. 0–1 short snippet (≤3 lines) only if essential.
+  - Medium change (single area or a few files): ≤6 bullets or 6–10 sentences. At most 1–2 short snippets total (≤8 lines each).
+  - Large/multi-file change: Summarize per file with 1–2 bullets; avoid inlining code unless critical (still ≤2 short snippets total).
+  - Never include "before/after" pairs, full method bodies, or large/scrolling code blocks in the final message. Prefer referencing file/symbol names instead.
+
+**Don’t**
+
+- Don’t use literal words “bold” or “monospace” in the content.
+- Don’t nest bullets or create deep hierarchies.
+- Don’t output ANSI escape codes directly — the CLI renderer applies them.
+- Don’t cram unrelated keywords into a single bullet; split for clarity.
+- Don’t let keyword lists run long — wrap or reformat for scanability.
+
+Generally, ensure your final answers adapt their shape and depth to the request. For example, answers to code explanations should have a precise, structured explanation with code references that answer the question directly. For tasks with a simple implementation, lead with the outcome and supplement only with what’s needed for clarity. Larger changes can be presented as a logical walkthrough of your approach, grouping related steps, explaining rationale where it adds value, and highlighting next actions to accelerate the user. Your answers should provide the right level of detail while being easily scannable.
+
+For casual greetings, acknowledgements, or other one-off conversational messages that are not delivering substantive information or structured results, respond naturally without section headers or bullet formatting.
+
+# Tool Guidelines
+
+## Shell commands
+
+When using the shell, you must adhere to the following guidelines:
+
+- The arguments to `shell` will be passed to execvp().
+- Always set the `workdir` param when using the shell function. Do not use `cd` unless absolutely necessary.
+- When searching for text or files, prefer using `rg` or `rg --files` respectively because `rg` is much faster than alternatives like `grep`. (If the `rg` command is not found, then use alternatives.)
+- Read files in chunks with a max chunk size of 250 lines. Do not use python scripts to attempt to output larger chunks of a file. Command line output will be truncated after 10 kilobytes or 256 lines of output, regardless of the command used.
+
+## apply_patch
+
+Use the `apply_patch` tool to edit files. Your patch language is a stripped‑down, file‑oriented diff format designed to be easy to parse and safe to apply. You can think of it as a high‑level envelope:
+
+*** Begin Patch
+[ one or more file sections ]
+*** End Patch
+
+Within that envelope, you get a sequence of file operations.
+You MUST include a header to specify the action you are taking.
+Each operation starts with one of three headers:
+
+*** Add File: <path> - create a new file. Every following line is a + line (the initial contents).
+*** Delete File: <path> - remove an existing file. Nothing follows.
+*** Update File: <path> - patch an existing file in place (optionally with a rename).
+
+Example patch:
+
+```
+*** Begin Patch
+*** Add File: hello.txt
++Hello world
+*** Update File: src/app.py
+*** Move to: src/main.py
+@@ def greet():
+-print("Hi")
++print("Hello, world!")
+*** Delete File: obsolete.txt
+*** End Patch
+```
+
+It is important to remember:
+
+- You must include a header with your intended action (Add/Delete/Update)
+- You must prefix new lines with `+` even when creating a new file
+
+## `update_plan`
+
+A tool named `update_plan` is available to you. You can use it to keep an up‑to‑date, step‑by‑step plan for the task.
+
+To create a new plan, call `update_plan` with a short list of 1‑sentence steps (no more than 5-7 words each) with a `status` for each step (`pending`, `in_progress`, or `completed`).
+
+When steps have been completed, use `update_plan` to mark each finished step as `completed` and the next step you are working on as `in_progress`. There should always be exactly one `in_progress` step until everything is done. You can mark multiple items as complete in a single `update_plan` call.
+
+If all steps are complete, ensure you call `update_plan` to mark all steps as `completed`.

codex-rs/core/gpt_5_codex_prompt.md

@@ -16,6 +16,7 @@ You are Codex, based on GPT-5. You are running as a coding agent in the Codex CL
     * If asked to make a commit or code edits and there are unrelated changes to your work or changes that you didn't make in those files, don't revert those changes.
     * If the changes are in files you've touched recently, you should read carefully and understand how you can work with the changes rather than reverting them.
     * If the changes are in unrelated files, just ignore them and don't revert them.
+- Do not amend a commit unless explicitly requested to do so.
 - While you are working, you might notice unexpected changes that you didn't make. If this happens, STOP IMMEDIATELY and ask the user how they would like to proceed.
 - **NEVER** use destructive commands like `git reset --hard` or `git checkout --` unless specifically requested or approved by the user.
 

codex-rs/core/src/auth.rs

@@ -1,12 +1,14 @@
 mod storage;
 
 use chrono::Utc;
+use reqwest::StatusCode;
 use serde::Deserialize;
 use serde::Serialize;
 #[cfg(test)]
 use serial_test::serial;
 use std::env;
 use std::fmt::Debug;
+use std::io::ErrorKind;
 use std::path::Path;
 use std::path::PathBuf;
 use std::sync::Arc;
@@ -22,10 +24,16 @@ use crate::auth::storage::AuthStorageBackend;
 use crate::auth::storage::create_auth_storage;
 use crate::config::Config;
 use crate::default_client::CodexHttpClient;
-use crate::token_data::PlanType;
+use crate::error::RefreshTokenFailedError;
+use crate::error::RefreshTokenFailedReason;
+use crate::token_data::KnownPlan as InternalKnownPlan;
+use crate::token_data::PlanType as InternalPlanType;
 use crate::token_data::TokenData;
 use crate::token_data::parse_id_token;
 use crate::util::try_parse_error_message;
+use codex_protocol::account::PlanType as AccountPlanType;
+use serde_json::Value;
+use thiserror::Error;
 
 #[derive(Debug, Clone)]
 pub struct CodexAuth {
@@ -46,18 +54,54 @@ impl PartialEq for CodexAuth {
 // TODO(pakrym): use token exp field to check for expiration instead
 const TOKEN_REFRESH_INTERVAL: i64 = 8;
 
+const REFRESH_TOKEN_EXPIRED_MESSAGE: &str = "Your access token could not be refreshed because your refresh token has expired. Please log out and sign in again.";
+const REFRESH_TOKEN_REUSED_MESSAGE: &str = "Your access token could not be refreshed because your refresh token was already used. Please log out and sign in again.";
+const REFRESH_TOKEN_INVALIDATED_MESSAGE: &str = "Your access token could not be refreshed because your refresh token was revoked. Please log out and sign in again.";
+const REFRESH_TOKEN_UNKNOWN_MESSAGE: &str =
+    "Your access token could not be refreshed. Please log out and sign in again.";
+const REFRESH_TOKEN_URL: &str = "https://auth.openai.com/oauth/token";
+pub const REFRESH_TOKEN_URL_OVERRIDE_ENV_VAR: &str = "CODEX_REFRESH_TOKEN_URL_OVERRIDE";
+
+#[derive(Debug, Error)]
+pub enum RefreshTokenError {
+    #[error("{0}")]
+    Permanent(#[from] RefreshTokenFailedError),
+    #[error(transparent)]
+    Transient(#[from] std::io::Error),
+}
+
+impl RefreshTokenError {
+    pub fn failed_reason(&self) -> Option<RefreshTokenFailedReason> {
+        match self {
+            Self::Permanent(error) => Some(error.reason),
+            Self::Transient(_) => None,
+        }
+    }
+
+    fn other_with_message(message: impl Into<String>) -> Self {
+        Self::Transient(std::io::Error::other(message.into()))
+    }
+}
+
+impl From<RefreshTokenError> for std::io::Error {
+    fn from(err: RefreshTokenError) -> Self {
+        match err {
+            RefreshTokenError::Permanent(failed) => std::io::Error::other(failed),
+            RefreshTokenError::Transient(inner) => inner,
+        }
+    }
+}
+
 impl CodexAuth {
-    pub async fn refresh_token(&self) -> Result<String, std::io::Error> {
+    pub async fn refresh_token(&self) -> Result<String, RefreshTokenError> {
         tracing::info!("Refreshing token");
 
-        let token_data = self
-            .get_current_token_data()
-            .ok_or(std::io::Error::other("Token data is not available."))?;
+        let token_data = self.get_current_token_data().ok_or_else(|| {
+            RefreshTokenError::Transient(std::io::Error::other("Token data is not available."))
+        })?;
         let token = token_data.refresh_token;
 
-        let refresh_response = try_refresh_token(token, &self.client)
-            .await
-            .map_err(std::io::Error::other)?;
+        let refresh_response = try_refresh_token(token, &self.client).await?;
 
         let updated = update_tokens(
             &self.storage,
@@ -65,7 +109,8 @@ impl CodexAuth {
             refresh_response.access_token,
             refresh_response.refresh_token,
         )
-        .await?;
+        .await
+        .map_err(RefreshTokenError::from)?;
 
         if let Ok(mut auth_lock) = self.auth_dot_json.lock() {
             *auth_lock = Some(updated.clone());
@@ -74,7 +119,7 @@ impl CodexAuth {
         let access = match updated.tokens {
             Some(t) => t.access_token,
             None => {
-                return Err(std::io::Error::other(
+                return Err(RefreshTokenError::other_with_message(
                     "Token data is not available after refresh.",
                 ));
             }
@@ -99,15 +144,21 @@ impl CodexAuth {
                 ..
             }) => {
                 if last_refresh < Utc::now() - chrono::Duration::days(TOKEN_REFRESH_INTERVAL) {
-                    let refresh_response = tokio::time::timeout(
+                    let refresh_result = tokio::time::timeout(
                         Duration::from_secs(60),
                         try_refresh_token(tokens.refresh_token.clone(), &self.client),
                     )
-                    .await
-                    .map_err(|_| {
-                        std::io::Error::other("timed out while refreshing OpenAI API key")
-                    })?
-                    .map_err(std::io::Error::other)?;
+                    .await;
+                    let refresh_response = match refresh_result {
+                        Ok(Ok(response)) => response,
+                        Ok(Err(err)) => return Err(err.into()),
+                        Err(_) => {
+                            return Err(std::io::Error::new(
+                                ErrorKind::TimedOut,
+                                "timed out while refreshing OpenAI API key",
+                            ));
+                        }
+                    };
 
                     let updated_auth_dot_json = update_tokens(
                         &self.storage,
@@ -153,7 +204,42 @@ impl CodexAuth {
         self.get_current_token_data().and_then(|t| t.id_token.email)
     }
 
-    pub(crate) fn get_plan_type(&self) -> Option<PlanType> {
+    /// Account-facing plan classification derived from the current token.
+    /// Returns a high-level `AccountPlanType` (e.g., Free/Plus/Pro/Team/…)
+    /// mapped from the ID token's internal plan value. Prefer this when you
+    /// need to make UI or product decisions based on the user's subscription.
+    pub fn account_plan_type(&self) -> Option<AccountPlanType> {
+        let map_known = |kp: &InternalKnownPlan| match kp {
+            InternalKnownPlan::Free => AccountPlanType::Free,
+            InternalKnownPlan::Plus => AccountPlanType::Plus,
+            InternalKnownPlan::Pro => AccountPlanType::Pro,
+            InternalKnownPlan::Team => AccountPlanType::Team,
+            InternalKnownPlan::Business => AccountPlanType::Business,
+            InternalKnownPlan::Enterprise => AccountPlanType::Enterprise,
+            InternalKnownPlan::Edu => AccountPlanType::Edu,
+        };
+
+        self.get_current_token_data()
+            .and_then(|t| t.id_token.chatgpt_plan_type)
+            .map(|pt| match pt {
+                InternalPlanType::Known(k) => map_known(&k),
+                InternalPlanType::Unknown(_) => AccountPlanType::Unknown,
+            })
+    }
+
+    /// Raw plan string from the ID token (including unknown/new plan types).
+    pub fn raw_plan_type(&self) -> Option<String> {
+        self.get_plan_type().map(|plan| match plan {
+            InternalPlanType::Known(k) => format!("{k:?}"),
+            InternalPlanType::Unknown(raw) => raw,
+        })
+    }
+
+    /// Raw internal plan value from the ID token.
+    /// Exposes the underlying `token_data::PlanType` without mapping it to the
+    /// public `AccountPlanType`. Use this when downstream code needs to inspect
+    /// internal/unknown plan strings exactly as issued in the token.
+    pub(crate) fn get_plan_type(&self) -> Option<InternalPlanType> {
         self.get_current_token_data()
             .and_then(|t| t.id_token.chatgpt_plan_type)
     }
@@ -257,7 +343,10 @@ pub fn save_auth(
 }
 
 /// Load CLI auth data using the configured credential store backend.
-/// Returns `None` when no credentials are stored.
+/// Returns `None` when no credentials are stored. This function is
+/// provided only for tests. Production code should not directly load
+/// from the auth.json storage. It should use the AuthManager abstraction
+/// instead.
 pub fn load_auth_dot_json(
     codex_home: &Path,
     auth_credentials_store_mode: AuthCredentialsStoreMode,
@@ -425,7 +514,7 @@ async fn update_tokens(
 async fn try_refresh_token(
     refresh_token: String,
     client: &CodexHttpClient,
-) -> std::io::Result<RefreshResponse> {
+) -> Result<RefreshResponse, RefreshTokenError> {
     let refresh_request = RefreshRequest {
         client_id: CLIENT_ID,
         grant_type: "refresh_token",
@@ -433,30 +522,93 @@ async fn try_refresh_token(
         scope: "openid profile email",
     };
 
+    let endpoint = refresh_token_endpoint();
+
     // Use shared client factory to include standard headers
     let response = client
-        .post("https://auth.openai.com/oauth/token")
+        .post(endpoint.as_str())
         .header("Content-Type", "application/json")
         .json(&refresh_request)
         .send()
         .await
-        .map_err(std::io::Error::other)?;
+        .map_err(|err| RefreshTokenError::Transient(std::io::Error::other(err)))?;
 
-    if response.status().is_success() {
+    let status = response.status();
+    if status.is_success() {
         let refresh_response = response
             .json::<RefreshResponse>()
             .await
-            .map_err(std::io::Error::other)?;
+            .map_err(|err| RefreshTokenError::Transient(std::io::Error::other(err)))?;
         Ok(refresh_response)
     } else {
-        Err(std::io::Error::other(format!(
-            "Failed to refresh token: {}: {}",
-            response.status(),
-            try_parse_error_message(&response.text().await.unwrap_or_default()),
-        )))
+        let body = response.text().await.unwrap_or_default();
+        if status == StatusCode::UNAUTHORIZED {
+            let failed = classify_refresh_token_failure(&body);
+            Err(RefreshTokenError::Permanent(failed))
+        } else {
+            let message = try_parse_error_message(&body);
+            Err(RefreshTokenError::Transient(std::io::Error::other(
+                format!("Failed to refresh token: {status}: {message}"),
+            )))
+        }
     }
 }
 
+fn classify_refresh_token_failure(body: &str) -> RefreshTokenFailedError {
+    let code = extract_refresh_token_error_code(body);
+
+    let normalized_code = code.as_deref().map(str::to_ascii_lowercase);
+    let reason = match normalized_code.as_deref() {
+        Some("refresh_token_expired") => RefreshTokenFailedReason::Expired,
+        Some("refresh_token_reused") => RefreshTokenFailedReason::Exhausted,
+        Some("refresh_token_invalidated") => RefreshTokenFailedReason::Revoked,
+        _ => RefreshTokenFailedReason::Other,
+    };
+
+    if reason == RefreshTokenFailedReason::Other {
+        tracing::warn!(
+            backend_code = normalized_code.as_deref(),
+            backend_body = body,
+            "Encountered unknown 401 response while refreshing token"
+        );
+    }
+
+    let message = match reason {
+        RefreshTokenFailedReason::Expired => REFRESH_TOKEN_EXPIRED_MESSAGE.to_string(),
+        RefreshTokenFailedReason::Exhausted => REFRESH_TOKEN_REUSED_MESSAGE.to_string(),
+        RefreshTokenFailedReason::Revoked => REFRESH_TOKEN_INVALIDATED_MESSAGE.to_string(),
+        RefreshTokenFailedReason::Other => REFRESH_TOKEN_UNKNOWN_MESSAGE.to_string(),
+    };
+
+    RefreshTokenFailedError::new(reason, message)
+}
+
+fn extract_refresh_token_error_code(body: &str) -> Option<String> {
+    if body.trim().is_empty() {
+        return None;
+    }
+
+    let Value::Object(map) = serde_json::from_str::<Value>(body).ok()? else {
+        return None;
+    };
+
+    if let Some(error_value) = map.get("error") {
+        match error_value {
+            Value::Object(obj) => {
+                if let Some(code) = obj.get("code").and_then(Value::as_str) {
+                    return Some(code.to_string());
+                }
+            }
+            Value::String(code) => {
+                return Some(code.to_string());
+            }
+            _ => {}
+        }
+    }
+
+    map.get("code").and_then(Value::as_str).map(str::to_string)
+}
+
 #[derive(Serialize)]
 struct RefreshRequest {
     client_id: &'static str,
@@ -475,6 +627,11 @@ struct RefreshResponse {
 // Shared constant for token refresh (client id used for oauth token refresh flow)
 pub const CLIENT_ID: &str = "app_EMoamEEZ73f0CkXaXp7hrann";
 
+fn refresh_token_endpoint() -> String {
+    std::env::var(REFRESH_TOKEN_URL_OVERRIDE_ENV_VAR)
+        .unwrap_or_else(|_| REFRESH_TOKEN_URL.to_string())
+}
+
 use std::sync::RwLock;
 
 /// Internal cached auth state.
@@ -492,8 +649,9 @@ mod tests {
     use crate::config::ConfigOverrides;
     use crate::config::ConfigToml;
     use crate::token_data::IdTokenInfo;
-    use crate::token_data::KnownPlan;
-    use crate::token_data::PlanType;
+    use crate::token_data::KnownPlan as InternalKnownPlan;
+    use crate::token_data::PlanType as InternalPlanType;
+    use codex_protocol::account::PlanType as AccountPlanType;
 
     use base64::Engine;
     use codex_protocol::config_types::ForcedLoginMethod;
@@ -610,7 +768,7 @@ mod tests {
                 tokens: Some(TokenData {
                     id_token: IdTokenInfo {
                         email: Some("user@example.com".to_string()),
-                        chatgpt_plan_type: Some(PlanType::Known(KnownPlan::Pro)),
+                        chatgpt_plan_type: Some(InternalPlanType::Known(InternalKnownPlan::Pro)),
                         chatgpt_account_id: None,
                         raw_jwt: fake_jwt,
                     },
@@ -864,6 +1022,54 @@ mod tests {
                 .contains("ChatGPT login is required, but an API key is currently being used.")
         );
     }
+
+    #[test]
+    fn plan_type_maps_known_plan() {
+        let codex_home = tempdir().unwrap();
+        let _jwt = write_auth_file(
+            AuthFileParams {
+                openai_api_key: None,
+                chatgpt_plan_type: "pro".to_string(),
+                chatgpt_account_id: None,
+            },
+            codex_home.path(),
+        )
+        .expect("failed to write auth file");
+
+        let auth = super::load_auth(codex_home.path(), false, AuthCredentialsStoreMode::File)
+            .expect("load auth")
+            .expect("auth available");
+
+        pretty_assertions::assert_eq!(auth.account_plan_type(), Some(AccountPlanType::Pro));
+        pretty_assertions::assert_eq!(
+            auth.get_plan_type(),
+            Some(InternalPlanType::Known(InternalKnownPlan::Pro))
+        );
+    }
+
+    #[test]
+    fn plan_type_maps_unknown_to_unknown() {
+        let codex_home = tempdir().unwrap();
+        let _jwt = write_auth_file(
+            AuthFileParams {
+                openai_api_key: None,
+                chatgpt_plan_type: "mystery-tier".to_string(),
+                chatgpt_account_id: None,
+            },
+            codex_home.path(),
+        )
+        .expect("failed to write auth file");
+
+        let auth = super::load_auth(codex_home.path(), false, AuthCredentialsStoreMode::File)
+            .expect("load auth")
+            .expect("auth available");
+
+        pretty_assertions::assert_eq!(auth.account_plan_type(), Some(AccountPlanType::Unknown));
+        pretty_assertions::assert_eq!(
+            auth.get_plan_type(),
+            Some(InternalPlanType::Unknown("mystery-tier".to_string()))
+        );
+    }
 }
 
 /// Central manager providing a single source of truth for auth.json derived
@@ -965,7 +1171,9 @@ impl AuthManager {
 
     /// Attempt to refresh the current auth token (if any). On success, reload
     /// the auth state from disk so other components observe refreshed token.
-    pub async fn refresh_token(&self) -> std::io::Result<Option<String>> {
+    /// If the token refresh fails in a permanent (non‑transient) way, logs out
+    /// to clear invalid auth state.
+    pub async fn refresh_token(&self) -> Result<Option<String>, RefreshTokenError> {
         let auth = match self.auth() {
             Some(a) => a,
             None => return Ok(None),

codex-rs/core/src/bash.rs

@@ -1,8 +1,13 @@
+use std::path::PathBuf;
+
 use tree_sitter::Node;
 use tree_sitter::Parser;
 use tree_sitter::Tree;
 use tree_sitter_bash::LANGUAGE as BASH;
 
+use crate::shell::ShellType;
+use crate::shell::detect_shell_type;
+
 /// Parse the provided bash source using tree-sitter-bash, returning a Tree on
 /// success or None if parsing failed.
 pub fn try_parse_shell(shell_lc_arg: &str) -> Option<Tree> {
@@ -88,17 +93,26 @@ pub fn try_parse_word_only_commands_sequence(tree: &Tree, src: &str) -> Option<V
     Some(commands)
 }
 
+pub fn extract_bash_command(command: &[String]) -> Option<(&str, &str)> {
+    let [shell, flag, script] = command else {
+        return None;
+    };
+    if !matches!(flag.as_str(), "-lc" | "-c")
+        || !matches!(
+            detect_shell_type(&PathBuf::from(shell)),
+            Some(ShellType::Zsh) | Some(ShellType::Bash)
+        )
+    {
+        return None;
+    }
+    Some((shell, script))
+}
+
 /// Returns the sequence of plain commands within a `bash -lc "..."` or
 /// `zsh -lc "..."` invocation when the script only contains word-only commands
 /// joined by safe operators.
 pub fn parse_shell_lc_plain_commands(command: &[String]) -> Option<Vec<Vec<String>>> {
-    let [shell, flag, script] = command else {
-        return None;
-    };
-
-    if flag != "-lc" || !(shell == "bash" || shell == "zsh") {
-        return None;
-    }
+    let (_, script) = extract_bash_command(command)?;
 
     let tree = try_parse_shell(script)?;
     try_parse_word_only_commands_sequence(&tree, script)

codex-rs/core/src/chat_completions.rs

@@ -338,7 +338,7 @@ pub(crate) async fn stream_chat_completions(
     debug!(
         "POST to {}: {}",
         provider.get_full_url(&None),
-        serde_json::to_string_pretty(&payload).unwrap_or_default()
+        payload.to_string()
     );
 
     let mut attempt = 0;
@@ -477,10 +477,14 @@ async fn append_reasoning_text(
         ..
     }) = reasoning_item
     {
+        let content_index = content.len() as i64;
         content.push(ReasoningItemContent::ReasoningText { text: text.clone() });
 
         let _ = tx_event
-            .send(Ok(ResponseEvent::ReasoningContentDelta(text.clone())))
+            .send(Ok(ResponseEvent::ReasoningContentDelta {
+                delta: text.clone(),
+                content_index,
+            }))
             .await;
     }
 }
@@ -898,20 +902,26 @@ where
                         continue;
                     }
                 }
-                Poll::Ready(Some(Ok(ResponseEvent::ReasoningContentDelta(delta)))) => {
+                Poll::Ready(Some(Ok(ResponseEvent::ReasoningContentDelta {
+                    delta,
+                    content_index,
+                }))) => {
                     // Always accumulate reasoning deltas so we can emit a final Reasoning item at Completed.
                     this.cumulative_reasoning.push_str(&delta);
                     if matches!(this.mode, AggregateMode::Streaming) {
                         // In streaming mode, also forward the delta immediately.
-                        return Poll::Ready(Some(Ok(ResponseEvent::ReasoningContentDelta(delta))));
+                        return Poll::Ready(Some(Ok(ResponseEvent::ReasoningContentDelta {
+                            delta,
+                            content_index,
+                        })));
                     } else {
                         continue;
                     }
                 }
-                Poll::Ready(Some(Ok(ResponseEvent::ReasoningSummaryDelta(_)))) => {
+                Poll::Ready(Some(Ok(ResponseEvent::ReasoningSummaryDelta { .. }))) => {
                     continue;
                 }
-                Poll::Ready(Some(Ok(ResponseEvent::ReasoningSummaryPartAdded))) => {
+                Poll::Ready(Some(Ok(ResponseEvent::ReasoningSummaryPartAdded { .. }))) => {
                     continue;
                 }
                 Poll::Ready(Some(Ok(ResponseEvent::OutputItemAdded(item)))) => {

codex-rs/core/src/client.rs

@@ -31,13 +31,14 @@ use tracing::warn;
 
 use crate::AuthManager;
 use crate::auth::CodexAuth;
+use crate::auth::RefreshTokenError;
 use crate::chat_completions::AggregateStreamExt;
 use crate::chat_completions::stream_chat_completions;
 use crate::client_common::Prompt;
+use crate::client_common::Reasoning;
 use crate::client_common::ResponseEvent;
 use crate::client_common::ResponseStream;
 use crate::client_common::ResponsesApiRequest;
-use crate::client_common::create_reasoning_param_for_request;
 use crate::client_common::create_text_param_for_request;
 use crate::config::Config;
 use crate::default_client::CodexHttpClient;
@@ -198,12 +199,18 @@ impl ModelClient {
         let auth_manager = self.auth_manager.clone();
 
         let full_instructions = prompt.get_full_instructions(&self.config.model_family);
-        let tools_json = create_tools_json_for_responses_api(&prompt.tools)?;
-        let reasoning = create_reasoning_param_for_request(
-            &self.config.model_family,
-            self.effort,
-            self.summary,
-        );
+        let tools_json: Vec<Value> = create_tools_json_for_responses_api(&prompt.tools)?;
+
+        let reasoning = if self.config.model_family.supports_reasoning_summaries {
+            Some(Reasoning {
+                effort: self
+                    .effort
+                    .or(self.config.model_family.default_reasoning_effort),
+                summary: Some(self.summary),
+            })
+        } else {
+            None
+        };
 
         let include: Vec<String> = if reasoning.is_some() {
             vec!["reasoning.encrypted_content".to_string()]
@@ -214,7 +221,9 @@ impl ModelClient {
         let input_with_instructions = prompt.get_formatted_input();
 
         let verbosity = if self.config.model_family.support_verbosity {
-            self.config.model_verbosity
+            self.config
+                .model_verbosity
+                .or(self.config.model_family.default_verbosity)
         } else {
             if self.config.model_verbosity.is_some() {
                 warn!(
@@ -293,10 +302,9 @@ impl ModelClient {
         let auth = auth_manager.as_ref().and_then(|m| m.auth());
 
         trace!(
-            "POST to {}: {:?}",
+            "POST to {}: {}",
             self.provider.get_full_url(&auth),
-            serde_json::to_string(payload_json)
-                .unwrap_or("<unable to serialize payload>".to_string())
+            payload_json.to_string()
         );
 
         let mut req_builder = self
@@ -389,12 +397,17 @@ impl ModelClient {
                     && let Some(manager) = auth_manager.as_ref()
                     && let Some(auth) = auth.as_ref()
                     && auth.mode == AuthMode::ChatGPT
+                    && let Err(err) = manager.refresh_token().await
                 {
-                    manager.refresh_token().await.map_err(|err| {
-                        StreamAttemptError::Fatal(CodexErr::Fatal(format!(
-                            "Failed to refresh ChatGPT credentials: {err}"
-                        )))
-                    })?;
+                    let stream_error = match err {
+                        RefreshTokenError::Permanent(failed) => {
+                            StreamAttemptError::Fatal(CodexErr::RefreshTokenFailed(failed))
+                        }
+                        RefreshTokenError::Transient(other) => {
+                            StreamAttemptError::RetryableTransportError(CodexErr::Io(other))
+                        }
+                    };
+                    return Err(stream_error);
                 }
 
                 // The OpenAI Responses endpoint returns structured JSON bodies even for 4xx/5xx
@@ -441,6 +454,8 @@ impl ModelClient {
                             return Err(StreamAttemptError::Fatal(codex_err));
                         } else if error.r#type.as_deref() == Some("usage_not_included") {
                             return Err(StreamAttemptError::Fatal(CodexErr::UsageNotIncluded));
+                        } else if is_quota_exceeded_error(&error) {
+                            return Err(StreamAttemptError::Fatal(CodexErr::QuotaExceeded));
                         }
                     }
                 }
@@ -545,6 +560,8 @@ struct SseEvent {
     response: Option<Value>,
     item: Option<Value>,
     delta: Option<String>,
+    summary_index: Option<i64>,
+    content_index: Option<i64>,
 }
 
 #[derive(Debug, Deserialize)]
@@ -804,16 +821,22 @@ async fn process_sse<S>(
                 }
             }
             "response.reasoning_summary_text.delta" => {
-                if let Some(delta) = event.delta {
-                    let event = ResponseEvent::ReasoningSummaryDelta(delta);
+                if let (Some(delta), Some(summary_index)) = (event.delta, event.summary_index) {
+                    let event = ResponseEvent::ReasoningSummaryDelta {
+                        delta,
+                        summary_index,
+                    };
                     if tx_event.send(Ok(event)).await.is_err() {
                         return;
                     }
                 }
             }
             "response.reasoning_text.delta" => {
-                if let Some(delta) = event.delta {
-                    let event = ResponseEvent::ReasoningContentDelta(delta);
+                if let (Some(delta), Some(content_index)) = (event.delta, event.content_index) {
+                    let event = ResponseEvent::ReasoningContentDelta {
+                        delta,
+                        content_index,
+                    };
                     if tx_event.send(Ok(event)).await.is_err() {
                         return;
                     }
@@ -838,6 +861,8 @@ async fn process_sse<S>(
                             Ok(error) => {
                                 if is_context_window_error(&error) {
                                     response_error = Some(CodexErr::ContextWindowExceeded);
+                                } else if is_quota_exceeded_error(&error) {
+                                    response_error = Some(CodexErr::QuotaExceeded);
                                 } else {
                                     let delay = try_parse_retry_after(&error);
                                     let message = error.message.clone().unwrap_or_default();
@@ -888,10 +913,12 @@ async fn process_sse<S>(
                 }
             }
             "response.reasoning_summary_part.added" => {
-                // Boundary between reasoning summary sections (e.g., titles).
-                let event = ResponseEvent::ReasoningSummaryPartAdded;
-                if tx_event.send(Ok(event)).await.is_err() {
-                    return;
+                if let Some(summary_index) = event.summary_index {
+                    // Boundary between reasoning summary sections (e.g., titles).
+                    let event = ResponseEvent::ReasoningSummaryPartAdded { summary_index };
+                    if tx_event.send(Ok(event)).await.is_err() {
+                        return;
+                    }
                 }
             }
             "response.reasoning_summary_text.done" => {}
@@ -931,8 +958,10 @@ async fn stream_from_fixture(
 fn rate_limit_regex() -> &'static Regex {
     static RE: OnceLock<Regex> = OnceLock::new();
 
+    // Match both OpenAI-style messages like "Please try again in 1.898s"
+    // and Azure OpenAI-style messages like "Try again in 35 seconds".
     #[expect(clippy::unwrap_used)]
-    RE.get_or_init(|| Regex::new(r"Please try again in (\d+(?:\.\d+)?)(s|ms)").unwrap())
+    RE.get_or_init(|| Regex::new(r"(?i)try again in\s*(\d+(?:\.\d+)?)\s*(s|ms|seconds?)").unwrap())
 }
 
 fn try_parse_retry_after(err: &Error) -> Option<Duration> {
@@ -940,7 +969,8 @@ fn try_parse_retry_after(err: &Error) -> Option<Duration> {
         return None;
     }
 
-    // parse the Please try again in 1.898s format using regex
+    // parse retry hints like "try again in 1.898s" or
+    // "Try again in 35 seconds" using regex
     let re = rate_limit_regex();
     if let Some(message) = &err.message
         && let Some(captures) = re.captures(message)
@@ -950,9 +980,9 @@ fn try_parse_retry_after(err: &Error) -> Option<Duration> {
 
         if let (Some(value), Some(unit)) = (seconds, unit) {
             let value = value.as_str().parse::<f64>().ok()?;
-            let unit = unit.as_str();
+            let unit = unit.as_str().to_ascii_lowercase();
 
-            if unit == "s" {
+            if unit == "s" || unit.starts_with("second") {
                 return Some(Duration::from_secs_f64(value));
             } else if unit == "ms" {
                 return Some(Duration::from_millis(value as u64));
@@ -966,6 +996,10 @@ fn is_context_window_error(error: &Error) -> bool {
     error.code.as_deref() == Some("context_length_exceeded")
 }
 
+fn is_quota_exceeded_error(error: &Error) -> bool {
+    error.code.as_deref() == Some("insufficient_quota")
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -1298,6 +1332,41 @@ mod tests {
         }
     }
 
+    #[tokio::test]
+    async fn quota_exceeded_error_is_fatal() {
+        let raw_error = r#"{"type":"response.failed","sequence_number":3,"response":{"id":"resp_fatal_quota","object":"response","created_at":1759771626,"status":"failed","background":false,"error":{"code":"insufficient_quota","message":"You exceeded your current quota, please check your plan and billing details. For more information on this error, read the docs: https://platform.openai.com/docs/guides/error-codes/api-errors."},"incomplete_details":null}}"#;
+
+        let sse1 = format!("event: response.failed\ndata: {raw_error}\n\n");
+        let provider = ModelProviderInfo {
+            name: "test".to_string(),
+            base_url: Some("https://test.com".to_string()),
+            env_key: Some("TEST_API_KEY".to_string()),
+            env_key_instructions: None,
+            experimental_bearer_token: None,
+            wire_api: WireApi::Responses,
+            query_params: None,
+            http_headers: None,
+            env_http_headers: None,
+            request_max_retries: Some(0),
+            stream_max_retries: Some(0),
+            stream_idle_timeout_ms: Some(1000),
+            requires_openai_auth: false,
+        };
+
+        let otel_event_manager = otel_event_manager();
+
+        let events = collect_events(&[sse1.as_bytes()], provider, otel_event_manager).await;
+
+        assert_eq!(events.len(), 1);
+
+        match &events[0] {
+            Err(err @ CodexErr::QuotaExceeded) => {
+                assert_eq!(err.to_string(), CodexErr::QuotaExceeded.to_string());
+            }
+            other => panic!("unexpected quota exceeded event: {other:?}"),
+        }
+    }
+
     // ────────────────────────────
     // Table-driven test from `main`
     // ────────────────────────────
@@ -1427,6 +1496,19 @@ mod tests {
         assert_eq!(delay, Some(Duration::from_secs_f64(1.898)));
     }
 
+    #[test]
+    fn test_try_parse_retry_after_azure() {
+        let err = Error {
+            r#type: None,
+            message: Some("Rate limit exceeded. Try again in 35 seconds.".to_string()),
+            code: Some("rate_limit_exceeded".to_string()),
+            plan_type: None,
+            resets_at: None,
+        };
+        let delay = try_parse_retry_after(&err);
+        assert_eq!(delay, Some(Duration::from_secs(35)));
+    }
+
     #[test]
     fn error_response_deserializes_schema_known_plan_type_and_serializes_back() {
         use crate::token_data::KnownPlan;

codex-rs/core/src/client_common.rs

@@ -203,9 +203,17 @@ pub enum ResponseEvent {
         token_usage: Option<TokenUsage>,
     },
     OutputTextDelta(String),
-    ReasoningSummaryDelta(String),
-    ReasoningContentDelta(String),
-    ReasoningSummaryPartAdded,
+    ReasoningSummaryDelta {
+        delta: String,
+        summary_index: i64,
+    },
+    ReasoningContentDelta {
+        delta: String,
+        content_index: i64,
+    },
+    ReasoningSummaryPartAdded {
+        summary_index: i64,
+    },
     RateLimits(RateLimitSnapshot),
 }
 
@@ -342,21 +350,6 @@ pub(crate) mod tools {
     }
 }
 
-pub(crate) fn create_reasoning_param_for_request(
-    model_family: &ModelFamily,
-    effort: Option<ReasoningEffortConfig>,
-    summary: ReasoningSummaryConfig,
-) -> Option<Reasoning> {
-    if !model_family.supports_reasoning_summaries {
-        return None;
-    }
-
-    Some(Reasoning {
-        effort,
-        summary: Some(summary),
-    })
-}
-
 pub(crate) fn create_text_param_for_request(
     verbosity: Option<VerbosityConfig>,
     output_schema: &Option<Value>,
@@ -421,6 +414,10 @@ mod tests {
                 slug: "gpt-5",
                 expects_apply_patch_instructions: true,
             },
+            InstructionsTestCase {
+                slug: "gpt-5.1",
+                expects_apply_patch_instructions: false,
+            },
             InstructionsTestCase {
                 slug: "codex-mini-latest",
                 expects_apply_patch_instructions: true,
@@ -433,6 +430,10 @@ mod tests {
                 slug: "gpt-5-codex",
                 expects_apply_patch_instructions: false,
             },
+            InstructionsTestCase {
+                slug: "gpt-5.1-codex",
+                expects_apply_patch_instructions: false,
+            },
         ];
         for test_case in test_cases {
             let model_family = find_family_for_model(test_case.slug).expect("known model slug");

codex-rs/core/src/codex.rs

@@ -6,6 +6,7 @@ use std::sync::atomic::AtomicU64;
 
 use crate::AuthManager;
 use crate::client_common::REVIEW_PROMPT;
+use crate::compact;
 use crate::features::Feature;
 use crate::function_tool::FunctionCallError;
 use crate::mcp::auth::McpAuthStatusEntry;
@@ -58,7 +59,7 @@ use crate::client_common::ResponseEvent;
 use crate::config::Config;
 use crate::config::types::McpServerTransportConfig;
 use crate::config::types::ShellEnvironmentPolicy;
-use crate::conversation_history::ConversationHistory;
+use crate::context_manager::ContextManager;
 use crate::environment_context::EnvironmentContext;
 use crate::error::CodexErr;
 use crate::error::Result as CodexResult;
@@ -66,6 +67,8 @@ use crate::error::Result as CodexResult;
 use crate::exec::StreamOutput;
 // Removed: legacy executor wiring replaced by ToolOrchestrator flows.
 // legacy normalize_exec_result no longer used after orchestrator migration
+use crate::compact::build_compacted_history;
+use crate::compact::collect_user_messages;
 use crate::mcp::auth::compute_auth_statuses;
 use crate::mcp_connection_manager::McpConnectionManager;
 use crate::model_family::find_family_for_model;
@@ -93,7 +96,9 @@ use crate::protocol::StreamErrorEvent;
 use crate::protocol::Submission;
 use crate::protocol::TokenCountEvent;
 use crate::protocol::TokenUsage;
+use crate::protocol::TokenUsageInfo;
 use crate::protocol::TurnDiffEvent;
+use crate::protocol::WarningEvent;
 use crate::rollout::RolloutRecorder;
 use crate::rollout::RolloutRecorderParams;
 use crate::shell;
@@ -128,10 +133,7 @@ use codex_protocol::protocol::InitialHistory;
 use codex_protocol::user_input::UserInput;
 use codex_utils_readiness::Readiness;
 use codex_utils_readiness::ReadinessFlag;
-
-pub mod compact;
-use self::compact::build_compacted_history;
-use self::compact::collect_user_messages;
+use codex_utils_tokenizer::warm_model_cache;
 
 /// The high-level interface to the Codex system.
 /// It operates as a queue pair where you send submissions and receive events.
@@ -553,7 +555,7 @@ impl Session {
                 None
             } else {
                 Some(format!(
-                    "You can either enable it using the CLI with `--enable {canonical}` or through the config.toml file with `[features].{canonical}`"
+                    "Enable it with `--enable {canonical}` or `[features].{canonical}` in config.toml. See https://github.com/openai/codex/blob/main/docs/config.md#feature-flags for details."
                 ))
             };
             post_session_configured_events.push(Event {
@@ -589,6 +591,9 @@ impl Session {
         // Create the mutable state for the Session.
         let state = SessionState::new(session_configuration.clone());
 
+        // Warm the tokenizer cache for the session model without blocking startup.
+        warm_model_cache(&session_configuration.model);
+
         let services = SessionServices {
             mcp_connection_manager,
             unified_exec_manager: UnifiedExecSessionManager::default(),
@@ -675,6 +680,34 @@ impl Session {
                 let rollout_items = conversation_history.get_rollout_items();
                 let persist = matches!(conversation_history, InitialHistory::Forked(_));
 
+                // If resuming, warn when the last recorded model differs from the current one.
+                if let InitialHistory::Resumed(_) = conversation_history
+                    && let Some(prev) = rollout_items.iter().rev().find_map(|it| {
+                        if let RolloutItem::TurnContext(ctx) = it {
+                            Some(ctx.model.as_str())
+                        } else {
+                            None
+                        }
+                    })
+                {
+                    let curr = turn_context.client.get_model();
+                    if prev != curr {
+                        warn!(
+                            "resuming session with different model: previous={prev}, current={curr}"
+                        );
+                        self.send_event(
+                                &turn_context,
+                                EventMsg::Warning(WarningEvent {
+                                    message: format!(
+                                        "This session was recorded with model `{prev}` but is resuming with `{curr}`. \
+                         Consider switching back to `{prev}` as it may affect Codex performance."
+                                    ),
+                                }),
+                            )
+                                .await;
+                    }
+                }
+
                 // Always add response items to conversation history
                 let reconstructed_history =
                     self.reconstruct_history_from_rollout(&turn_context, &rollout_items);
@@ -945,7 +978,7 @@ impl Session {
         turn_context: &TurnContext,
         rollout_items: &[RolloutItem],
     ) -> Vec<ResponseItem> {
-        let mut history = ConversationHistory::new();
+        let mut history = ContextManager::new();
         for item in rollout_items {
             match item {
                 RolloutItem::ResponseItem(response_item) => {
@@ -968,7 +1001,7 @@ impl Session {
     }
 
     /// Append ResponseItems to the in-memory conversation history only.
-    async fn record_into_history(&self, items: &[ResponseItem]) {
+    pub(crate) async fn record_into_history(&self, items: &[ResponseItem]) {
         let mut state = self.state.lock().await;
         state.record_items(items.iter());
     }
@@ -1020,7 +1053,7 @@ impl Session {
         items
     }
 
-    async fn persist_rollout_items(&self, items: &[RolloutItem]) {
+    pub(crate) async fn persist_rollout_items(&self, items: &[RolloutItem]) {
         let recorder = {
             let guard = self.services.rollout.lock().await;
             guard.clone()
@@ -1032,12 +1065,12 @@ impl Session {
         }
     }
 
-    pub(crate) async fn clone_history(&self) -> ConversationHistory {
+    pub(crate) async fn clone_history(&self) -> ContextManager {
         let state = self.state.lock().await;
         state.clone_history()
     }
 
-    async fn update_token_usage_info(
+    pub(crate) async fn update_token_usage_info(
         &self,
         turn_context: &TurnContext,
         token_usage: Option<&TokenUsage>,
@@ -1054,7 +1087,37 @@ impl Session {
         self.send_token_count_event(turn_context).await;
     }
 
-    async fn update_rate_limits(
+    pub(crate) async fn override_last_token_usage_estimate(
+        &self,
+        turn_context: &TurnContext,
+        estimated_total_tokens: i64,
+    ) {
+        {
+            let mut state = self.state.lock().await;
+            let mut info = state.token_info().unwrap_or(TokenUsageInfo {
+                total_token_usage: TokenUsage::default(),
+                last_token_usage: TokenUsage::default(),
+                model_context_window: None,
+            });
+
+            info.last_token_usage = TokenUsage {
+                input_tokens: 0,
+                cached_input_tokens: 0,
+                output_tokens: 0,
+                reasoning_output_tokens: 0,
+                total_tokens: estimated_total_tokens.max(0),
+            };
+
+            if info.model_context_window.is_none() {
+                info.model_context_window = turn_context.client.get_model_context_window();
+            }
+
+            state.set_token_info(Some(info));
+        }
+        self.send_token_count_event(turn_context).await;
+    }
+
+    pub(crate) async fn update_rate_limits(
         &self,
         turn_context: &TurnContext,
         new_rate_limits: RateLimitSnapshot,
@@ -1075,7 +1138,7 @@ impl Session {
         self.send_event(turn_context, event).await;
     }
 
-    async fn set_total_tokens_full(&self, turn_context: &TurnContext) {
+    pub(crate) async fn set_total_tokens_full(&self, turn_context: &TurnContext) {
         let context_window = turn_context.client.get_model_context_window();
         if let Some(context_window) = context_window {
             {
@@ -1118,7 +1181,11 @@ impl Session {
         self.send_event(turn_context, event).await;
     }
 
-    async fn notify_stream_error(&self, turn_context: &TurnContext, message: impl Into<String>) {
+    pub(crate) async fn notify_stream_error(
+        &self,
+        turn_context: &TurnContext,
+        message: impl Into<String>,
+    ) {
         let event = EventMsg::StreamError(StreamErrorEvent {
             message: message.into(),
         });
@@ -1643,8 +1710,7 @@ async fn spawn_review_thread(
     let mut review_features = config.features.clone();
     review_features
         .disable(crate::features::Feature::WebSearchRequest)
-        .disable(crate::features::Feature::ViewImageTool)
-        .disable(crate::features::Feature::StreamableShell);
+        .disable(crate::features::Feature::ViewImageTool);
     let tools_config = ToolsConfig::new(&ToolsConfigParams {
         model_family: &review_model_family,
         features: &review_features,
@@ -1773,19 +1839,14 @@ pub(crate) async fn run_task(
             sess.clone_history().await.get_history_for_prompt()
         };
 
-        let turn_input_messages: Vec<String> = turn_input
+        let turn_input_messages = turn_input
             .iter()
-            .filter_map(|item| match item {
-                ResponseItem::Message { content, .. } => Some(content),
+            .filter_map(|item| match parse_turn_item(item) {
+                Some(TurnItem::UserMessage(user_message)) => Some(user_message),
                 _ => None,
             })
-            .flat_map(|content| {
-                content.iter().filter_map(|item| match item {
-                    ContentItem::OutputText { text } => Some(text.clone()),
-                    _ => None,
-                })
-            })
-            .collect();
+            .map(|user_message| user_message.message())
+            .collect::<Vec<String>>();
         match run_turn(
             Arc::clone(&sess),
             Arc::clone(&turn_context),
@@ -1933,6 +1994,8 @@ async fn run_turn(
                 return Err(CodexErr::UsageLimitReached(e));
             }
             Err(CodexErr::UsageNotIncluded) => return Err(CodexErr::UsageNotIncluded),
+            Err(e @ CodexErr::QuotaExceeded) => return Err(e),
+            Err(e @ CodexErr::RefreshTokenFailed(_)) => return Err(e),
             Err(e) => {
                 // Use the configured provider-specific stream retry budget.
                 let max_retries = turn_context.client.get_provider().stream_max_retries();
@@ -1951,7 +2014,7 @@ async fn run_turn(
                     // at a seemingly frozen screen.
                     sess.notify_stream_error(
                         &turn_context,
-                        format!("Re-connecting... {retries}/{max_retries}"),
+                        format!("Reconnecting... {retries}/{max_retries}"),
                     )
                     .await;
 
@@ -2173,13 +2236,17 @@ async fn try_run_turn(
                     error_or_panic("ReasoningSummaryDelta without active item".to_string());
                 }
             }
-            ResponseEvent::ReasoningSummaryDelta(delta) => {
+            ResponseEvent::ReasoningSummaryDelta {
+                delta,
+                summary_index,
+            } => {
                 if let Some(active) = active_item.as_ref() {
                     let event = ReasoningContentDeltaEvent {
                         thread_id: sess.conversation_id.to_string(),
                         turn_id: turn_context.sub_id.clone(),
                         item_id: active.id(),
-                        delta: delta.clone(),
+                        delta,
+                        summary_index,
                     };
                     sess.send_event(&turn_context, EventMsg::ReasoningContentDelta(event))
                         .await;
@@ -2187,18 +2254,29 @@ async fn try_run_turn(
                     error_or_panic("ReasoningSummaryDelta without active item".to_string());
                 }
             }
-            ResponseEvent::ReasoningSummaryPartAdded => {
-                let event =
-                    EventMsg::AgentReasoningSectionBreak(AgentReasoningSectionBreakEvent {});
-                sess.send_event(&turn_context, event).await;
+            ResponseEvent::ReasoningSummaryPartAdded { summary_index } => {
+                if let Some(active) = active_item.as_ref() {
+                    let event =
+                        EventMsg::AgentReasoningSectionBreak(AgentReasoningSectionBreakEvent {
+                            item_id: active.id(),
+                            summary_index,
+                        });
+                    sess.send_event(&turn_context, event).await;
+                } else {
+                    error_or_panic("ReasoningSummaryPartAdded without active item".to_string());
+                }
             }
-            ResponseEvent::ReasoningContentDelta(delta) => {
+            ResponseEvent::ReasoningContentDelta {
+                delta,
+                content_index,
+            } => {
                 if let Some(active) = active_item.as_ref() {
                     let event = ReasoningRawContentDeltaEvent {
                         thread_id: sess.conversation_id.to_string(),
                         turn_id: turn_context.sub_id.clone(),
                         item_id: active.id(),
-                        delta: delta.clone(),
+                        delta,
+                        content_index,
                     };
                     sess.send_event(&turn_context, EventMsg::ReasoningRawContentDelta(event))
                         .await;
@@ -2324,6 +2402,7 @@ mod tests {
     use crate::tools::context::ToolOutput;
     use crate::tools::context::ToolPayload;
     use crate::tools::handlers::ShellHandler;
+    use crate::tools::handlers::UnifiedExecHandler;
     use crate::tools::registry::ToolHandler;
     use crate::turn_diff_tracker::TurnDiffTracker;
     use codex_app_server_protocol::AuthMode;
@@ -2839,7 +2918,7 @@ mod tests {
         turn_context: &TurnContext,
     ) -> (Vec<RolloutItem>, Vec<ResponseItem>) {
         let mut rollout_items = Vec::new();
-        let mut live_history = ConversationHistory::new();
+        let mut live_history = ContextManager::new();
 
         let initial_context = session.build_initial_context(turn_context);
         for item in &initial_context {
@@ -3063,6 +3142,48 @@ mod tests {
         assert!(exec_output.output.contains("hi"));
     }
 
+    #[tokio::test]
+    async fn unified_exec_rejects_escalated_permissions_when_policy_not_on_request() {
+        use crate::protocol::AskForApproval;
+        use crate::turn_diff_tracker::TurnDiffTracker;
+
+        let (session, mut turn_context_raw) = make_session_and_context();
+        turn_context_raw.approval_policy = AskForApproval::OnFailure;
+        let session = Arc::new(session);
+        let turn_context = Arc::new(turn_context_raw);
+        let tracker = Arc::new(tokio::sync::Mutex::new(TurnDiffTracker::new()));
+
+        let handler = UnifiedExecHandler;
+        let resp = handler
+            .handle(ToolInvocation {
+                session: Arc::clone(&session),
+                turn: Arc::clone(&turn_context),
+                tracker: Arc::clone(&tracker),
+                call_id: "exec-call".to_string(),
+                tool_name: "exec_command".to_string(),
+                payload: ToolPayload::Function {
+                    arguments: serde_json::json!({
+                        "cmd": "echo hi",
+                        "with_escalated_permissions": true,
+                        "justification": "need unsandboxed execution",
+                    })
+                    .to_string(),
+                },
+            })
+            .await;
+
+        let Err(FunctionCallError::RespondToModel(output)) = resp else {
+            panic!("expected error result");
+        };
+
+        let expected = format!(
+            "approval policy is {policy:?}; reject command — you cannot ask for escalated permissions if the approval policy is {policy:?}",
+            policy = turn_context.approval_policy
+        );
+
+        pretty_assertions::assert_eq!(output, expected);
+    }
+
     #[test]
     fn mcp_init_error_display_prompts_for_github_pat() {
         let server_name = "github";

codex-rs/core/src/codex_delegate.rs

@@ -158,6 +158,11 @@ async fn forward_events(
 ) {
     while let Ok(event) = codex.next_event().await {
         match event {
+            // ignore all legacy delta events
+            Event {
+                id: _,
+                msg: EventMsg::AgentMessageDelta(_) | EventMsg::AgentReasoningDelta(_),
+            } => continue,
             Event {
                 id: _,
                 msg: EventMsg::SessionConfigured(_),

codex-rs/core/src/command_safety/is_dangerous_command.rs

@@ -1,4 +1,38 @@
+use codex_protocol::protocol::AskForApproval;
+use codex_protocol::protocol::SandboxPolicy;
+
 use crate::bash::parse_shell_lc_plain_commands;
+use crate::is_safe_command::is_known_safe_command;
+
+pub fn requires_initial_appoval(
+    policy: AskForApproval,
+    sandbox_policy: &SandboxPolicy,
+    command: &[String],
+    with_escalated_permissions: bool,
+) -> bool {
+    if is_known_safe_command(command) {
+        return false;
+    }
+    match policy {
+        AskForApproval::Never | AskForApproval::OnFailure => false,
+        AskForApproval::OnRequest => {
+            // In DangerFullAccess, only prompt if the command looks dangerous.
+            if matches!(sandbox_policy, SandboxPolicy::DangerFullAccess) {
+                return command_might_be_dangerous(command);
+            }
+
+            // In restricted sandboxes (ReadOnly/WorkspaceWrite), do not prompt for
+            // non‑escalated, non‑dangerous commands — let the sandbox enforce
+            // restrictions (e.g., block network/write) without a user prompt.
+            let wants_escalation: bool = with_escalated_permissions;
+            if wants_escalation {
+                return true;
+            }
+            command_might_be_dangerous(command)
+        }
+        AskForApproval::UnlessTrusted => !is_known_safe_command(command),
+    }
+}
 
 pub fn command_might_be_dangerous(command: &[String]) -> bool {
     if is_dangerous_to_call_with_exec(command) {

codex-rs/core/src/command_safety/is_safe_command.rs

@@ -1,4 +1,5 @@
 use crate::bash::parse_shell_lc_plain_commands;
+use crate::command_safety::windows_safe_commands::is_safe_command_windows;
 
 pub fn is_known_safe_command(command: &[String]) -> bool {
     let command: Vec<String> = command
@@ -11,12 +12,9 @@ pub fn is_known_safe_command(command: &[String]) -> bool {
             }
         })
         .collect();
-    #[cfg(target_os = "windows")]
-    {
-        use super::windows_safe_commands::is_safe_command_windows;
-        if is_safe_command_windows(&command) {
-            return true;
-        }
+
+    if is_safe_command_windows(&command) {
+        return true;
     }
 
     if is_safe_to_call_with_exec(&command) {

codex-rs/core/src/command_safety/mod.rs

@@ -1,4 +1,3 @@
 pub mod is_dangerous_command;
 pub mod is_safe_command;
-#[cfg(target_os = "windows")]
 pub mod windows_safe_commands;

codex-rs/core/src/compact.rs

@@ -1,10 +1,10 @@
 use std::sync::Arc;
 
-use super::Session;
-use super::TurnContext;
-use super::get_last_assistant_message_from_turn;
 use crate::Prompt;
 use crate::client_common::ResponseEvent;
+use crate::codex::Session;
+use crate::codex::TurnContext;
+use crate::codex::get_last_assistant_message_from_turn;
 use crate::error::CodexErr;
 use crate::error::Result as CodexResult;
 use crate::protocol::AgentMessageEvent;
@@ -16,7 +16,6 @@ use crate::protocol::TurnContextItem;
 use crate::protocol::WarningEvent;
 use crate::truncate::truncate_middle;
 use crate::util::backoff;
-use askama::Template;
 use codex_protocol::items::TurnItem;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ResponseInputItem;
@@ -26,16 +25,9 @@ use codex_protocol::user_input::UserInput;
 use futures::prelude::*;
 use tracing::error;
 
-pub const SUMMARIZATION_PROMPT: &str = include_str!("../../templates/compact/prompt.md");
+pub const SUMMARIZATION_PROMPT: &str = include_str!("../templates/compact/prompt.md");
 const COMPACT_USER_MESSAGE_MAX_TOKENS: usize = 20_000;
 
-#[derive(Template)]
-#[template(path = "compact/history_bridge.md", escape = "none")]
-struct HistoryBridgeTemplate<'a> {
-    user_messages_text: &'a str,
-    summary_text: &'a str,
-}
-
 pub(crate) async fn run_inline_auto_compact_task(
     sess: Arc<Session>,
     turn_context: Arc<TurnContext>,
@@ -150,6 +142,7 @@ async fn run_compact_task_inner(
     let history_snapshot = sess.clone_history().await.get_history();
     let summary_text = get_last_assistant_message_from_turn(&history_snapshot).unwrap_or_default();
     let user_messages = collect_user_messages(&history_snapshot);
+
     let initial_context = sess.build_initial_context(turn_context.as_ref());
     let mut new_history = build_compacted_history(initial_context, &user_messages, &summary_text);
     let ghost_snapshots: Vec<ResponseItem> = history_snapshot
@@ -160,6 +153,15 @@ async fn run_compact_task_inner(
     new_history.extend(ghost_snapshots);
     sess.replace_history(new_history).await;
 
+    if let Some(estimated_tokens) = sess
+        .clone_history()
+        .await
+        .estimate_token_count(&turn_context)
+    {
+        sess.override_last_token_usage_estimate(&turn_context, estimated_tokens)
+            .await;
+    }
+
     let rollout_item = RolloutItem::Compacted(CompactedItem {
         message: summary_text.clone(),
     });
@@ -171,7 +173,7 @@ async fn run_compact_task_inner(
     sess.send_event(&turn_context, event).await;
 
     let warning = EventMsg::Warning(WarningEvent {
-        message: "Heads up: Long conversations and multiple compactions can cause the model to be less accurate. Start new a new conversation when possible to keep conversations small and targeted.".to_string(),
+        message: "Heads up: Long conversations and multiple compactions can cause the model to be less accurate. Start a new conversation when possible to keep conversations small and targeted.".to_string(),
     });
     sess.send_event(&turn_context, warning).await;
 }
@@ -224,33 +226,47 @@ fn build_compacted_history_with_limit(
     summary_text: &str,
     max_bytes: usize,
 ) -> Vec<ResponseItem> {
-    let mut user_messages_text = if user_messages.is_empty() {
-        "(none)".to_string()
-    } else {
-        user_messages.join("\n\n")
-    };
-    // Truncate the concatenated prior user messages so the bridge message
-    // stays well under the context window (approx. 4 bytes/token).
-    if user_messages_text.len() > max_bytes {
-        user_messages_text = truncate_middle(&user_messages_text, max_bytes).0;
+    let mut selected_messages: Vec<String> = Vec::new();
+    if max_bytes > 0 {
+        let mut remaining = max_bytes;
+        for message in user_messages.iter().rev() {
+            if remaining == 0 {
+                break;
+            }
+            if message.len() <= remaining {
+                selected_messages.push(message.clone());
+                remaining = remaining.saturating_sub(message.len());
+            } else {
+                let (truncated, _) = truncate_middle(message, remaining);
+                selected_messages.push(truncated);
+                break;
+            }
+        }
+        selected_messages.reverse();
     }
+
+    for message in &selected_messages {
+        history.push(ResponseItem::Message {
+            id: None,
+            role: "user".to_string(),
+            content: vec![ContentItem::InputText {
+                text: message.clone(),
+            }],
+        });
+    }
+
     let summary_text = if summary_text.is_empty() {
         "(no summary available)".to_string()
     } else {
         summary_text.to_string()
     };
-    let Ok(bridge) = HistoryBridgeTemplate {
-        user_messages_text: &user_messages_text,
-        summary_text: &summary_text,
-    }
-    .render() else {
-        return vec![];
-    };
+
     history.push(ResponseItem::Message {
         id: None,
         role: "user".to_string(),
-        content: vec![ContentItem::InputText { text: bridge }],
+        content: vec![ContentItem::InputText { text: summary_text }],
     });
+
     history
 }
 
@@ -390,30 +406,55 @@ mod tests {
             "SUMMARY",
             max_bytes,
         );
+        assert_eq!(history.len(), 2);
 
-        // Expect exactly one bridge message added to history (plus any initial context we provided, which is none).
-        assert_eq!(history.len(), 1);
+        let truncated_message = &history[0];
+        let summary_message = &history[1];
 
-        // Extract the text content of the bridge message.
-        let bridge_text = match &history[0] {
+        let truncated_text = match truncated_message {
             ResponseItem::Message { role, content, .. } if role == "user" => {
                 content_items_to_text(content).unwrap_or_default()
             }
             other => panic!("unexpected item in history: {other:?}"),
         };
 
-        // The bridge should contain the truncation marker and not the full original payload.
         assert!(
-            bridge_text.contains("tokens truncated"),
-            "expected truncation marker in bridge message"
+            truncated_text.contains("tokens truncated"),
+            "expected truncation marker in truncated user message"
         );
         assert!(
-            !bridge_text.contains(&big),
-            "bridge should not include the full oversized user text"
+            !truncated_text.contains(&big),
+            "truncated user message should not include the full oversized user text"
         );
+
+        let summary_text = match summary_message {
+            ResponseItem::Message { role, content, .. } if role == "user" => {
+                content_items_to_text(content).unwrap_or_default()
+            }
+            other => panic!("unexpected item in history: {other:?}"),
+        };
+        assert_eq!(summary_text, "SUMMARY");
+    }
+
+    #[test]
+    fn build_compacted_history_appends_summary_message() {
+        let initial_context: Vec<ResponseItem> = Vec::new();
+        let user_messages = vec!["first user message".to_string()];
+        let summary_text = "summary text";
+
+        let history = build_compacted_history(initial_context, &user_messages, summary_text);
         assert!(
-            bridge_text.contains("SUMMARY"),
-            "bridge should include the provided summary text"
+            !history.is_empty(),
+            "expected compacted history to include summary"
         );
+
+        let last = history.last().expect("history should have a summary entry");
+        let summary = match last {
+            ResponseItem::Message { role, content, .. } if role == "user" => {
+                content_items_to_text(content).unwrap_or_default()
+            }
+            other => panic!("expected summary message, found {other:?}"),
+        };
+        assert_eq!(summary, summary_text);
     }
 }

codex-rs/core/src/config/edit.rs

@@ -3,6 +3,8 @@ use crate::config::types::McpServerConfig;
 use crate::config::types::Notice;
 use anyhow::Context;
 use codex_protocol::config_types::ReasoningEffort;
+use codex_protocol::config_types::TrustLevel;
+use codex_utils_tokenizer::warm_model_cache;
 use std::collections::BTreeMap;
 use std::path::Path;
 use std::path::PathBuf;
@@ -23,13 +25,19 @@ pub enum ConfigEdit {
     },
     /// Toggle the acknowledgement flag under `[notice]`.
     SetNoticeHideFullAccessWarning(bool),
+    /// Toggle the Windows world-writable directories warning acknowledgement flag.
+    SetNoticeHideWorldWritableWarning(bool),
+    /// Toggle the rate limit model nudge acknowledgement flag.
+    SetNoticeHideRateLimitModelNudge(bool),
     /// Toggle the Windows onboarding acknowledgement flag.
     SetWindowsWslSetupAcknowledged(bool),
+    /// Toggle the model migration prompt acknowledgement flag.
+    SetNoticeHideModelMigrationPrompt(String, bool),
     /// Replace the entire `[mcp_servers]` table.
     ReplaceMcpServers(BTreeMap<String, McpServerConfig>),
-    /// Set trust_level = "trusted" under `[projects."<path>"]`,
+    /// Set trust_level under `[projects."<path>"]`,
     /// migrating inline tables to explicit tables.
-    SetProjectTrusted(PathBuf),
+    SetProjectTrustLevel { path: PathBuf, level: TrustLevel },
     /// Set the value stored at the exact dotted path.
     SetPath {
         segments: Vec<String>,
@@ -223,6 +231,9 @@ impl ConfigDocument {
     fn apply(&mut self, edit: &ConfigEdit) -> anyhow::Result<bool> {
         match edit {
             ConfigEdit::SetModel { model, effort } => Ok({
+                if let Some(model) = &model {
+                    warm_model_cache(model)
+                }
                 let mut mutated = false;
                 mutated |= self.write_profile_value(
                     &["model"],
@@ -239,6 +250,23 @@ impl ConfigDocument {
                 &[Notice::TABLE_KEY, "hide_full_access_warning"],
                 value(*acknowledged),
             )),
+            ConfigEdit::SetNoticeHideWorldWritableWarning(acknowledged) => Ok(self.write_value(
+                Scope::Global,
+                &[Notice::TABLE_KEY, "hide_world_writable_warning"],
+                value(*acknowledged),
+            )),
+            ConfigEdit::SetNoticeHideRateLimitModelNudge(acknowledged) => Ok(self.write_value(
+                Scope::Global,
+                &[Notice::TABLE_KEY, "hide_rate_limit_model_nudge"],
+                value(*acknowledged),
+            )),
+            ConfigEdit::SetNoticeHideModelMigrationPrompt(migration_config, acknowledged) => {
+                Ok(self.write_value(
+                    Scope::Global,
+                    &[Notice::TABLE_KEY, migration_config.as_str()],
+                    value(*acknowledged),
+                ))
+            }
             ConfigEdit::SetWindowsWslSetupAcknowledged(acknowledged) => Ok(self.write_value(
                 Scope::Global,
                 &["windows_wsl_setup_acknowledged"],
@@ -247,10 +275,14 @@ impl ConfigDocument {
             ConfigEdit::ReplaceMcpServers(servers) => Ok(self.replace_mcp_servers(servers)),
             ConfigEdit::SetPath { segments, value } => Ok(self.insert(segments, value.clone())),
             ConfigEdit::ClearPath { segments } => Ok(self.clear_owned(segments)),
-            ConfigEdit::SetProjectTrusted(project_path) => {
+            ConfigEdit::SetProjectTrustLevel { path, level } => {
                 // Delegate to the existing, tested logic in config.rs to
                 // ensure tables are explicit and migration is preserved.
-                crate::config::set_project_trusted_inner(&mut self.doc, project_path.as_path())?;
+                crate::config::set_project_trust_level_inner(
+                    &mut self.doc,
+                    path.as_path(),
+                    *level,
+                )?;
                 Ok(true)
             }
         }
@@ -473,6 +505,27 @@ impl ConfigEditsBuilder {
         self
     }
 
+    pub fn set_hide_world_writable_warning(mut self, acknowledged: bool) -> Self {
+        self.edits
+            .push(ConfigEdit::SetNoticeHideWorldWritableWarning(acknowledged));
+        self
+    }
+
+    pub fn set_hide_rate_limit_model_nudge(mut self, acknowledged: bool) -> Self {
+        self.edits
+            .push(ConfigEdit::SetNoticeHideRateLimitModelNudge(acknowledged));
+        self
+    }
+
+    pub fn set_hide_model_migration_prompt(mut self, model: &str, acknowledged: bool) -> Self {
+        self.edits
+            .push(ConfigEdit::SetNoticeHideModelMigrationPrompt(
+                model.to_string(),
+                acknowledged,
+            ));
+        self
+    }
+
     pub fn set_windows_wsl_setup_acknowledged(mut self, acknowledged: bool) -> Self {
         self.edits
             .push(ConfigEdit::SetWindowsWslSetupAcknowledged(acknowledged));
@@ -485,9 +538,15 @@ impl ConfigEditsBuilder {
         self
     }
 
-    pub fn set_project_trusted<P: Into<PathBuf>>(mut self, project_path: P) -> Self {
-        self.edits
-            .push(ConfigEdit::SetProjectTrusted(project_path.into()));
+    pub fn set_project_trust_level<P: Into<PathBuf>>(
+        mut self,
+        project_path: P,
+        trust_level: TrustLevel,
+    ) -> Self {
+        self.edits.push(ConfigEdit::SetProjectTrustLevel {
+            path: project_path.into(),
+            level: trust_level,
+        });
         self
     }
 
@@ -720,6 +779,63 @@ hide_full_access_warning = true
         assert_eq!(contents, expected);
     }
 
+    #[test]
+    fn blocking_set_hide_rate_limit_model_nudge_preserves_table() {
+        let tmp = tempdir().expect("tmpdir");
+        let codex_home = tmp.path();
+        std::fs::write(
+            codex_home.join(CONFIG_TOML_FILE),
+            r#"[notice]
+existing = "value"
+"#,
+        )
+        .expect("seed");
+
+        apply_blocking(
+            codex_home,
+            None,
+            &[ConfigEdit::SetNoticeHideRateLimitModelNudge(true)],
+        )
+        .expect("persist");
+
+        let contents =
+            std::fs::read_to_string(codex_home.join(CONFIG_TOML_FILE)).expect("read config");
+        let expected = r#"[notice]
+existing = "value"
+hide_rate_limit_model_nudge = true
+"#;
+        assert_eq!(contents, expected);
+    }
+    #[test]
+    fn blocking_set_hide_gpt5_1_migration_prompt_preserves_table() {
+        let tmp = tempdir().expect("tmpdir");
+        let codex_home = tmp.path();
+        std::fs::write(
+            codex_home.join(CONFIG_TOML_FILE),
+            r#"[notice]
+existing = "value"
+"#,
+        )
+        .expect("seed");
+        apply_blocking(
+            codex_home,
+            None,
+            &[ConfigEdit::SetNoticeHideModelMigrationPrompt(
+                "hide_gpt5_1_migration_prompt".to_string(),
+                true,
+            )],
+        )
+        .expect("persist");
+
+        let contents =
+            std::fs::read_to_string(codex_home.join(CONFIG_TOML_FILE)).expect("read config");
+        let expected = r#"[notice]
+existing = "value"
+hide_gpt5_1_migration_prompt = true
+"#;
+        assert_eq!(contents, expected);
+    }
+
     #[test]
     fn blocking_replace_mcp_servers_round_trips() {
         let tmp = tempdir().expect("tmpdir");

codex-rs/core/src/config/mod.rs

@@ -38,6 +38,7 @@ use codex_protocol::config_types::ForcedLoginMethod;
 use codex_protocol::config_types::ReasoningEffort;
 use codex_protocol::config_types::ReasoningSummary;
 use codex_protocol::config_types::SandboxMode;
+use codex_protocol::config_types::TrustLevel;
 use codex_protocol::config_types::Verbosity;
 use codex_rmcp_client::OAuthCredentialsStoreMode;
 use dirs::home_dir;
@@ -241,8 +242,6 @@ pub struct Config {
     /// When `true`, run a model-based assessment for commands denied by the sandbox.
     pub experimental_sandbox_command_assessment: bool,
 
-    pub use_experimental_streamable_shell_tool: bool,
-
     /// If set to `true`, used only the experimental unified exec tool.
     pub use_experimental_unified_exec_tool: bool,
 
@@ -384,15 +383,16 @@ fn ensure_no_inline_bearer_tokens(value: &TomlValue) -> std::io::Result<()> {
     Ok(())
 }
 
-pub(crate) fn set_project_trusted_inner(
+pub(crate) fn set_project_trust_level_inner(
     doc: &mut DocumentMut,
     project_path: &Path,
+    trust_level: TrustLevel,
 ) -> anyhow::Result<()> {
     // Ensure we render a human-friendly structure:
     //
     // [projects]
     // [projects."/path/to/project"]
-    // trust_level = "trusted"
+    // trust_level = "trusted" or "untrusted"
     //
     // rather than inline tables like:
     //
@@ -448,17 +448,21 @@ pub(crate) fn set_project_trusted_inner(
         return Err(anyhow::anyhow!("project table missing for {project_key}"));
     };
     proj_tbl.set_implicit(false);
-    proj_tbl["trust_level"] = toml_edit::value("trusted");
+    proj_tbl["trust_level"] = toml_edit::value(trust_level.to_string());
     Ok(())
 }
 
-/// Patch `CODEX_HOME/config.toml` project state.
+/// Patch `CODEX_HOME/config.toml` project state to set trust level.
 /// Use with caution.
-pub fn set_project_trusted(codex_home: &Path, project_path: &Path) -> anyhow::Result<()> {
+pub fn set_project_trust_level(
+    codex_home: &Path,
+    project_path: &Path,
+    trust_level: TrustLevel,
+) -> anyhow::Result<()> {
     use crate::config::edit::ConfigEditsBuilder;
 
     ConfigEditsBuilder::new(codex_home)
-        .set_project_trusted(project_path)
+        .set_project_trust_level(project_path, trust_level)
         .apply_blocking()
 }
 
@@ -655,7 +659,6 @@ pub struct ConfigToml {
     /// Legacy, now use features
     pub experimental_instructions_file: Option<PathBuf>,
     pub experimental_compact_prompt_file: Option<PathBuf>,
-    pub experimental_use_exec_command_tool: Option<bool>,
     pub experimental_use_unified_exec_tool: Option<bool>,
     pub experimental_use_rmcp_client: Option<bool>,
     pub experimental_use_freeform_apply_patch: Option<bool>,
@@ -689,15 +692,16 @@ impl From<ConfigToml> for UserSavedConfig {
 
 #[derive(Deserialize, Debug, Clone, PartialEq, Eq)]
 pub struct ProjectConfig {
-    pub trust_level: Option<String>,
+    pub trust_level: Option<TrustLevel>,
 }
 
 impl ProjectConfig {
     pub fn is_trusted(&self) -> bool {
-        match &self.trust_level {
-            Some(trust_level) => trust_level == "trusted",
-            None => false,
-        }
+        matches!(self.trust_level, Some(TrustLevel::Trusted))
+    }
+
+    pub fn is_untrusted(&self) -> bool {
+        matches!(self.trust_level, Some(TrustLevel::Untrusted))
     }
 }
 
@@ -738,9 +742,9 @@ impl ConfigToml {
             .or(profile_sandbox_mode)
             .or(self.sandbox_mode)
             .or_else(|| {
-                // if no sandbox_mode is set, but user has marked directory as trusted, use WorkspaceWrite
+                // if no sandbox_mode is set, but user has marked directory as trusted or untrusted, use WorkspaceWrite
                 self.get_active_project(resolved_cwd).and_then(|p| {
-                    if p.is_trusted() {
+                    if p.is_trusted() || p.is_untrusted() {
                         Some(SandboxMode::WorkspaceWrite)
                     } else {
                         None
@@ -961,6 +965,9 @@ impl Config {
                 if active_project.is_trusted() {
                     // If no explicit approval policy is set, but we trust cwd, default to OnRequest
                     AskForApproval::OnRequest
+                } else if active_project.is_untrusted() {
+                    // If project is explicitly marked untrusted, require approval for non-safe commands
+                    AskForApproval::UnlessTrusted
                 } else {
                     AskForApproval::default()
                 }
@@ -999,7 +1006,6 @@ impl Config {
 
         let include_apply_patch_tool_flag = features.enabled(Feature::ApplyPatchFreeform);
         let tools_web_search_request = features.enabled(Feature::WebSearchRequest);
-        let use_experimental_streamable_shell_tool = features.enabled(Feature::StreamableShell);
         let use_experimental_unified_exec_tool = features.enabled(Feature::UnifiedExec);
         let use_experimental_use_rmcp_client = features.enabled(Feature::RmcpClient);
         let experimental_sandbox_command_assessment =
@@ -1156,7 +1162,6 @@ impl Config {
             include_apply_patch_tool: include_apply_patch_tool_flag,
             tools_web_search_request,
             experimental_sandbox_command_assessment,
-            use_experimental_streamable_shell_tool,
             use_experimental_unified_exec_tool,
             use_experimental_use_rmcp_client,
             features,
@@ -1328,7 +1333,7 @@ persistence = "none"
     }
 
     #[test]
-    fn tui_config_missing_notifications_field_defaults_to_disabled() {
+    fn tui_config_missing_notifications_field_defaults_to_enabled() {
         let cfg = r#"
 [tui]
 "#;
@@ -1337,7 +1342,7 @@ persistence = "none"
             .expect("TUI config without notifications should succeed");
         let tui = parsed.tui.expect("config should include tui section");
 
-        assert_eq!(tui.notifications, Notifications::Enabled(false));
+        assert_eq!(tui.notifications, Notifications::Enabled(true));
     }
 
     #[test]
@@ -1715,7 +1720,6 @@ trust_level = "trusted"
     fn legacy_toggles_map_to_features() -> std::io::Result<()> {
         let codex_home = TempDir::new()?;
         let cfg = ConfigToml {
-            experimental_use_exec_command_tool: Some(true),
             experimental_use_unified_exec_tool: Some(true),
             experimental_use_rmcp_client: Some(true),
             experimental_use_freeform_apply_patch: Some(true),
@@ -1729,12 +1733,11 @@ trust_level = "trusted"
         )?;
 
         assert!(config.features.enabled(Feature::ApplyPatchFreeform));
-        assert!(config.features.enabled(Feature::StreamableShell));
         assert!(config.features.enabled(Feature::UnifiedExec));
         assert!(config.features.enabled(Feature::RmcpClient));
 
         assert!(config.include_apply_patch_tool);
-        assert!(config.use_experimental_streamable_shell_tool);
+
         assert!(config.use_experimental_unified_exec_tool);
         assert!(config.use_experimental_use_rmcp_client);
 
@@ -2902,7 +2905,6 @@ model_verbosity = "high"
                 include_apply_patch_tool: false,
                 tools_web_search_request: false,
                 experimental_sandbox_command_assessment: false,
-                use_experimental_streamable_shell_tool: false,
                 use_experimental_unified_exec_tool: false,
                 use_experimental_use_rmcp_client: false,
                 features: Features::with_defaults(),
@@ -2974,7 +2976,6 @@ model_verbosity = "high"
             include_apply_patch_tool: false,
             tools_web_search_request: false,
             experimental_sandbox_command_assessment: false,
-            use_experimental_streamable_shell_tool: false,
             use_experimental_unified_exec_tool: false,
             use_experimental_use_rmcp_client: false,
             features: Features::with_defaults(),
@@ -3061,7 +3062,6 @@ model_verbosity = "high"
             include_apply_patch_tool: false,
             tools_web_search_request: false,
             experimental_sandbox_command_assessment: false,
-            use_experimental_streamable_shell_tool: false,
             use_experimental_unified_exec_tool: false,
             use_experimental_use_rmcp_client: false,
             features: Features::with_defaults(),
@@ -3134,7 +3134,6 @@ model_verbosity = "high"
             include_apply_patch_tool: false,
             tools_web_search_request: false,
             experimental_sandbox_command_assessment: false,
-            use_experimental_streamable_shell_tool: false,
             use_experimental_unified_exec_tool: false,
             use_experimental_use_rmcp_client: false,
             features: Features::with_defaults(),
@@ -3175,7 +3174,7 @@ model_verbosity = "high"
         let project_dir = Path::new("/some/path");
         let mut doc = DocumentMut::new();
 
-        set_project_trusted_inner(&mut doc, project_dir)?;
+        set_project_trust_level_inner(&mut doc, project_dir, TrustLevel::Trusted)?;
 
         let contents = doc.to_string();
 
@@ -3215,7 +3214,7 @@ trust_level = "trusted"
         let mut doc = initial.parse::<DocumentMut>()?;
 
         // Run the function; it should convert to explicit tables and set trusted
-        set_project_trusted_inner(&mut doc, project_dir)?;
+        set_project_trust_level_inner(&mut doc, project_dir, TrustLevel::Trusted)?;
 
         let contents = doc.to_string();
 
@@ -3242,7 +3241,7 @@ model = "foo""#;
 
         // Approve a new directory
         let new_project = Path::new("/Users/mbolin/code/codex2");
-        set_project_trusted_inner(&mut doc, new_project)?;
+        set_project_trust_level_inner(&mut doc, new_project, TrustLevel::Trusted)?;
 
         let contents = doc.to_string();
 
@@ -3265,6 +3264,87 @@ trust_level = "trusted"
 
         Ok(())
     }
+
+    #[test]
+    fn test_untrusted_project_gets_workspace_write_sandbox() -> anyhow::Result<()> {
+        let config_with_untrusted = r#"
+[projects."/tmp/test"]
+trust_level = "untrusted"
+"#;
+
+        let cfg = toml::from_str::<ConfigToml>(config_with_untrusted)
+            .expect("TOML deserialization should succeed");
+
+        let resolution = cfg.derive_sandbox_policy(None, None, &PathBuf::from("/tmp/test"));
+
+        // Verify that untrusted projects get WorkspaceWrite (or ReadOnly on Windows due to downgrade)
+        if cfg!(target_os = "windows") {
+            assert!(
+                matches!(resolution.policy, SandboxPolicy::ReadOnly),
+                "Expected ReadOnly on Windows, got {:?}",
+                resolution.policy
+            );
+        } else {
+            assert!(
+                matches!(resolution.policy, SandboxPolicy::WorkspaceWrite { .. }),
+                "Expected WorkspaceWrite for untrusted project, got {:?}",
+                resolution.policy
+            );
+        }
+
+        Ok(())
+    }
+
+    #[test]
+    fn test_untrusted_project_gets_unless_trusted_approval_policy() -> std::io::Result<()> {
+        let codex_home = TempDir::new()?;
+        let test_project_dir = TempDir::new()?;
+        let test_path = test_project_dir.path();
+
+        let mut projects = std::collections::HashMap::new();
+        projects.insert(
+            test_path.to_string_lossy().to_string(),
+            ProjectConfig {
+                trust_level: Some(TrustLevel::Untrusted),
+            },
+        );
+
+        let cfg = ConfigToml {
+            projects: Some(projects),
+            ..Default::default()
+        };
+
+        let config = Config::load_from_base_config_with_overrides(
+            cfg,
+            ConfigOverrides {
+                cwd: Some(test_path.to_path_buf()),
+                ..Default::default()
+            },
+            codex_home.path().to_path_buf(),
+        )?;
+
+        // Verify that untrusted projects get UnlessTrusted approval policy
+        assert_eq!(
+            config.approval_policy,
+            AskForApproval::UnlessTrusted,
+            "Expected UnlessTrusted approval policy for untrusted project"
+        );
+
+        // Verify that untrusted projects still get WorkspaceWrite sandbox (or ReadOnly on Windows)
+        if cfg!(target_os = "windows") {
+            assert!(
+                matches!(config.sandbox_policy, SandboxPolicy::ReadOnly),
+                "Expected ReadOnly on Windows"
+            );
+        } else {
+            assert!(
+                matches!(config.sandbox_policy, SandboxPolicy::WorkspaceWrite { .. }),
+                "Expected WorkspaceWrite sandbox for untrusted project"
+            );
+        }
+
+        Ok(())
+    }
 }
 
 #[cfg(test)]

codex-rs/core/src/config/profile.rs

@@ -25,7 +25,6 @@ pub struct ConfigProfile {
     pub experimental_compact_prompt_file: Option<PathBuf>,
     pub include_apply_patch_tool: Option<bool>,
     pub experimental_use_unified_exec_tool: Option<bool>,
-    pub experimental_use_exec_command_tool: Option<bool>,
     pub experimental_use_rmcp_client: Option<bool>,
     pub experimental_use_freeform_apply_patch: Option<bool>,
     pub experimental_sandbox_command_assessment: Option<bool>,

codex-rs/core/src/config/types.rs

@@ -338,7 +338,7 @@ pub enum Notifications {
 
 impl Default for Notifications {
     fn default() -> Self {
-        Self::Enabled(false)
+        Self::Enabled(true)
     }
 }
 
@@ -346,7 +346,7 @@ impl Default for Notifications {
 #[derive(Deserialize, Debug, Clone, PartialEq, Default)]
 pub struct Tui {
     /// Enable desktop notifications from the TUI when the terminal is unfocused.
-    /// Defaults to `false`.
+    /// Defaults to `true`.
     #[serde(default)]
     pub notifications: Notifications,
 }
@@ -358,6 +358,12 @@ pub struct Tui {
 pub struct Notice {
     /// Tracks whether the user has acknowledged the full access warning prompt.
     pub hide_full_access_warning: Option<bool>,
+    /// Tracks whether the user has acknowledged the Windows world-writable directories warning.
+    pub hide_world_writable_warning: Option<bool>,
+    /// Tracks whether the user opted out of the rate limit model switch reminder.
+    pub hide_rate_limit_model_nudge: Option<bool>,
+    /// Tracks whether the user has seen the model migration prompt
+    pub hide_gpt5_1_migration_prompt: Option<bool>,
 }
 
 impl Notice {

codex-rs/core/src/context_manager/history.rs

@@ -0,0 +1,216 @@
+use crate::codex::TurnContext;
+use crate::context_manager::normalize;
+use crate::context_manager::truncate;
+use crate::context_manager::truncate::format_output_for_model_body;
+use crate::context_manager::truncate::globally_truncate_function_output_items;
+use codex_protocol::models::FunctionCallOutputPayload;
+use codex_protocol::models::ResponseItem;
+use codex_protocol::protocol::TokenUsage;
+use codex_protocol::protocol::TokenUsageInfo;
+use codex_utils_tokenizer::Tokenizer;
+use std::ops::Deref;
+
+const CONTEXT_WINDOW_HARD_LIMIT_FACTOR: f64 = 1.1;
+const CONTEXT_WINDOW_HARD_LIMIT_BYTES: usize =
+    (truncate::MODEL_FORMAT_MAX_BYTES as f64 * CONTEXT_WINDOW_HARD_LIMIT_FACTOR) as usize;
+const CONTEXT_WINDOW_HARD_LIMIT_LINES: usize =
+    (truncate::MODEL_FORMAT_MAX_LINES as f64 * CONTEXT_WINDOW_HARD_LIMIT_FACTOR) as usize;
+
+/// Transcript of conversation history
+#[derive(Debug, Clone, Default)]
+pub(crate) struct ContextManager {
+    /// The oldest items are at the beginning of the vector.
+    items: Vec<ResponseItem>,
+    token_info: Option<TokenUsageInfo>,
+}
+
+impl ContextManager {
+    pub(crate) fn new() -> Self {
+        Self {
+            items: Vec::new(),
+            token_info: TokenUsageInfo::new_or_append(&None, &None, None),
+        }
+    }
+
+    pub(crate) fn token_info(&self) -> Option<TokenUsageInfo> {
+        self.token_info.clone()
+    }
+
+    pub(crate) fn set_token_info(&mut self, info: Option<TokenUsageInfo>) {
+        self.token_info = info;
+    }
+
+    pub(crate) fn set_token_usage_full(&mut self, context_window: i64) {
+        match &mut self.token_info {
+            Some(info) => info.fill_to_context_window(context_window),
+            None => {
+                self.token_info = Some(TokenUsageInfo::full_context_window(context_window));
+            }
+        }
+    }
+
+    /// `items` is ordered from oldest to newest.
+    pub(crate) fn record_items<I>(&mut self, items: I)
+    where
+        I: IntoIterator,
+        I::Item: std::ops::Deref<Target = ResponseItem>,
+    {
+        for item in items {
+            let item_ref = item.deref();
+            let is_ghost_snapshot = matches!(item_ref, ResponseItem::GhostSnapshot { .. });
+            if !is_api_message(item_ref) && !is_ghost_snapshot {
+                continue;
+            }
+
+            let processed = Self::process_item(&item);
+            self.items.push(processed);
+        }
+    }
+
+    pub(crate) fn get_history(&mut self) -> Vec<ResponseItem> {
+        self.normalize_history();
+        self.contents()
+    }
+
+    // Returns the history prepared for sending to the model.
+    // With extra response items filtered out and GhostCommits removed.
+    pub(crate) fn get_history_for_prompt(&mut self) -> Vec<ResponseItem> {
+        let mut history = self.get_history();
+        Self::remove_ghost_snapshots(&mut history);
+        history
+    }
+
+    // Estimate the number of tokens in the history. Return None if no tokenizer
+    // is available. This does not consider the reasoning traces.
+    // /!\ The value is a lower bound estimate and does not represent the exact
+    // context length.
+    pub(crate) fn estimate_token_count(&self, turn_context: &TurnContext) -> Option<i64> {
+        let model = turn_context.client.get_model();
+        let tokenizer = Tokenizer::for_model(model.as_str()).ok()?;
+        let model_family = turn_context.client.get_model_family();
+
+        Some(
+            self.items
+                .iter()
+                .map(|item| {
+                    serde_json::to_string(&item)
+                        .map(|item| tokenizer.count(&item))
+                        .unwrap_or_default()
+                })
+                .sum::<i64>()
+                + tokenizer.count(model_family.base_instructions.as_str()),
+        )
+    }
+
+    pub(crate) fn remove_first_item(&mut self) {
+        if !self.items.is_empty() {
+            // Remove the oldest item (front of the list). Items are ordered from
+            // oldest → newest, so index 0 is the first entry recorded.
+            let removed = self.items.remove(0);
+            // If the removed item participates in a call/output pair, also remove
+            // its corresponding counterpart to keep the invariants intact without
+            // running a full normalization pass.
+            normalize::remove_corresponding_for(&mut self.items, &removed);
+        }
+    }
+
+    pub(crate) fn replace(&mut self, items: Vec<ResponseItem>) {
+        self.items = items;
+    }
+
+    pub(crate) fn update_token_info(
+        &mut self,
+        usage: &TokenUsage,
+        model_context_window: Option<i64>,
+    ) {
+        self.token_info = TokenUsageInfo::new_or_append(
+            &self.token_info,
+            &Some(usage.clone()),
+            model_context_window,
+        );
+    }
+
+    /// This function enforces a couple of invariants on the in-memory history:
+    /// 1. every call (function/custom) has a corresponding output entry
+    /// 2. every output has a corresponding call entry
+    fn normalize_history(&mut self) {
+        // all function/tool calls must have a corresponding output
+        normalize::ensure_call_outputs_present(&mut self.items);
+
+        // all outputs must have a corresponding function/tool call
+        normalize::remove_orphan_outputs(&mut self.items);
+    }
+
+    /// Returns a clone of the contents in the transcript.
+    fn contents(&self) -> Vec<ResponseItem> {
+        self.items.clone()
+    }
+
+    fn remove_ghost_snapshots(items: &mut Vec<ResponseItem>) {
+        items.retain(|item| !matches!(item, ResponseItem::GhostSnapshot { .. }));
+    }
+
+    fn process_item(item: &ResponseItem) -> ResponseItem {
+        match item {
+            ResponseItem::FunctionCallOutput { call_id, output } => {
+                let truncated = format_output_for_model_body(
+                    output.content.as_str(),
+                    CONTEXT_WINDOW_HARD_LIMIT_BYTES,
+                    CONTEXT_WINDOW_HARD_LIMIT_LINES,
+                );
+                let truncated_items = output
+                    .content_items
+                    .as_ref()
+                    .map(|items| globally_truncate_function_output_items(items));
+                ResponseItem::FunctionCallOutput {
+                    call_id: call_id.clone(),
+                    output: FunctionCallOutputPayload {
+                        content: truncated,
+                        content_items: truncated_items,
+                        success: output.success,
+                    },
+                }
+            }
+            ResponseItem::CustomToolCallOutput { call_id, output } => {
+                let truncated = format_output_for_model_body(
+                    output,
+                    CONTEXT_WINDOW_HARD_LIMIT_BYTES,
+                    CONTEXT_WINDOW_HARD_LIMIT_LINES,
+                );
+                ResponseItem::CustomToolCallOutput {
+                    call_id: call_id.clone(),
+                    output: truncated,
+                }
+            }
+            ResponseItem::Message { .. }
+            | ResponseItem::Reasoning { .. }
+            | ResponseItem::LocalShellCall { .. }
+            | ResponseItem::FunctionCall { .. }
+            | ResponseItem::WebSearchCall { .. }
+            | ResponseItem::CustomToolCall { .. }
+            | ResponseItem::GhostSnapshot { .. }
+            | ResponseItem::Other => item.clone(),
+        }
+    }
+}
+
+/// API messages include every non-system item (user/assistant messages, reasoning,
+/// tool calls, tool outputs, shell calls, and web-search calls).
+fn is_api_message(message: &ResponseItem) -> bool {
+    match message {
+        ResponseItem::Message { role, .. } => role.as_str() != "system",
+        ResponseItem::FunctionCallOutput { .. }
+        | ResponseItem::FunctionCall { .. }
+        | ResponseItem::CustomToolCall { .. }
+        | ResponseItem::CustomToolCallOutput { .. }
+        | ResponseItem::LocalShellCall { .. }
+        | ResponseItem::Reasoning { .. }
+        | ResponseItem::WebSearchCall { .. } => true,
+        ResponseItem::GhostSnapshot { .. } => false,
+        ResponseItem::Other => false,
+    }
+}
+
+#[cfg(test)]
+#[path = "history_tests.rs"]
+mod tests;

codex-rs/core/src/context_manager/history_tests.rs

@@ -0,0 +1,867 @@
+use super::*;
+use crate::context_manager::MODEL_FORMAT_MAX_LINES;
+use crate::context_manager::truncate;
+use codex_git::GhostCommit;
+use codex_protocol::models::ContentItem;
+use codex_protocol::models::FunctionCallOutputContentItem;
+use codex_protocol::models::FunctionCallOutputPayload;
+use codex_protocol::models::LocalShellAction;
+use codex_protocol::models::LocalShellExecAction;
+use codex_protocol::models::LocalShellStatus;
+use codex_protocol::models::ReasoningItemContent;
+use codex_protocol::models::ReasoningItemReasoningSummary;
+use pretty_assertions::assert_eq;
+use regex_lite::Regex;
+
+fn assistant_msg(text: &str) -> ResponseItem {
+    ResponseItem::Message {
+        id: None,
+        role: "assistant".to_string(),
+        content: vec![ContentItem::OutputText {
+            text: text.to_string(),
+        }],
+    }
+}
+
+fn create_history_with_items(items: Vec<ResponseItem>) -> ContextManager {
+    let mut h = ContextManager::new();
+    h.record_items(items.iter());
+    h
+}
+
+fn user_msg(text: &str) -> ResponseItem {
+    ResponseItem::Message {
+        id: None,
+        role: "user".to_string(),
+        content: vec![ContentItem::OutputText {
+            text: text.to_string(),
+        }],
+    }
+}
+
+fn reasoning_msg(text: &str) -> ResponseItem {
+    ResponseItem::Reasoning {
+        id: String::new(),
+        summary: vec![ReasoningItemReasoningSummary::SummaryText {
+            text: "summary".to_string(),
+        }],
+        content: Some(vec![ReasoningItemContent::ReasoningText {
+            text: text.to_string(),
+        }]),
+        encrypted_content: None,
+    }
+}
+
+#[test]
+fn filters_non_api_messages() {
+    let mut h = ContextManager::default();
+    // System message is not API messages; Other is ignored.
+    let system = ResponseItem::Message {
+        id: None,
+        role: "system".to_string(),
+        content: vec![ContentItem::OutputText {
+            text: "ignored".to_string(),
+        }],
+    };
+    let reasoning = reasoning_msg("thinking...");
+    h.record_items([&system, &reasoning, &ResponseItem::Other]);
+
+    // User and assistant should be retained.
+    let u = user_msg("hi");
+    let a = assistant_msg("hello");
+    h.record_items([&u, &a]);
+
+    let items = h.contents();
+    assert_eq!(
+        items,
+        vec![
+            ResponseItem::Reasoning {
+                id: String::new(),
+                summary: vec![ReasoningItemReasoningSummary::SummaryText {
+                    text: "summary".to_string(),
+                }],
+                content: Some(vec![ReasoningItemContent::ReasoningText {
+                    text: "thinking...".to_string(),
+                }]),
+                encrypted_content: None,
+            },
+            ResponseItem::Message {
+                id: None,
+                role: "user".to_string(),
+                content: vec![ContentItem::OutputText {
+                    text: "hi".to_string()
+                }]
+            },
+            ResponseItem::Message {
+                id: None,
+                role: "assistant".to_string(),
+                content: vec![ContentItem::OutputText {
+                    text: "hello".to_string()
+                }]
+            }
+        ]
+    );
+}
+
+#[test]
+fn get_history_for_prompt_drops_ghost_commits() {
+    let items = vec![ResponseItem::GhostSnapshot {
+        ghost_commit: GhostCommit::new("ghost-1".to_string(), None, Vec::new(), Vec::new()),
+    }];
+    let mut history = create_history_with_items(items);
+    let filtered = history.get_history_for_prompt();
+    assert_eq!(filtered, vec![]);
+}
+
+#[test]
+fn remove_first_item_removes_matching_output_for_function_call() {
+    let items = vec![
+        ResponseItem::FunctionCall {
+            id: None,
+            name: "do_it".to_string(),
+            arguments: "{}".to_string(),
+            call_id: "call-1".to_string(),
+        },
+        ResponseItem::FunctionCallOutput {
+            call_id: "call-1".to_string(),
+            output: FunctionCallOutputPayload {
+                content: "ok".to_string(),
+                ..Default::default()
+            },
+        },
+    ];
+    let mut h = create_history_with_items(items);
+    h.remove_first_item();
+    assert_eq!(h.contents(), vec![]);
+}
+
+#[test]
+fn remove_first_item_removes_matching_call_for_output() {
+    let items = vec![
+        ResponseItem::FunctionCallOutput {
+            call_id: "call-2".to_string(),
+            output: FunctionCallOutputPayload {
+                content: "ok".to_string(),
+                ..Default::default()
+            },
+        },
+        ResponseItem::FunctionCall {
+            id: None,
+            name: "do_it".to_string(),
+            arguments: "{}".to_string(),
+            call_id: "call-2".to_string(),
+        },
+    ];
+    let mut h = create_history_with_items(items);
+    h.remove_first_item();
+    assert_eq!(h.contents(), vec![]);
+}
+
+#[test]
+fn remove_first_item_handles_local_shell_pair() {
+    let items = vec![
+        ResponseItem::LocalShellCall {
+            id: None,
+            call_id: Some("call-3".to_string()),
+            status: LocalShellStatus::Completed,
+            action: LocalShellAction::Exec(LocalShellExecAction {
+                command: vec!["echo".to_string(), "hi".to_string()],
+                timeout_ms: None,
+                working_directory: None,
+                env: None,
+                user: None,
+            }),
+        },
+        ResponseItem::FunctionCallOutput {
+            call_id: "call-3".to_string(),
+            output: FunctionCallOutputPayload {
+                content: "ok".to_string(),
+                ..Default::default()
+            },
+        },
+    ];
+    let mut h = create_history_with_items(items);
+    h.remove_first_item();
+    assert_eq!(h.contents(), vec![]);
+}
+
+#[test]
+fn remove_first_item_handles_custom_tool_pair() {
+    let items = vec![
+        ResponseItem::CustomToolCall {
+            id: None,
+            status: None,
+            call_id: "tool-1".to_string(),
+            name: "my_tool".to_string(),
+            input: "{}".to_string(),
+        },
+        ResponseItem::CustomToolCallOutput {
+            call_id: "tool-1".to_string(),
+            output: "ok".to_string(),
+        },
+    ];
+    let mut h = create_history_with_items(items);
+    h.remove_first_item();
+    assert_eq!(h.contents(), vec![]);
+}
+
+#[test]
+fn normalization_retains_local_shell_outputs() {
+    let items = vec![
+        ResponseItem::LocalShellCall {
+            id: None,
+            call_id: Some("shell-1".to_string()),
+            status: LocalShellStatus::Completed,
+            action: LocalShellAction::Exec(LocalShellExecAction {
+                command: vec!["echo".to_string(), "hi".to_string()],
+                timeout_ms: None,
+                working_directory: None,
+                env: None,
+                user: None,
+            }),
+        },
+        ResponseItem::FunctionCallOutput {
+            call_id: "shell-1".to_string(),
+            output: FunctionCallOutputPayload {
+                content: "ok".to_string(),
+                ..Default::default()
+            },
+        },
+    ];
+
+    let mut history = create_history_with_items(items.clone());
+    let normalized = history.get_history();
+    assert_eq!(normalized, items);
+}
+
+#[test]
+fn record_items_truncates_function_call_output_content() {
+    let mut history = ContextManager::new();
+    let long_line = "a very long line to trigger truncation\n";
+    let long_output = long_line.repeat(2_500);
+    let item = ResponseItem::FunctionCallOutput {
+        call_id: "call-100".to_string(),
+        output: FunctionCallOutputPayload {
+            content: long_output.clone(),
+            success: Some(true),
+            ..Default::default()
+        },
+    };
+
+    history.record_items([&item]);
+
+    assert_eq!(history.items.len(), 1);
+    match &history.items[0] {
+        ResponseItem::FunctionCallOutput { output, .. } => {
+            assert_ne!(output.content, long_output);
+            assert!(
+                output.content.starts_with("Total output lines:"),
+                "expected truncated summary, got {}",
+                output.content
+            );
+        }
+        other => panic!("unexpected history item: {other:?}"),
+    }
+}
+
+#[test]
+fn record_items_truncates_custom_tool_call_output_content() {
+    let mut history = ContextManager::new();
+    let line = "custom output that is very long\n";
+    let long_output = line.repeat(2_500);
+    let item = ResponseItem::CustomToolCallOutput {
+        call_id: "tool-200".to_string(),
+        output: long_output.clone(),
+    };
+
+    history.record_items([&item]);
+
+    assert_eq!(history.items.len(), 1);
+    match &history.items[0] {
+        ResponseItem::CustomToolCallOutput { output, .. } => {
+            assert_ne!(output, &long_output);
+            assert!(
+                output.starts_with("Total output lines:"),
+                "expected truncated summary, got {output}"
+            );
+        }
+        other => panic!("unexpected history item: {other:?}"),
+    }
+}
+
+fn assert_truncated_message_matches(message: &str, line: &str, total_lines: usize) {
+    let pattern = truncated_message_pattern(line, total_lines);
+    let regex = Regex::new(&pattern).unwrap_or_else(|err| {
+        panic!("failed to compile regex {pattern}: {err}");
+    });
+    let captures = regex
+        .captures(message)
+        .unwrap_or_else(|| panic!("message failed to match pattern {pattern}: {message}"));
+    let body = captures
+        .name("body")
+        .expect("missing body capture")
+        .as_str();
+    assert!(
+        body.len() <= truncate::MODEL_FORMAT_MAX_BYTES,
+        "body exceeds byte limit: {} bytes",
+        body.len()
+    );
+}
+
+fn truncated_message_pattern(line: &str, total_lines: usize) -> String {
+    let head_lines = MODEL_FORMAT_MAX_LINES / 2;
+    let tail_lines = MODEL_FORMAT_MAX_LINES - head_lines;
+    let head_take = head_lines.min(total_lines);
+    let tail_take = tail_lines.min(total_lines.saturating_sub(head_take));
+    let omitted = total_lines.saturating_sub(head_take + tail_take);
+    let escaped_line = regex_lite::escape(line);
+    if omitted == 0 {
+        return format!(
+            r"(?s)^Total output lines: {total_lines}\n\n(?P<body>{escaped_line}.*\n\[\.{{3}} output truncated to fit {max_bytes} bytes \.{{3}}]\n\n.*)$",
+            max_bytes = truncate::MODEL_FORMAT_MAX_BYTES,
+        );
+    }
+    format!(
+        r"(?s)^Total output lines: {total_lines}\n\n(?P<body>{escaped_line}.*\n\[\.{{3}} omitted {omitted} of {total_lines} lines \.{{3}}]\n\n.*)$",
+    )
+}
+
+#[test]
+fn format_exec_output_truncates_large_error() {
+    let line = "very long execution error line that should trigger truncation\n";
+    let large_error = line.repeat(2_500); // way beyond both byte and line limits
+
+    let truncated = truncate::format_output_for_model_body(
+        &large_error,
+        truncate::MODEL_FORMAT_MAX_BYTES,
+        truncate::MODEL_FORMAT_MAX_LINES,
+    );
+
+    let total_lines = large_error.lines().count();
+    assert_truncated_message_matches(&truncated, line, total_lines);
+    assert_ne!(truncated, large_error);
+}
+
+#[test]
+fn format_exec_output_marks_byte_truncation_without_omitted_lines() {
+    let long_line = "a".repeat(truncate::MODEL_FORMAT_MAX_BYTES + 50);
+    let truncated = truncate::format_output_for_model_body(
+        &long_line,
+        truncate::MODEL_FORMAT_MAX_BYTES,
+        truncate::MODEL_FORMAT_MAX_LINES,
+    );
+
+    assert_ne!(truncated, long_line);
+    let marker_line = format!(
+        "[... output truncated to fit {} bytes ...]",
+        truncate::MODEL_FORMAT_MAX_BYTES
+    );
+    assert!(
+        truncated.contains(&marker_line),
+        "missing byte truncation marker: {truncated}"
+    );
+    assert!(
+        !truncated.contains("omitted"),
+        "line omission marker should not appear when no lines were dropped: {truncated}"
+    );
+}
+
+#[test]
+fn format_exec_output_returns_original_when_within_limits() {
+    let content = "example output\n".repeat(10);
+
+    assert_eq!(
+        truncate::format_output_for_model_body(
+            &content,
+            truncate::MODEL_FORMAT_MAX_BYTES,
+            truncate::MODEL_FORMAT_MAX_LINES
+        ),
+        content
+    );
+}
+
+#[test]
+fn format_exec_output_reports_omitted_lines_and_keeps_head_and_tail() {
+    let total_lines = truncate::MODEL_FORMAT_MAX_LINES + 100;
+    let content: String = (0..total_lines)
+        .map(|idx| format!("line-{idx}\n"))
+        .collect();
+
+    let truncated = truncate::format_output_for_model_body(
+        &content,
+        truncate::MODEL_FORMAT_MAX_BYTES,
+        truncate::MODEL_FORMAT_MAX_LINES,
+    );
+    let omitted = total_lines - truncate::MODEL_FORMAT_MAX_LINES;
+    let expected_marker = format!("[... omitted {omitted} of {total_lines} lines ...]");
+
+    assert!(
+        truncated.contains(&expected_marker),
+        "missing omitted marker: {truncated}"
+    );
+    assert!(
+        truncated.contains("line-0\n"),
+        "expected head line to remain: {truncated}"
+    );
+
+    let last_line = format!("line-{}\n", total_lines - 1);
+    assert!(
+        truncated.contains(&last_line),
+        "expected tail line to remain: {truncated}"
+    );
+}
+
+#[test]
+fn format_exec_output_prefers_line_marker_when_both_limits_exceeded() {
+    let total_lines = truncate::MODEL_FORMAT_MAX_LINES + 42;
+    let long_line = "x".repeat(256);
+    let content: String = (0..total_lines)
+        .map(|idx| format!("line-{idx}-{long_line}\n"))
+        .collect();
+
+    let truncated = truncate::format_output_for_model_body(
+        &content,
+        truncate::MODEL_FORMAT_MAX_BYTES,
+        truncate::MODEL_FORMAT_MAX_LINES,
+    );
+
+    assert!(
+        truncated.contains("[... omitted 42 of 298 lines ...]"),
+        "expected omitted marker when line count exceeds limit: {truncated}"
+    );
+    assert!(
+        !truncated.contains("output truncated to fit"),
+        "line omission marker should take precedence over byte marker: {truncated}"
+    );
+}
+
+#[test]
+fn truncates_across_multiple_under_limit_texts_and_reports_omitted() {
+    // Arrange: several text items, none exceeding per-item limit, but total exceeds budget.
+    let budget = truncate::MODEL_FORMAT_MAX_BYTES;
+    let t1_len = (budget / 2).saturating_sub(10);
+    let t2_len = (budget / 2).saturating_sub(10);
+    let remaining_after_t1_t2 = budget.saturating_sub(t1_len + t2_len);
+    let t3_len = 50; // gets truncated to remaining_after_t1_t2
+    let t4_len = 5; // omitted
+    let t5_len = 7; // omitted
+
+    let t1 = "a".repeat(t1_len);
+    let t2 = "b".repeat(t2_len);
+    let t3 = "c".repeat(t3_len);
+    let t4 = "d".repeat(t4_len);
+    let t5 = "e".repeat(t5_len);
+
+    let item = ResponseItem::FunctionCallOutput {
+        call_id: "call-omit".to_string(),
+        output: FunctionCallOutputPayload {
+            content: "irrelevant".to_string(),
+            content_items: Some(vec![
+                FunctionCallOutputContentItem::InputText { text: t1 },
+                FunctionCallOutputContentItem::InputText { text: t2 },
+                FunctionCallOutputContentItem::InputImage {
+                    image_url: "img:mid".to_string(),
+                },
+                FunctionCallOutputContentItem::InputText { text: t3 },
+                FunctionCallOutputContentItem::InputText { text: t4 },
+                FunctionCallOutputContentItem::InputText { text: t5 },
+            ]),
+            success: Some(true),
+        },
+    };
+
+    let mut history = ContextManager::new();
+    history.record_items([&item]);
+    assert_eq!(history.items.len(), 1);
+    let json = serde_json::to_value(&history.items[0]).expect("serialize to json");
+
+    let output = json
+        .get("output")
+        .expect("output field")
+        .as_array()
+        .expect("array output");
+
+    // Expect: t1 (full), t2 (full), image, t3 (truncated), summary mentioning 2 omitted.
+    assert_eq!(output.len(), 5);
+
+    let first = output[0].as_object().expect("first obj");
+    assert_eq!(first.get("type").unwrap(), "input_text");
+    let first_text = first.get("text").unwrap().as_str().unwrap();
+    assert_eq!(first_text.len(), t1_len);
+
+    let second = output[1].as_object().expect("second obj");
+    assert_eq!(second.get("type").unwrap(), "input_text");
+    let second_text = second.get("text").unwrap().as_str().unwrap();
+    assert_eq!(second_text.len(), t2_len);
+
+    assert_eq!(
+        output[2],
+        serde_json::json!({"type": "input_image", "image_url": "img:mid"})
+    );
+
+    let fourth = output[3].as_object().expect("fourth obj");
+    assert_eq!(fourth.get("type").unwrap(), "input_text");
+    let fourth_text = fourth.get("text").unwrap().as_str().unwrap();
+    assert_eq!(fourth_text.len(), remaining_after_t1_t2);
+
+    let summary = output[4].as_object().expect("summary obj");
+    assert_eq!(summary.get("type").unwrap(), "input_text");
+    let summary_text = summary.get("text").unwrap().as_str().unwrap();
+    assert!(summary_text.contains("omitted 2 text items"));
+}
+
+//TODO(aibrahim): run CI in release mode.
+#[cfg(not(debug_assertions))]
+#[test]
+fn normalize_adds_missing_output_for_function_call() {
+    let items = vec![ResponseItem::FunctionCall {
+        id: None,
+        name: "do_it".to_string(),
+        arguments: "{}".to_string(),
+        call_id: "call-x".to_string(),
+    }];
+    let mut h = create_history_with_items(items);
+
+    h.normalize_history();
+
+    assert_eq!(
+        h.contents(),
+        vec![
+            ResponseItem::FunctionCall {
+                id: None,
+                name: "do_it".to_string(),
+                arguments: "{}".to_string(),
+                call_id: "call-x".to_string(),
+            },
+            ResponseItem::FunctionCallOutput {
+                call_id: "call-x".to_string(),
+                output: FunctionCallOutputPayload {
+                    content: "aborted".to_string(),
+                    ..Default::default()
+                },
+            },
+        ]
+    );
+}
+
+#[cfg(not(debug_assertions))]
+#[test]
+fn normalize_adds_missing_output_for_custom_tool_call() {
+    let items = vec![ResponseItem::CustomToolCall {
+        id: None,
+        status: None,
+        call_id: "tool-x".to_string(),
+        name: "custom".to_string(),
+        input: "{}".to_string(),
+    }];
+    let mut h = create_history_with_items(items);
+
+    h.normalize_history();
+
+    assert_eq!(
+        h.contents(),
+        vec![
+            ResponseItem::CustomToolCall {
+                id: None,
+                status: None,
+                call_id: "tool-x".to_string(),
+                name: "custom".to_string(),
+                input: "{}".to_string(),
+            },
+            ResponseItem::CustomToolCallOutput {
+                call_id: "tool-x".to_string(),
+                output: "aborted".to_string(),
+            },
+        ]
+    );
+}
+
+#[cfg(not(debug_assertions))]
+#[test]
+fn normalize_adds_missing_output_for_local_shell_call_with_id() {
+    let items = vec![ResponseItem::LocalShellCall {
+        id: None,
+        call_id: Some("shell-1".to_string()),
+        status: LocalShellStatus::Completed,
+        action: LocalShellAction::Exec(LocalShellExecAction {
+            command: vec!["echo".to_string(), "hi".to_string()],
+            timeout_ms: None,
+            working_directory: None,
+            env: None,
+            user: None,
+        }),
+    }];
+    let mut h = create_history_with_items(items);
+
+    h.normalize_history();
+
+    assert_eq!(
+        h.contents(),
+        vec![
+            ResponseItem::LocalShellCall {
+                id: None,
+                call_id: Some("shell-1".to_string()),
+                status: LocalShellStatus::Completed,
+                action: LocalShellAction::Exec(LocalShellExecAction {
+                    command: vec!["echo".to_string(), "hi".to_string()],
+                    timeout_ms: None,
+                    working_directory: None,
+                    env: None,
+                    user: None,
+                }),
+            },
+            ResponseItem::FunctionCallOutput {
+                call_id: "shell-1".to_string(),
+                output: FunctionCallOutputPayload {
+                    content: "aborted".to_string(),
+                    ..Default::default()
+                },
+            },
+        ]
+    );
+}
+
+#[cfg(not(debug_assertions))]
+#[test]
+fn normalize_removes_orphan_function_call_output() {
+    let items = vec![ResponseItem::FunctionCallOutput {
+        call_id: "orphan-1".to_string(),
+        output: FunctionCallOutputPayload {
+            content: "ok".to_string(),
+            ..Default::default()
+        },
+    }];
+    let mut h = create_history_with_items(items);
+
+    h.normalize_history();
+
+    assert_eq!(h.contents(), vec![]);
+}
+
+#[cfg(not(debug_assertions))]
+#[test]
+fn normalize_removes_orphan_custom_tool_call_output() {
+    let items = vec![ResponseItem::CustomToolCallOutput {
+        call_id: "orphan-2".to_string(),
+        output: "ok".to_string(),
+    }];
+    let mut h = create_history_with_items(items);
+
+    h.normalize_history();
+
+    assert_eq!(h.contents(), vec![]);
+}
+
+#[cfg(not(debug_assertions))]
+#[test]
+fn normalize_mixed_inserts_and_removals() {
+    let items = vec![
+        // Will get an inserted output
+        ResponseItem::FunctionCall {
+            id: None,
+            name: "f1".to_string(),
+            arguments: "{}".to_string(),
+            call_id: "c1".to_string(),
+        },
+        // Orphan output that should be removed
+        ResponseItem::FunctionCallOutput {
+            call_id: "c2".to_string(),
+            output: FunctionCallOutputPayload {
+                content: "ok".to_string(),
+                ..Default::default()
+            },
+        },
+        // Will get an inserted custom tool output
+        ResponseItem::CustomToolCall {
+            id: None,
+            status: None,
+            call_id: "t1".to_string(),
+            name: "tool".to_string(),
+            input: "{}".to_string(),
+        },
+        // Local shell call also gets an inserted function call output
+        ResponseItem::LocalShellCall {
+            id: None,
+            call_id: Some("s1".to_string()),
+            status: LocalShellStatus::Completed,
+            action: LocalShellAction::Exec(LocalShellExecAction {
+                command: vec!["echo".to_string()],
+                timeout_ms: None,
+                working_directory: None,
+                env: None,
+                user: None,
+            }),
+        },
+    ];
+    let mut h = create_history_with_items(items);
+
+    h.normalize_history();
+
+    assert_eq!(
+        h.contents(),
+        vec![
+            ResponseItem::FunctionCall {
+                id: None,
+                name: "f1".to_string(),
+                arguments: "{}".to_string(),
+                call_id: "c1".to_string(),
+            },
+            ResponseItem::FunctionCallOutput {
+                call_id: "c1".to_string(),
+                output: FunctionCallOutputPayload {
+                    content: "aborted".to_string(),
+                    ..Default::default()
+                },
+            },
+            ResponseItem::CustomToolCall {
+                id: None,
+                status: None,
+                call_id: "t1".to_string(),
+                name: "tool".to_string(),
+                input: "{}".to_string(),
+            },
+            ResponseItem::CustomToolCallOutput {
+                call_id: "t1".to_string(),
+                output: "aborted".to_string(),
+            },
+            ResponseItem::LocalShellCall {
+                id: None,
+                call_id: Some("s1".to_string()),
+                status: LocalShellStatus::Completed,
+                action: LocalShellAction::Exec(LocalShellExecAction {
+                    command: vec!["echo".to_string()],
+                    timeout_ms: None,
+                    working_directory: None,
+                    env: None,
+                    user: None,
+                }),
+            },
+            ResponseItem::FunctionCallOutput {
+                call_id: "s1".to_string(),
+                output: FunctionCallOutputPayload {
+                    content: "aborted".to_string(),
+                    ..Default::default()
+                },
+            },
+        ]
+    );
+}
+
+// In debug builds we panic on normalization errors instead of silently fixing them.
+#[cfg(debug_assertions)]
+#[test]
+#[should_panic]
+fn normalize_adds_missing_output_for_function_call_panics_in_debug() {
+    let items = vec![ResponseItem::FunctionCall {
+        id: None,
+        name: "do_it".to_string(),
+        arguments: "{}".to_string(),
+        call_id: "call-x".to_string(),
+    }];
+    let mut h = create_history_with_items(items);
+    h.normalize_history();
+}
+
+#[cfg(debug_assertions)]
+#[test]
+#[should_panic]
+fn normalize_adds_missing_output_for_custom_tool_call_panics_in_debug() {
+    let items = vec![ResponseItem::CustomToolCall {
+        id: None,
+        status: None,
+        call_id: "tool-x".to_string(),
+        name: "custom".to_string(),
+        input: "{}".to_string(),
+    }];
+    let mut h = create_history_with_items(items);
+    h.normalize_history();
+}
+
+#[cfg(debug_assertions)]
+#[test]
+#[should_panic]
+fn normalize_adds_missing_output_for_local_shell_call_with_id_panics_in_debug() {
+    let items = vec![ResponseItem::LocalShellCall {
+        id: None,
+        call_id: Some("shell-1".to_string()),
+        status: LocalShellStatus::Completed,
+        action: LocalShellAction::Exec(LocalShellExecAction {
+            command: vec!["echo".to_string(), "hi".to_string()],
+            timeout_ms: None,
+            working_directory: None,
+            env: None,
+            user: None,
+        }),
+    }];
+    let mut h = create_history_with_items(items);
+    h.normalize_history();
+}
+
+#[cfg(debug_assertions)]
+#[test]
+#[should_panic]
+fn normalize_removes_orphan_function_call_output_panics_in_debug() {
+    let items = vec![ResponseItem::FunctionCallOutput {
+        call_id: "orphan-1".to_string(),
+        output: FunctionCallOutputPayload {
+            content: "ok".to_string(),
+            ..Default::default()
+        },
+    }];
+    let mut h = create_history_with_items(items);
+    h.normalize_history();
+}
+
+#[cfg(debug_assertions)]
+#[test]
+#[should_panic]
+fn normalize_removes_orphan_custom_tool_call_output_panics_in_debug() {
+    let items = vec![ResponseItem::CustomToolCallOutput {
+        call_id: "orphan-2".to_string(),
+        output: "ok".to_string(),
+    }];
+    let mut h = create_history_with_items(items);
+    h.normalize_history();
+}
+
+#[cfg(debug_assertions)]
+#[test]
+#[should_panic]
+fn normalize_mixed_inserts_and_removals_panics_in_debug() {
+    let items = vec![
+        ResponseItem::FunctionCall {
+            id: None,
+            name: "f1".to_string(),
+            arguments: "{}".to_string(),
+            call_id: "c1".to_string(),
+        },
+        ResponseItem::FunctionCallOutput {
+            call_id: "c2".to_string(),
+            output: FunctionCallOutputPayload {
+                content: "ok".to_string(),
+                ..Default::default()
+            },
+        },
+        ResponseItem::CustomToolCall {
+            id: None,
+            status: None,
+            call_id: "t1".to_string(),
+            name: "tool".to_string(),
+            input: "{}".to_string(),
+        },
+        ResponseItem::LocalShellCall {
+            id: None,
+            call_id: Some("s1".to_string()),
+            status: LocalShellStatus::Completed,
+            action: LocalShellAction::Exec(LocalShellExecAction {
+                command: vec!["echo".to_string()],
+                timeout_ms: None,
+                working_directory: None,
+                env: None,
+                user: None,
+            }),
+        },
+    ];
+    let mut h = create_history_with_items(items);
+    h.normalize_history();
+}

codex-rs/core/src/context_manager/mod.rs

@@ -0,0 +1,8 @@
+mod history;
+mod normalize;
+mod truncate;
+
+pub(crate) use history::ContextManager;
+pub(crate) use truncate::MODEL_FORMAT_MAX_BYTES;
+pub(crate) use truncate::MODEL_FORMAT_MAX_LINES;
+pub(crate) use truncate::format_output_for_model_body;

codex-rs/core/src/context_manager/normalize.rs

@@ -0,0 +1,213 @@
+use std::collections::HashSet;
+
+use codex_protocol::models::FunctionCallOutputPayload;
+use codex_protocol::models::ResponseItem;
+
+use crate::util::error_or_panic;
+
+pub(crate) fn ensure_call_outputs_present(items: &mut Vec<ResponseItem>) {
+    // Collect synthetic outputs to insert immediately after their calls.
+    // Store the insertion position (index of call) alongside the item so
+    // we can insert in reverse order and avoid index shifting.
+    let mut missing_outputs_to_insert: Vec<(usize, ResponseItem)> = Vec::new();
+
+    for (idx, item) in items.iter().enumerate() {
+        match item {
+            ResponseItem::FunctionCall { call_id, .. } => {
+                let has_output = items.iter().any(|i| match i {
+                    ResponseItem::FunctionCallOutput {
+                        call_id: existing, ..
+                    } => existing == call_id,
+                    _ => false,
+                });
+
+                if !has_output {
+                    error_or_panic(format!(
+                        "Function call output is missing for call id: {call_id}"
+                    ));
+                    missing_outputs_to_insert.push((
+                        idx,
+                        ResponseItem::FunctionCallOutput {
+                            call_id: call_id.clone(),
+                            output: FunctionCallOutputPayload {
+                                content: "aborted".to_string(),
+                                ..Default::default()
+                            },
+                        },
+                    ));
+                }
+            }
+            ResponseItem::CustomToolCall { call_id, .. } => {
+                let has_output = items.iter().any(|i| match i {
+                    ResponseItem::CustomToolCallOutput {
+                        call_id: existing, ..
+                    } => existing == call_id,
+                    _ => false,
+                });
+
+                if !has_output {
+                    error_or_panic(format!(
+                        "Custom tool call output is missing for call id: {call_id}"
+                    ));
+                    missing_outputs_to_insert.push((
+                        idx,
+                        ResponseItem::CustomToolCallOutput {
+                            call_id: call_id.clone(),
+                            output: "aborted".to_string(),
+                        },
+                    ));
+                }
+            }
+            // LocalShellCall is represented in upstream streams by a FunctionCallOutput
+            ResponseItem::LocalShellCall { call_id, .. } => {
+                if let Some(call_id) = call_id.as_ref() {
+                    let has_output = items.iter().any(|i| match i {
+                        ResponseItem::FunctionCallOutput {
+                            call_id: existing, ..
+                        } => existing == call_id,
+                        _ => false,
+                    });
+
+                    if !has_output {
+                        error_or_panic(format!(
+                            "Local shell call output is missing for call id: {call_id}"
+                        ));
+                        missing_outputs_to_insert.push((
+                            idx,
+                            ResponseItem::FunctionCallOutput {
+                                call_id: call_id.clone(),
+                                output: FunctionCallOutputPayload {
+                                    content: "aborted".to_string(),
+                                    ..Default::default()
+                                },
+                            },
+                        ));
+                    }
+                }
+            }
+            _ => {}
+        }
+    }
+
+    // Insert synthetic outputs in reverse index order to avoid re-indexing.
+    for (idx, output_item) in missing_outputs_to_insert.into_iter().rev() {
+        items.insert(idx + 1, output_item);
+    }
+}
+
+pub(crate) fn remove_orphan_outputs(items: &mut Vec<ResponseItem>) {
+    let function_call_ids: HashSet<String> = items
+        .iter()
+        .filter_map(|i| match i {
+            ResponseItem::FunctionCall { call_id, .. } => Some(call_id.clone()),
+            _ => None,
+        })
+        .collect();
+
+    let local_shell_call_ids: HashSet<String> = items
+        .iter()
+        .filter_map(|i| match i {
+            ResponseItem::LocalShellCall {
+                call_id: Some(call_id),
+                ..
+            } => Some(call_id.clone()),
+            _ => None,
+        })
+        .collect();
+
+    let custom_tool_call_ids: HashSet<String> = items
+        .iter()
+        .filter_map(|i| match i {
+            ResponseItem::CustomToolCall { call_id, .. } => Some(call_id.clone()),
+            _ => None,
+        })
+        .collect();
+
+    items.retain(|item| match item {
+        ResponseItem::FunctionCallOutput { call_id, .. } => {
+            let has_match =
+                function_call_ids.contains(call_id) || local_shell_call_ids.contains(call_id);
+            if !has_match {
+                error_or_panic(format!(
+                    "Orphan function call output for call id: {call_id}"
+                ));
+            }
+            has_match
+        }
+        ResponseItem::CustomToolCallOutput { call_id, .. } => {
+            let has_match = custom_tool_call_ids.contains(call_id);
+            if !has_match {
+                error_or_panic(format!(
+                    "Orphan custom tool call output for call id: {call_id}"
+                ));
+            }
+            has_match
+        }
+        _ => true,
+    });
+}
+
+pub(crate) fn remove_corresponding_for(items: &mut Vec<ResponseItem>, item: &ResponseItem) {
+    match item {
+        ResponseItem::FunctionCall { call_id, .. } => {
+            remove_first_matching(items, |i| {
+                matches!(
+                    i,
+                    ResponseItem::FunctionCallOutput {
+                        call_id: existing, ..
+                    } if existing == call_id
+                )
+            });
+        }
+        ResponseItem::FunctionCallOutput { call_id, .. } => {
+            if let Some(pos) = items.iter().position(|i| {
+                matches!(i, ResponseItem::FunctionCall { call_id: existing, .. } if existing == call_id)
+            }) {
+                items.remove(pos);
+            } else if let Some(pos) = items.iter().position(|i| {
+                matches!(i, ResponseItem::LocalShellCall { call_id: Some(existing), .. } if existing == call_id)
+            }) {
+                items.remove(pos);
+            }
+        }
+        ResponseItem::CustomToolCall { call_id, .. } => {
+            remove_first_matching(items, |i| {
+                matches!(
+                    i,
+                    ResponseItem::CustomToolCallOutput {
+                        call_id: existing, ..
+                    } if existing == call_id
+                )
+            });
+        }
+        ResponseItem::CustomToolCallOutput { call_id, .. } => {
+            remove_first_matching(
+                items,
+                |i| matches!(i, ResponseItem::CustomToolCall { call_id: existing, .. } if existing == call_id),
+            );
+        }
+        ResponseItem::LocalShellCall {
+            call_id: Some(call_id),
+            ..
+        } => {
+            remove_first_matching(items, |i| {
+                matches!(
+                    i,
+                    ResponseItem::FunctionCallOutput {
+                        call_id: existing, ..
+                    } if existing == call_id
+                )
+            });
+        }
+        _ => {}
+    }
+}
+
+fn remove_first_matching<F>(items: &mut Vec<ResponseItem>, predicate: F)
+where
+    F: Fn(&ResponseItem) -> bool,
+{
+    if let Some(pos) = items.iter().position(predicate) {
+        items.remove(pos);
+    }
+}

codex-rs/core/src/context_manager/truncate.rs

@@ -0,0 +1,148 @@
+use codex_protocol::models::FunctionCallOutputContentItem;
+use codex_utils_string::take_bytes_at_char_boundary;
+use codex_utils_string::take_last_bytes_at_char_boundary;
+
+use crate::util::error_or_panic;
+
+// Model-formatting limits: clients get full streams; only content sent to the model is truncated.
+pub const MODEL_FORMAT_MAX_BYTES: usize = 10 * 1024; // 10 KiB
+pub const MODEL_FORMAT_MAX_LINES: usize = 256; // lines
+
+pub(crate) fn globally_truncate_function_output_items(
+    items: &[FunctionCallOutputContentItem],
+) -> Vec<FunctionCallOutputContentItem> {
+    let mut out: Vec<FunctionCallOutputContentItem> = Vec::with_capacity(items.len());
+    let mut remaining = MODEL_FORMAT_MAX_BYTES;
+    let mut omitted_text_items = 0usize;
+
+    for it in items {
+        match it {
+            FunctionCallOutputContentItem::InputText { text } => {
+                if remaining == 0 {
+                    omitted_text_items += 1;
+                    continue;
+                }
+
+                let len = text.len();
+                if len <= remaining {
+                    out.push(FunctionCallOutputContentItem::InputText { text: text.clone() });
+                    remaining -= len;
+                } else {
+                    let slice = take_bytes_at_char_boundary(text, remaining);
+                    if !slice.is_empty() {
+                        out.push(FunctionCallOutputContentItem::InputText {
+                            text: slice.to_string(),
+                        });
+                    }
+                    remaining = 0;
+                }
+            }
+            // todo(aibrahim): handle input images; resize
+            FunctionCallOutputContentItem::InputImage { image_url } => {
+                out.push(FunctionCallOutputContentItem::InputImage {
+                    image_url: image_url.clone(),
+                });
+            }
+        }
+    }
+
+    if omitted_text_items > 0 {
+        out.push(FunctionCallOutputContentItem::InputText {
+            text: format!("[omitted {omitted_text_items} text items ...]"),
+        });
+    }
+
+    out
+}
+
+pub(crate) fn format_output_for_model_body(
+    content: &str,
+    limit_bytes: usize,
+    limit_lines: usize,
+) -> String {
+    // Head+tail truncation for the model: show the beginning and end with an elision.
+    // Clients still receive full streams; only this formatted summary is capped.
+    let total_lines = content.lines().count();
+    if content.len() <= limit_bytes && total_lines <= limit_lines {
+        return content.to_string();
+    }
+    let output = truncate_formatted_exec_output(content, total_lines, limit_bytes, limit_lines);
+    format!("Total output lines: {total_lines}\n\n{output}")
+}
+
+fn truncate_formatted_exec_output(
+    content: &str,
+    total_lines: usize,
+    limit_bytes: usize,
+    limit_lines: usize,
+) -> String {
+    debug_panic_on_double_truncation(content);
+    let head_lines: usize = limit_lines / 2;
+    let tail_lines: usize = limit_lines - head_lines; // 128
+    let head_bytes: usize = limit_bytes / 2;
+    let segments: Vec<&str> = content.split_inclusive('\n').collect();
+    let head_take = head_lines.min(segments.len());
+    let tail_take = tail_lines.min(segments.len().saturating_sub(head_take));
+    let omitted = segments.len().saturating_sub(head_take + tail_take);
+
+    let head_slice_end: usize = segments
+        .iter()
+        .take(head_take)
+        .map(|segment| segment.len())
+        .sum();
+    let tail_slice_start: usize = if tail_take == 0 {
+        content.len()
+    } else {
+        content.len()
+            - segments
+                .iter()
+                .rev()
+                .take(tail_take)
+                .map(|segment| segment.len())
+                .sum::<usize>()
+    };
+    let head_slice = &content[..head_slice_end];
+    let tail_slice = &content[tail_slice_start..];
+    let truncated_by_bytes = content.len() > limit_bytes;
+    // this is a bit wrong. We are counting metadata lines and not just shell output lines.
+    let marker = if omitted > 0 {
+        Some(format!(
+            "\n[... omitted {omitted} of {total_lines} lines ...]\n\n"
+        ))
+    } else if truncated_by_bytes {
+        Some(format!(
+            "\n[... output truncated to fit {limit_bytes} bytes ...]\n\n"
+        ))
+    } else {
+        None
+    };
+
+    let marker_len = marker.as_ref().map_or(0, String::len);
+    let base_head_budget = head_bytes.min(limit_bytes);
+    let head_budget = base_head_budget.min(limit_bytes.saturating_sub(marker_len));
+    let head_part = take_bytes_at_char_boundary(head_slice, head_budget);
+    let mut result = String::with_capacity(limit_bytes.min(content.len()));
+
+    result.push_str(head_part);
+    if let Some(marker_text) = marker.as_ref() {
+        result.push_str(marker_text);
+    }
+
+    let remaining = limit_bytes.saturating_sub(result.len());
+    if remaining == 0 {
+        return result;
+    }
+
+    let tail_part = take_last_bytes_at_char_boundary(tail_slice, remaining);
+    result.push_str(tail_part);
+
+    result
+}
+
+fn debug_panic_on_double_truncation(content: &str) {
+    if content.contains("Total output lines:") && content.contains("omitted") {
+        error_or_panic(format!(
+            "FunctionCallOutput content was already truncated before ContextManager::record_items; this would cause double truncation {content}"
+        ));
+    }
+}

codex-rs/core/src/custom_prompts.rs

@@ -32,12 +32,11 @@ pub async fn discover_prompts_in_excluding(
 
     while let Ok(Some(entry)) = entries.next_entry().await {
         let path = entry.path();
-        let is_file = entry
-            .file_type()
+        let is_file_like = fs::metadata(&path)
             .await
-            .map(|ft| ft.is_file())
+            .map(|m| m.is_file())
             .unwrap_or(false);
-        if !is_file {
+        if !is_file_like {
             continue;
         }
         // Only include Markdown files with a .md extension.
@@ -197,6 +196,25 @@ mod tests {
         assert_eq!(names, vec!["good"]);
     }
 
+    #[tokio::test]
+    #[cfg(unix)]
+    async fn discovers_symlinked_md_files() {
+        let tmp = tempdir().expect("create TempDir");
+        let dir = tmp.path();
+
+        // Create a real file
+        fs::write(dir.join("real.md"), b"real content").unwrap();
+
+        // Create a symlink to the real file
+        std::os::unix::fs::symlink(dir.join("real.md"), dir.join("link.md")).unwrap();
+
+        let found = discover_prompts_in(dir).await;
+        let names: Vec<String> = found.into_iter().map(|e| e.name).collect();
+
+        // Both real and link should be discovered, sorted alphabetically
+        assert_eq!(names, vec!["link", "real"]);
+    }
+
     #[tokio::test]
     async fn parses_frontmatter_and_strips_from_body() {
         let tmp = tempdir().expect("create TempDir");

codex-rs/core/src/environment_context.rs

@@ -329,7 +329,6 @@ mod tests {
             Some(workspace_write_policy(vec!["/repo"], false)),
             Some(Shell::Bash(BashShell {
                 shell_path: "/bin/bash".into(),
-                bashrc_path: "/home/user/.bashrc".into(),
             })),
         );
         let context2 = EnvironmentContext::new(
@@ -338,7 +337,6 @@ mod tests {
             Some(workspace_write_policy(vec!["/repo"], false)),
             Some(Shell::Zsh(ZshShell {
                 shell_path: "/bin/zsh".into(),
-                zshrc_path: "/home/user/.zshrc".into(),
             })),
         );
 

codex-rs/core/src/error.rs

@@ -109,6 +109,9 @@ pub enum CodexErr {
     #[error("{0}")]
     ConnectionFailed(ConnectionFailedError),
 
+    #[error("Quota exceeded. Check your plan and billing details.")]
+    QuotaExceeded,
+
     #[error(
         "To use Codex with your ChatGPT plan, upgrade to Plus: https://openai.com/chatgpt/pricing."
     )]
@@ -135,6 +138,9 @@ pub enum CodexErr {
     #[error("unsupported operation: {0}")]
     UnsupportedOperation(String),
 
+    #[error("{0}")]
+    RefreshTokenFailed(RefreshTokenFailedError),
+
     #[error("Fatal error: {0}")]
     Fatal(String),
 
@@ -201,6 +207,30 @@ impl std::fmt::Display for ResponseStreamFailed {
     }
 }
 
+#[derive(Debug, Clone, PartialEq, Eq, Error)]
+#[error("{message}")]
+pub struct RefreshTokenFailedError {
+    pub reason: RefreshTokenFailedReason,
+    pub message: String,
+}
+
+impl RefreshTokenFailedError {
+    pub fn new(reason: RefreshTokenFailedReason, message: impl Into<String>) -> Self {
+        Self {
+            reason,
+            message: message.into(),
+        }
+    }
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum RefreshTokenFailedReason {
+    Expired,
+    Exhausted,
+    Revoked,
+    Other,
+}
+
 #[derive(Debug)]
 pub struct UnexpectedResponseError {
     pub status: StatusCode,
@@ -208,18 +238,44 @@ pub struct UnexpectedResponseError {
     pub request_id: Option<String>,
 }
 
+const CLOUDFLARE_BLOCKED_MESSAGE: &str =
+    "Access blocked by Cloudflare. This usually happens when connecting from a restricted region";
+
+impl UnexpectedResponseError {
+    fn friendly_message(&self) -> Option<String> {
+        if self.status != StatusCode::FORBIDDEN {
+            return None;
+        }
+
+        if !self.body.contains("Cloudflare") || !self.body.contains("blocked") {
+            return None;
+        }
+
+        let mut message = format!("{CLOUDFLARE_BLOCKED_MESSAGE} (status {})", self.status);
+        if let Some(id) = &self.request_id {
+            message.push_str(&format!(", request id: {id}"));
+        }
+
+        Some(message)
+    }
+}
+
 impl std::fmt::Display for UnexpectedResponseError {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        write!(
-            f,
-            "unexpected status {}: {}{}",
-            self.status,
-            self.body,
-            self.request_id
-                .as_ref()
-                .map(|id| format!(", request id: {id}"))
-                .unwrap_or_default()
-        )
+        if let Some(friendly) = self.friendly_message() {
+            write!(f, "{friendly}")
+        } else {
+            write!(
+                f,
+                "unexpected status {}: {}{}",
+                self.status,
+                self.body,
+                self.request_id
+                    .as_ref()
+                    .map(|id| format!(", request id: {id}"))
+                    .unwrap_or_default()
+            )
+        }
     }
 }
 
@@ -255,7 +311,7 @@ impl std::fmt::Display for UsageLimitReachedError {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
         let message = match self.plan_type.as_ref() {
             Some(PlanType::Known(KnownPlan::Plus)) => format!(
-                "You've hit your usage limit. Upgrade to Pro (https://openai.com/chatgpt/pricing), visit chatgpt.com/codex/settings/usage to purchase more credits{}",
+                "You've hit your usage limit. Upgrade to Pro (https://openai.com/chatgpt/pricing), visit https://chatgpt.com/codex/settings/usage to purchase more credits{}",
                 retry_suffix_after_or(self.resets_at.as_ref())
             ),
             Some(PlanType::Known(KnownPlan::Team)) | Some(PlanType::Known(KnownPlan::Business)) => {
@@ -269,7 +325,7 @@ impl std::fmt::Display for UsageLimitReachedError {
                     .to_string()
             }
             Some(PlanType::Known(KnownPlan::Pro)) => format!(
-                "You've hit your usage limit. Visit chatgpt.com/codex/settings/usage to purchase more credits{}",
+                "You've hit your usage limit. Visit https://chatgpt.com/codex/settings/usage to purchase more credits{}",
                 retry_suffix_after_or(self.resets_at.as_ref())
             ),
             Some(PlanType::Known(KnownPlan::Enterprise))
@@ -460,7 +516,7 @@ mod tests {
         };
         assert_eq!(
             err.to_string(),
-            "You've hit your usage limit. Upgrade to Pro (https://openai.com/chatgpt/pricing), visit chatgpt.com/codex/settings/usage to purchase more credits or try again later."
+            "You've hit your usage limit. Upgrade to Pro (https://openai.com/chatgpt/pricing), visit https://chatgpt.com/codex/settings/usage to purchase more credits or try again later."
         );
     }
 
@@ -613,7 +669,7 @@ mod tests {
                 rate_limits: Some(rate_limit_snapshot()),
             };
             let expected = format!(
-                "You've hit your usage limit. Visit chatgpt.com/codex/settings/usage to purchase more credits or try again at {expected_time}."
+                "You've hit your usage limit. Visit https://chatgpt.com/codex/settings/usage to purchase more credits or try again at {expected_time}."
             );
             assert_eq!(err.to_string(), expected);
         });
@@ -635,6 +691,35 @@ mod tests {
         });
     }
 
+    #[test]
+    fn unexpected_status_cloudflare_html_is_simplified() {
+        let err = UnexpectedResponseError {
+            status: StatusCode::FORBIDDEN,
+            body: "<html><body>Cloudflare error: Sorry, you have been blocked</body></html>"
+                .to_string(),
+            request_id: Some("ray-id".to_string()),
+        };
+        let status = StatusCode::FORBIDDEN.to_string();
+        assert_eq!(
+            err.to_string(),
+            format!("{CLOUDFLARE_BLOCKED_MESSAGE} (status {status}), request id: ray-id")
+        );
+    }
+
+    #[test]
+    fn unexpected_status_non_html_is_unchanged() {
+        let err = UnexpectedResponseError {
+            status: StatusCode::FORBIDDEN,
+            body: "plain text error".to_string(),
+            request_id: None,
+        };
+        let status = StatusCode::FORBIDDEN.to_string();
+        assert_eq!(
+            err.to_string(),
+            format!("unexpected status {status}: plain text error")
+        );
+    }
+
     #[test]
     fn usage_limit_reached_includes_hours_and_minutes() {
         let base = Utc.with_ymd_and_hms(2024, 1, 1, 0, 0, 0).unwrap();
@@ -647,7 +732,7 @@ mod tests {
                 rate_limits: Some(rate_limit_snapshot()),
             };
             let expected = format!(
-                "You've hit your usage limit. Upgrade to Pro (https://openai.com/chatgpt/pricing), visit chatgpt.com/codex/settings/usage to purchase more credits or try again at {expected_time}."
+                "You've hit your usage limit. Upgrade to Pro (https://openai.com/chatgpt/pricing), visit https://chatgpt.com/codex/settings/usage to purchase more credits or try again at {expected_time}."
             );
             assert_eq!(err.to_string(), expected);
         });

codex-rs/core/src/event_mapping.rs

@@ -14,6 +14,7 @@ use tracing::warn;
 use uuid::Uuid;
 
 use crate::user_instructions::UserInstructions;
+use crate::user_shell_command::is_user_shell_command_text;
 
 fn is_session_prefix(text: &str) -> bool {
     let trimmed = text.trim_start();
@@ -31,7 +32,7 @@ fn parse_user_message(message: &[ContentItem]) -> Option<UserMessageItem> {
     for content_item in message.iter() {
         match content_item {
             ContentItem::InputText { text } => {
-                if is_session_prefix(text) {
+                if is_session_prefix(text) || is_user_shell_command_text(text) {
                     return None;
                 }
                 content.push(UserInput::Text { text: text.clone() });
@@ -197,7 +198,14 @@ mod tests {
                     text: "# AGENTS.md instructions for test_directory\n\n<INSTRUCTIONS>\ntest_text\n</INSTRUCTIONS>".to_string(),
                 }],
             },
-        ];
+        ResponseItem::Message {
+            id: None,
+            role: "user".to_string(),
+            content: vec![ContentItem::InputText {
+                text: "<user_shell_command>echo 42</user_shell_command>".to_string(),
+            }],
+        },
+    ];
 
         for item in items {
             let turn_item = parse_turn_item(&item);

codex-rs/core/src/exec.rs

@@ -171,6 +171,7 @@ async fn exec_windows_sandbox(
     params: ExecParams,
     sandbox_policy: &SandboxPolicy,
 ) -> Result<RawExecToolCallOutput> {
+    use crate::config::find_codex_home;
     use codex_windows_sandbox::run_windows_sandbox_capture;
 
     let ExecParams {
@@ -188,8 +189,17 @@ async fn exec_windows_sandbox(
     };
 
     let sandbox_cwd = cwd.clone();
+    let logs_base_dir = find_codex_home().ok();
     let spawn_res = tokio::task::spawn_blocking(move || {
-        run_windows_sandbox_capture(policy_str, &sandbox_cwd, command, &cwd, env, timeout_ms)
+        run_windows_sandbox_capture(
+            policy_str,
+            &sandbox_cwd,
+            command,
+            &cwd,
+            env,
+            timeout_ms,
+            logs_base_dir.as_deref(),
+        )
     })
     .await;
 
@@ -303,6 +313,10 @@ pub(crate) mod errors {
                 SandboxTransformError::MissingLinuxSandboxExecutable => {
                     CodexErr::LandlockSandboxExecutableNotProvided
                 }
+                #[cfg(not(target_os = "macos"))]
+                SandboxTransformError::SeatbeltUnavailable => CodexErr::UnsupportedOperation(
+                    "seatbelt sandbox is only available on macOS".to_string(),
+                ),
             }
         }
     }
@@ -504,6 +518,7 @@ async fn consume_truncated_output(
                 }
                 Err(_) => {
                     // timeout
+                    kill_child_process_group(&mut child)?;
                     child.start_kill()?;
                     // Debatable whether `child.wait().await` should be called here.
                     (synthetic_exit_status(EXIT_CODE_SIGNAL_BASE + TIMEOUT_CODE), true)
@@ -511,13 +526,58 @@ async fn consume_truncated_output(
             }
         }
         _ = tokio::signal::ctrl_c() => {
+            kill_child_process_group(&mut child)?;
             child.start_kill()?;
             (synthetic_exit_status(EXIT_CODE_SIGNAL_BASE + SIGKILL_CODE), false)
         }
     };
 
-    let stdout = stdout_handle.await??;
-    let stderr = stderr_handle.await??;
+    // Wait for the stdout/stderr collection tasks but guard against them
+    // hanging forever. In the normal case, both pipes are closed once the child
+    // terminates so the tasks exit quickly. However, if the child process
+    // spawned grandchildren that inherited its stdout/stderr file descriptors
+    // those pipes may stay open after we `kill` the direct child on timeout.
+    // That would cause the `read_capped` tasks to block on `read()`
+    // indefinitely, effectively hanging the whole agent.
+
+    const IO_DRAIN_TIMEOUT_MS: u64 = 2_000; // 2 s should be plenty for local pipes
+
+    // We need mutable bindings so we can `abort()` them on timeout.
+    use tokio::task::JoinHandle;
+
+    async fn await_with_timeout(
+        handle: &mut JoinHandle<std::io::Result<StreamOutput<Vec<u8>>>>,
+        timeout: Duration,
+    ) -> std::io::Result<StreamOutput<Vec<u8>>> {
+        match tokio::time::timeout(timeout, &mut *handle).await {
+            Ok(join_res) => match join_res {
+                Ok(io_res) => io_res,
+                Err(join_err) => Err(std::io::Error::other(join_err)),
+            },
+            Err(_elapsed) => {
+                // Timeout: abort the task to avoid hanging on open pipes.
+                handle.abort();
+                Ok(StreamOutput {
+                    text: Vec::new(),
+                    truncated_after_lines: None,
+                })
+            }
+        }
+    }
+
+    let mut stdout_handle = stdout_handle;
+    let mut stderr_handle = stderr_handle;
+
+    let stdout = await_with_timeout(
+        &mut stdout_handle,
+        Duration::from_millis(IO_DRAIN_TIMEOUT_MS),
+    )
+    .await?;
+    let stderr = await_with_timeout(
+        &mut stderr_handle,
+        Duration::from_millis(IO_DRAIN_TIMEOUT_MS),
+    )
+    .await?;
 
     drop(agg_tx);
 
@@ -607,6 +667,38 @@ fn synthetic_exit_status(code: i32) -> ExitStatus {
     std::process::ExitStatus::from_raw(code as u32)
 }
 
+#[cfg(unix)]
+fn kill_child_process_group(child: &mut Child) -> io::Result<()> {
+    use std::io::ErrorKind;
+
+    if let Some(pid) = child.id() {
+        let pid = pid as libc::pid_t;
+        let pgid = unsafe { libc::getpgid(pid) };
+        if pgid == -1 {
+            let err = std::io::Error::last_os_error();
+            if err.kind() != ErrorKind::NotFound {
+                return Err(err);
+            }
+            return Ok(());
+        }
+
+        let result = unsafe { libc::killpg(pgid, libc::SIGKILL) };
+        if result == -1 {
+            let err = std::io::Error::last_os_error();
+            if err.kind() != ErrorKind::NotFound {
+                return Err(err);
+            }
+        }
+    }
+
+    Ok(())
+}
+
+#[cfg(not(unix))]
+fn kill_child_process_group(_: &mut Child) -> io::Result<()> {
+    Ok(())
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -679,4 +771,51 @@ mod tests {
         let output = make_exec_output(exit_code, "", "", "");
         assert!(is_likely_sandbox_denied(SandboxType::LinuxSeccomp, &output));
     }
+
+    #[cfg(unix)]
+    #[tokio::test]
+    async fn kill_child_process_group_kills_grandchildren_on_timeout() -> Result<()> {
+        let command = vec![
+            "/bin/bash".to_string(),
+            "-c".to_string(),
+            "sleep 60 & echo $!; sleep 60".to_string(),
+        ];
+        let env: HashMap<String, String> = std::env::vars().collect();
+        let params = ExecParams {
+            command,
+            cwd: std::env::current_dir()?,
+            timeout_ms: Some(500),
+            env,
+            with_escalated_permissions: None,
+            justification: None,
+            arg0: None,
+        };
+
+        let output = exec(params, SandboxType::None, &SandboxPolicy::ReadOnly, None).await?;
+        assert!(output.timed_out);
+
+        let stdout = output.stdout.from_utf8_lossy().text;
+        let pid_line = stdout.lines().next().unwrap_or("").trim();
+        let pid: i32 = pid_line.parse().map_err(|error| {
+            io::Error::new(
+                io::ErrorKind::InvalidData,
+                format!("Failed to parse pid from stdout '{pid_line}': {error}"),
+            )
+        })?;
+
+        let mut killed = false;
+        for _ in 0..20 {
+            // Use kill(pid, 0) to check if the process is alive.
+            if unsafe { libc::kill(pid, 0) } == -1
+                && let Some(libc::ESRCH) = std::io::Error::last_os_error().raw_os_error()
+            {
+                killed = true;
+                break;
+            }
+            tokio::time::sleep(Duration::from_millis(100)).await;
+        }
+
+        assert!(killed, "grandchild process with pid {pid} is still alive");
+        Ok(())
+    }
 }

codex-rs/core/src/features.rs

@@ -29,8 +29,9 @@ pub enum Stage {
 pub enum Feature {
     /// Use the single unified PTY-backed exec tool.
     UnifiedExec,
-    /// Use the streamable exec-command/write-stdin tool pair.
-    StreamableShell,
+    /// Use the shell command tool that takes `command` as a single string of
+    /// shell instead of an array of args passed to `execvp(3)`.
+    ShellCommandTool,
     /// Enable experimental RMCP features such as OAuth login.
     RmcpClient,
     /// Include the freeform apply_patch tool.
@@ -118,8 +119,9 @@ impl Features {
         self.enabled.contains(&f)
     }
 
-    pub fn enable(&mut self, f: Feature) {
+    pub fn enable(&mut self, f: Feature) -> &mut Self {
         self.enabled.insert(f);
+        self
     }
 
     pub fn disable(&mut self, f: Feature) -> &mut Self {
@@ -178,7 +180,6 @@ impl Features {
         let base_legacy = LegacyFeatureToggles {
             experimental_sandbox_command_assessment: cfg.experimental_sandbox_command_assessment,
             experimental_use_freeform_apply_patch: cfg.experimental_use_freeform_apply_patch,
-            experimental_use_exec_command_tool: cfg.experimental_use_exec_command_tool,
             experimental_use_unified_exec_tool: cfg.experimental_use_unified_exec_tool,
             experimental_use_rmcp_client: cfg.experimental_use_rmcp_client,
             tools_web_search: cfg.tools.as_ref().and_then(|t| t.web_search),
@@ -197,7 +198,7 @@ impl Features {
                 .experimental_sandbox_command_assessment,
             experimental_use_freeform_apply_patch: config_profile
                 .experimental_use_freeform_apply_patch,
-            experimental_use_exec_command_tool: config_profile.experimental_use_exec_command_tool,
+
             experimental_use_unified_exec_tool: config_profile.experimental_use_unified_exec_tool,
             experimental_use_rmcp_client: config_profile.experimental_use_rmcp_client,
             tools_web_search: config_profile.tools_web_search,
@@ -253,8 +254,8 @@ pub const FEATURES: &[FeatureSpec] = &[
         default_enabled: false,
     },
     FeatureSpec {
-        id: Feature::StreamableShell,
-        key: "streamable_shell",
+        id: Feature::ShellCommandTool,
+        key: "shell_command_tool",
         stage: Stage::Experimental,
         default_enabled: false,
     },
@@ -292,7 +293,7 @@ pub const FEATURES: &[FeatureSpec] = &[
         id: Feature::GhostCommit,
         key: "ghost_commit",
         stage: Stage::Experimental,
-        default_enabled: false,
+        default_enabled: true,
     },
     FeatureSpec {
         id: Feature::WindowsSandbox,

codex-rs/core/src/features/legacy.rs

@@ -17,10 +17,6 @@ const ALIASES: &[Alias] = &[
         legacy_key: "experimental_use_unified_exec_tool",
         feature: Feature::UnifiedExec,
     },
-    Alias {
-        legacy_key: "experimental_use_exec_command_tool",
-        feature: Feature::StreamableShell,
-    },
     Alias {
         legacy_key: "experimental_use_rmcp_client",
         feature: Feature::RmcpClient,
@@ -54,7 +50,6 @@ pub struct LegacyFeatureToggles {
     pub include_apply_patch_tool: Option<bool>,
     pub experimental_sandbox_command_assessment: Option<bool>,
     pub experimental_use_freeform_apply_patch: Option<bool>,
-    pub experimental_use_exec_command_tool: Option<bool>,
     pub experimental_use_unified_exec_tool: Option<bool>,
     pub experimental_use_rmcp_client: Option<bool>,
     pub tools_web_search: Option<bool>,
@@ -81,12 +76,6 @@ impl LegacyFeatureToggles {
             self.experimental_use_freeform_apply_patch,
             "experimental_use_freeform_apply_patch",
         );
-        set_if_some(
-            features,
-            Feature::StreamableShell,
-            self.experimental_use_exec_command_tool,
-            "experimental_use_exec_command_tool",
-        );
         set_if_some(
             features,
             Feature::UnifiedExec,
@@ -123,7 +112,7 @@ fn set_if_some(
     if let Some(enabled) = maybe_value {
         set_feature(features, feature, enabled);
         log_alias(alias_key, feature);
-        features.record_legacy_usage_force(alias_key, feature);
+        features.record_legacy_usage(alias_key, feature);
     }
 }
 

codex-rs/core/src/lib.rs

@@ -18,7 +18,7 @@ mod codex_delegate;
 mod command_safety;
 pub mod config;
 pub mod config_loader;
-mod conversation_history;
+mod context_manager;
 pub mod custom_prompts;
 mod environment_context;
 pub mod error;
@@ -75,16 +75,19 @@ pub use rollout::find_conversation_path_by_id_str;
 pub use rollout::list::ConversationItem;
 pub use rollout::list::ConversationsPage;
 pub use rollout::list::Cursor;
+pub use rollout::list::parse_cursor;
 pub use rollout::list::read_head_for_summary;
 mod function_tool;
 mod state;
 mod tasks;
 mod user_notification;
+mod user_shell_command;
 pub mod util;
 
 pub use apply_patch::CODEX_APPLY_PATCH_ARG1;
 pub use command_safety::is_safe_command;
 pub use safety::get_platform_sandbox;
+pub use safety::set_windows_sandbox_enabled;
 // Re-export the protocol types from the standalone `codex-protocol` crate so existing
 // `codex_core::protocol::...` references continue to work across the workspace.
 pub use codex_protocol::protocol;
@@ -97,11 +100,12 @@ pub use client_common::Prompt;
 pub use client_common::REVIEW_PROMPT;
 pub use client_common::ResponseEvent;
 pub use client_common::ResponseStream;
-pub use codex::compact::content_items_to_text;
 pub use codex_protocol::models::ContentItem;
 pub use codex_protocol::models::LocalShellAction;
 pub use codex_protocol::models::LocalShellExecAction;
 pub use codex_protocol::models::LocalShellStatus;
 pub use codex_protocol::models::ResponseItem;
+pub use compact::content_items_to_text;
 pub use event_mapping::parse_turn_item;
+pub mod compact;
 pub mod otel_init;

codex-rs/core/src/model_family.rs

@@ -1,10 +1,16 @@
+use codex_protocol::config_types::ReasoningEffort;
+use codex_protocol::config_types::Verbosity;
+
 use crate::config::types::ReasoningSummaryFormat;
 use crate::tools::handlers::apply_patch::ApplyPatchToolType;
+use crate::tools::spec::ConfigShellToolType;
 
 /// The `instructions` field in the payload sent to a model should always start
 /// with this content.
 const BASE_INSTRUCTIONS: &str = include_str!("../prompt.md");
+
 const GPT_5_CODEX_INSTRUCTIONS: &str = include_str!("../gpt_5_codex_prompt.md");
+const GPT_5_1_INSTRUCTIONS: &str = include_str!("../gpt_5_1_prompt.md");
 
 /// A model family is a group of models that share certain characteristics.
 #[derive(Debug, Clone, PartialEq, Eq, Hash)]
@@ -26,15 +32,12 @@ pub struct ModelFamily {
     // `summary` is optional).
     pub supports_reasoning_summaries: bool,
 
+    // The reasoning effort to use for this model family when none is explicitly chosen.
+    pub default_reasoning_effort: Option<ReasoningEffort>,
+
     // Define if we need a special handling of reasoning summary
     pub reasoning_summary_format: ReasoningSummaryFormat,
 
-    // This should be set to true when the model expects a tool named
-    // "local_shell" to be provided. Its contract must be understood natively by
-    // the model such that its description can be omitted.
-    // See https://platform.openai.com/docs/guides/tools-local-shell
-    pub uses_local_shell_tool: bool,
-
     /// Whether this model supports parallel tool calls when using the
     /// Responses API.
     pub supports_parallel_tool_calls: bool,
@@ -57,6 +60,12 @@ pub struct ModelFamily {
 
     /// If the model family supports setting the verbosity level when using Responses API.
     pub support_verbosity: bool,
+
+    // The default verbosity level for this model family when using Responses API.
+    pub default_verbosity: Option<Verbosity>,
+
+    /// Preferred shell tool type for this model family when features do not override it.
+    pub shell_type: ConfigShellToolType,
 }
 
 macro_rules! model_family {
@@ -64,20 +73,24 @@ macro_rules! model_family {
         $slug:expr, $family:expr $(, $key:ident : $value:expr )* $(,)?
     ) => {{
         // defaults
+        #[allow(unused_mut)]
         let mut mf = ModelFamily {
             slug: $slug.to_string(),
             family: $family.to_string(),
             needs_special_apply_patch_instructions: false,
             supports_reasoning_summaries: false,
             reasoning_summary_format: ReasoningSummaryFormat::None,
-            uses_local_shell_tool: false,
             supports_parallel_tool_calls: false,
             apply_patch_tool_type: None,
             base_instructions: BASE_INSTRUCTIONS.to_string(),
             experimental_supported_tools: Vec::new(),
             effective_context_window_percent: 95,
             support_verbosity: false,
+            shell_type: ConfigShellToolType::Default,
+            default_verbosity: None,
+            default_reasoning_effort: None,
         };
+
         // apply overrides
         $(
             mf.$key = $value;
@@ -105,8 +118,8 @@ pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
         model_family!(
             slug, "codex-mini-latest",
             supports_reasoning_summaries: true,
-            uses_local_shell_tool: true,
             needs_special_apply_patch_instructions: true,
+            shell_type: ConfigShellToolType::Local,
         )
     } else if slug.starts_with("gpt-4.1") {
         model_family!(
@@ -153,7 +166,10 @@ pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
         )
 
     // Production models.
-    } else if slug.starts_with("gpt-5-codex") || slug.starts_with("codex-") {
+    } else if slug.starts_with("gpt-5-codex")
+        || slug.starts_with("gpt-5.1-codex")
+        || slug.starts_with("codex-")
+    {
         model_family!(
             slug, slug,
             supports_reasoning_summaries: true,
@@ -162,6 +178,16 @@ pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
             apply_patch_tool_type: Some(ApplyPatchToolType::Freeform),
             support_verbosity: false,
         )
+    } else if slug.starts_with("gpt-5.1") {
+        model_family!(
+            slug, "gpt-5.1",
+            supports_reasoning_summaries: true,
+            apply_patch_tool_type: Some(ApplyPatchToolType::Freeform),
+            support_verbosity: true,
+            default_verbosity: Some(Verbosity::Low),
+            base_instructions: GPT_5_1_INSTRUCTIONS.to_string(),
+            default_reasoning_effort: Some(ReasoningEffort::Medium),
+        )
     } else if slug.starts_with("gpt-5") {
         model_family!(
             slug, "gpt-5",
@@ -181,12 +207,14 @@ pub fn derive_default_model_family(model: &str) -> ModelFamily {
         needs_special_apply_patch_instructions: false,
         supports_reasoning_summaries: false,
         reasoning_summary_format: ReasoningSummaryFormat::None,
-        uses_local_shell_tool: false,
         supports_parallel_tool_calls: false,
         apply_patch_tool_type: None,
         base_instructions: BASE_INSTRUCTIONS.to_string(),
         experimental_supported_tools: Vec::new(),
         effective_context_window_percent: 95,
         support_verbosity: false,
+        shell_type: ConfigShellToolType::Default,
+        default_verbosity: None,
+        default_reasoning_effort: None,
     }
 }

codex-rs/core/src/openai_model_info.rs

@@ -70,7 +70,7 @@ pub(crate) fn get_model_info(model_family: &ModelFamily) -> Option<ModelInfo> {
         // https://platform.openai.com/docs/models/gpt-3.5-turbo
         "gpt-3.5-turbo" => Some(ModelInfo::new(16_385, 4_096)),
 
-        _ if slug.starts_with("gpt-5-codex") => {
+        _ if slug.starts_with("gpt-5-codex") || slug.starts_with("gpt-5.1-codex") => {
             Some(ModelInfo::new(CONTEXT_WINDOW_272K, MAX_OUTPUT_TOKENS_128K))
         }