Clippy

<img width="769" height="803" alt="Screenshot 2025-11-14 at 10 25 49"
src="https://github.com/user-attachments/assets/12fbc21e-cab9-4d0a-a484-1aeb60219f96"
/>

.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/cla.yml

@@ -16,10 +16,27 @@ jobs:
     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
@@ -170,7 +170,7 @@ jobs:
 
       # Install and restore sccache cache
       - name: Install sccache
-        uses: taiki-e/install-action@0c5db7f7f897c03b771660e91d065338615679f4 # v2
+        uses: taiki-e/install-action@44c6d64aa62cd779e873306675c7a58e86d6d532 # v2
         with:
           tool: sccache
           version: 0.7.5
@@ -228,7 +228,7 @@ jobs:
 
       - name: Install cargo-chef
         if: ${{ matrix.profile == 'release' }}
-        uses: taiki-e/install-action@0c5db7f7f897c03b771660e91d065338615679f4 # v2
+        uses: taiki-e/install-action@44c6d64aa62cd779e873306675c7a58e86d6d532 # v2
         with:
           tool: cargo-chef
           version: 0.1.71
@@ -370,7 +370,7 @@ jobs:
             cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-
 
       - name: Install sccache
-        uses: taiki-e/install-action@0c5db7f7f897c03b771660e91d065338615679f4 # v2
+        uses: taiki-e/install-action@44c6d64aa62cd779e873306675c7a58e86d6d532 # v2
         with:
           tool: sccache
           version: 0.7.5
@@ -399,17 +399,17 @@ 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'

.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

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:
 

codex-rs/Cargo.lock

@@ -211,6 +211,7 @@ dependencies = [
  "parking_lot",
  "percent-encoding",
  "windows-sys 0.59.0",
+ "wl-clipboard-rs",
  "x11rb",
 ]
 
@@ -237,46 +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]]
@@ -981,21 +980,23 @@ 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]]
@@ -1047,6 +1048,7 @@ dependencies = [
  "codex-app-server-protocol",
  "codex-core",
  "codex-protocol",
+ "once_cell",
  "serde",
  "toml",
 ]
@@ -1066,6 +1068,7 @@ dependencies = [
  "chrono",
  "codex-app-server-protocol",
  "codex-apply-patch",
+ "codex-arg0",
  "codex-async-utils",
  "codex-file-search",
  "codex-git",
@@ -1080,6 +1083,7 @@ dependencies = [
  "codex-windows-sandbox",
  "core-foundation 0.9.4",
  "core_test_support",
+ "ctor 0.5.0",
  "dirs",
  "dunce",
  "env-flags",
@@ -1111,6 +1115,7 @@ dependencies = [
  "similar",
  "strum_macros 0.27.2",
  "tempfile",
+ "test-case",
  "test-log",
  "thiserror 2.0.17",
  "time",
@@ -1365,16 +1370,6 @@ 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"
@@ -1452,8 +1447,10 @@ dependencies = [
  "codex-login",
  "codex-ollama",
  "codex-protocol",
+ "codex-windows-sandbox",
  "color-eyre",
  "crossterm",
+ "derive_more 2.0.1",
  "diffy",
  "dirs",
  "dunce",
@@ -1463,6 +1460,7 @@ dependencies = [
  "lazy_static",
  "libc",
  "mcp-types",
+ "objc",
  "opentelemetry-appender-tracing",
  "pathdiff",
  "pretty_assertions",
@@ -1564,6 +1562,7 @@ version = "0.1.0"
 dependencies = [
  "anyhow",
  "dirs-next",
+ "dunce",
  "rand 0.8.5",
  "serde",
  "serde_json",
@@ -1657,6 +1656,15 @@ 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"
@@ -2002,7 +2010,7 @@ 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",
@@ -2015,6 +2023,7 @@ 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",
@@ -2878,15 +2887,6 @@ 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"
@@ -3530,12 +3530,6 @@ 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"
@@ -3641,6 +3635,15 @@ dependencies = [
  "url",
 ]
 
+[[package]]
+name = "malloc_buf"
+version = "0.0.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "62bb907fe88d54d8d9ce32a3cceab4218ed2f6b7d35617cafe9adf84e43919cb"
+dependencies = [
+ "libc",
+]
+
 [[package]]
 name = "maplit"
 version = "1.0.2"
@@ -4028,6 +4031,15 @@ dependencies = [
  "url",
 ]
 
+[[package]]
+name = "objc"
+version = "0.2.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "915b1b472bc21c53464d6c8461c9d3af805ba1ef837e1cac254428f4a77177b1"
+dependencies = [
+ "malloc_buf",
+]
+
 [[package]]
 name = "objc2"
 version = "0.6.2"
@@ -4302,6 +4314,16 @@ 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"
@@ -4449,7 +4471,7 @@ checksum = "3af6b589e163c5a788fab00ce0c0366f6efbb9959c2f9874b224936af7fce7e1"
 dependencies = [
  "base64",
  "indexmap 2.12.0",
- "quick-xml",
+ "quick-xml 0.38.0",
  "serde",
  "time",
 ]
@@ -4678,6 +4700,15 @@ 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"
@@ -6150,6 +6181,39 @@ 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"
@@ -6715,6 +6779,18 @@ 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"
@@ -7052,6 +7128,76 @@ 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"
@@ -7623,6 +7769,25 @@ 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"
@@ -7791,9 +7956,9 @@ 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

@@ -25,7 +25,6 @@ members = [
     "ollama",
     "process-hardening",
     "protocol",
-    "protocol-ts",
     "rmcp-client",
     "responses-api-proxy",
     "stdio-to-uds",
@@ -75,7 +74,6 @@ 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 +85,7 @@ 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 +94,8 @@ 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 +122,8 @@ 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 +140,7 @@ 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"
@@ -213,7 +212,7 @@ which = "6"
 wildmatch = "2.5.0"
 
 wiremock = "0.6"
-zeroize = "1.8.1"
+zeroize = "1.8.2"
 
 [workspace.lints]
 rust = {}
@@ -280,8 +279,8 @@ opt-level = 0
 [patch.crates-io]
 # Uncomment to debug local changes.
 # ratatui = { path = "../../ratatui" }
-ratatui = { git = "https://github.com/nornagon/ratatui", branch = "nornagon-v0.29.0-patch" }
 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.
 # rmcp = { path = "../../rust-sdk/crates/rmcp" }

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/src/export.rs

@@ -13,10 +13,7 @@ use crate::export_server_responses;
 use anyhow::Context;
 use anyhow::Result;
 use anyhow::anyhow;
-use codex_protocol::parse_command::ParsedCommand;
 use codex_protocol::protocol::EventMsg;
-use codex_protocol::protocol::FileChange;
-use codex_protocol::protocol::SandboxPolicy;
 use schemars::JsonSchema;
 use schemars::schema_for;
 use serde::Serialize;
@@ -92,6 +89,8 @@ pub fn generate_ts(out_dir: &Path, prettier: Option<&Path>) -> Result<()> {
     {
         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()))?;
@@ -118,10 +117,6 @@ pub fn generate_json(out_dir: &Path) -> Result<()> {
         |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"),
-        |d| write_json_schema_with_return::<FileChange>(d, "FileChange"),
-        |d| write_json_schema_with_return::<crate::protocol::v1::InputItem>(d, "InputItem"),
-        |d| write_json_schema_with_return::<ParsedCommand>(d, "ParsedCommand"),
-        |d| write_json_schema_with_return::<SandboxPolicy>(d, "SandboxPolicy"),
     ];
 
     let mut schemas: Vec<GeneratedSchema> = Vec::new();
@@ -150,13 +145,10 @@ fn build_schema_bundle(schemas: Vec<GeneratedSchema>) -> Result<Value> {
         "ClientNotification",
         "ClientRequest",
         "EventMsg",
-        "FileChange",
-        "InputItem",
-        "ParsedCommand",
-        "SandboxPolicy",
         "ServerNotification",
         "ServerRequest",
     ];
+    const IGNORED_DEFINITIONS: &[&str] = &["Option<()>"];
 
     let namespaced_types = collect_namespaced_types(&schemas);
     let mut definitions = Map::new();
@@ -169,6 +161,10 @@ fn build_schema_bundle(schemas: Vec<GeneratedSchema>) -> Result<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);
         }
@@ -179,6 +175,9 @@ fn build_schema_bundle(schemas: Vec<GeneratedSchema>) -> Result<Value> {
             && let Value::Object(defs_obj) = defs
         {
             for (def_name, mut def_schema) in defs_obj {
+                if IGNORED_DEFINITIONS.contains(&def_name.as_str()) {
+                    continue;
+                }
                 if SPECIAL_DEFINITIONS.contains(&def_name.as_str()) {
                     continue;
                 }
@@ -384,14 +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()
         {
@@ -666,6 +657,8 @@ fn ts_files_in_recursive(dir: &Path) -> Result<Vec<PathBuf>> {
     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)?

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

@@ -46,7 +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,
             }
@@ -58,6 +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,
@@ -101,105 +102,78 @@ macro_rules! client_request_definitions {
 }
 
 client_request_definitions! {
-    /// NEW APIs
-    // Thread lifecycle
-    #[serde(rename = "thread/start")]
-    #[ts(rename = "thread/start")]
-    ThreadStart {
-        params: v2::ThreadStartParams,
-        response: v2::ThreadStartResponse,
-    },
-    #[serde(rename = "thread/resume")]
-    #[ts(rename = "thread/resume")]
-    ThreadResume {
-        params: v2::ThreadResumeParams,
-        response: v2::ThreadResumeResponse,
-    },
-    #[serde(rename = "thread/archive")]
-    #[ts(rename = "thread/archive")]
-    ThreadArchive {
-        params: v2::ThreadArchiveParams,
-        response: v2::ThreadArchiveResponse,
-    },
-    #[serde(rename = "thread/list")]
-    #[ts(rename = "thread/list")]
-    ThreadList {
-        params: v2::ThreadListParams,
-        response: v2::ThreadListResponse,
-    },
-    #[serde(rename = "thread/compact")]
-    #[ts(rename = "thread/compact")]
-    ThreadCompact {
-        params: v2::ThreadCompactParams,
-        response: v2::ThreadCompactResponse,
-    },
-    #[serde(rename = "turn/start")]
-    #[ts(rename = "turn/start")]
-    TurnStart {
-        params: v2::TurnStartParams,
-        response: v2::TurnStartResponse,
-    },
-    #[serde(rename = "turn/interrupt")]
-    #[ts(rename = "turn/interrupt")]
-    TurnInterrupt {
-        params: v2::TurnInterruptParams,
-        response: v2::TurnInterruptResponse,
-    },
-
-    #[serde(rename = "model/list")]
-    #[ts(rename = "model/list")]
-    ModelList {
-        params: v2::ModelListParams,
-        response: v2::ModelListResponse,
-    },
-
-    #[serde(rename = "account/login/start")]
-    #[ts(rename = "account/login/start")]
-    LoginAccount {
-        params: v2::LoginAccountParams,
-        response: v2::LoginAccountResponse,
-    },
-
-    #[serde(rename = "account/login/cancel")]
-    #[ts(rename = "account/login/cancel")]
-    CancelLoginAccount {
-        params: v2::CancelLoginAccountParams,
-        response: v2::CancelLoginAccountResponse,
-    },
-
-    #[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")]
-    FeedbackUpload {
-        params: v2::FeedbackUploadParams,
-        response: v2::FeedbackUploadResponse,
-    },
-
-    #[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,
@@ -263,6 +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,
@@ -533,6 +508,9 @@ server_notification_definitions! {
     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")]
@@ -758,12 +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)?,
         );
@@ -772,19 +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()),
+            email: "user@example.com".to_string(),
             plan_type: PlanType::Plus,
         };
         assert_eq!(

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)]

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

@@ -6,6 +6,9 @@ use codex_protocol::ConversationId;
 use codex_protocol::account::PlanType;
 use codex_protocol::config_types::ReasoningEffort;
 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;
@@ -54,8 +57,8 @@ v2_enum_from_core!(
 );
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq, JsonSchema, TS)]
-#[serde(tag = "mode", rename_all = "camelCase")]
-#[ts(tag = "mode")]
+#[serde(tag = "type", rename_all = "camelCase")]
+#[ts(tag = "type")]
 #[ts(export_to = "v2/")]
 pub enum SandboxPolicy {
     DangerFullAccess,
@@ -123,14 +126,11 @@ impl From<codex_protocol::protocol::SandboxPolicy> for SandboxPolicy {
 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)]
@@ -193,11 +193,20 @@ pub struct GetAccountRateLimitsResponse {
     pub rate_limits: RateLimitSnapshot,
 }
 
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[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: Account,
+    pub account: Option<Account>,
+    pub requires_openai_auth: bool,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Default, JsonSchema, TS)]
@@ -282,11 +291,39 @@ pub struct ThreadStartResponse {
     pub thread: Thread,
 }
 
-#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[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)]
@@ -348,6 +385,13 @@ pub struct ThreadCompactResponse {}
 #[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)]
@@ -446,6 +490,17 @@ impl UserInput {
     }
 }
 
+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")]
@@ -461,7 +516,10 @@ pub enum ThreadItem {
     },
     Reasoning {
         id: String,
-        text: String,
+        #[serde(default)]
+        summary: Vec<String>,
+        #[serde(default)]
+        content: Vec<String>,
     },
     CommandExecution {
         id: String,
@@ -503,6 +561,36 @@ pub enum ThreadItem {
     },
 }
 
+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/")]
@@ -628,6 +716,32 @@ pub struct AgentMessageDeltaNotification {
     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/")]
@@ -697,3 +811,101 @@ pub struct AccountLoginCompletedNotification {
     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/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/codex_message_processor.rs

@@ -4,11 +4,15 @@ use crate::fuzzy_file_search::run_fuzzy_file_search;
 use crate::models::supported_models;
 use crate::outgoing_message::OutgoingMessageSender;
 use crate::outgoing_message::OutgoingNotification;
+use chrono::DateTime;
+use chrono::Utc;
+use codex_app_server_protocol::Account;
 use codex_app_server_protocol::AccountLoginCompletedNotification;
 use codex_app_server_protocol::AccountRateLimitsUpdatedNotification;
 use codex_app_server_protocol::AccountUpdatedNotification;
 use codex_app_server_protocol::AddConversationListenerParams;
 use codex_app_server_protocol::AddConversationSubscriptionResponse;
+use codex_app_server_protocol::AgentMessageDeltaNotification;
 use codex_app_server_protocol::ApplyPatchApprovalParams;
 use codex_app_server_protocol::ApplyPatchApprovalResponse;
 use codex_app_server_protocol::ArchiveConversationParams;
@@ -20,6 +24,7 @@ use codex_app_server_protocol::CancelLoginAccountParams;
 use codex_app_server_protocol::CancelLoginAccountResponse;
 use codex_app_server_protocol::CancelLoginChatGptResponse;
 use codex_app_server_protocol::ClientRequest;
+use codex_app_server_protocol::ConversationGitInfo;
 use codex_app_server_protocol::ConversationSummary;
 use codex_app_server_protocol::ExecCommandApprovalParams;
 use codex_app_server_protocol::ExecCommandApprovalResponse;
@@ -29,7 +34,9 @@ use codex_app_server_protocol::FeedbackUploadParams;
 use codex_app_server_protocol::FeedbackUploadResponse;
 use codex_app_server_protocol::FuzzyFileSearchParams;
 use codex_app_server_protocol::FuzzyFileSearchResponse;
+use codex_app_server_protocol::GetAccountParams;
 use codex_app_server_protocol::GetAccountRateLimitsResponse;
+use codex_app_server_protocol::GetAccountResponse;
 use codex_app_server_protocol::GetAuthStatusParams;
 use codex_app_server_protocol::GetAuthStatusResponse;
 use codex_app_server_protocol::GetConversationSummaryParams;
@@ -40,6 +47,8 @@ use codex_app_server_protocol::GitDiffToRemoteResponse;
 use codex_app_server_protocol::InputItem as WireInputItem;
 use codex_app_server_protocol::InterruptConversationParams;
 use codex_app_server_protocol::InterruptConversationResponse;
+use codex_app_server_protocol::ItemCompletedNotification;
+use codex_app_server_protocol::ItemStartedNotification;
 use codex_app_server_protocol::JSONRPCErrorError;
 use codex_app_server_protocol::ListConversationsParams;
 use codex_app_server_protocol::ListConversationsResponse;
@@ -54,6 +63,9 @@ use codex_app_server_protocol::ModelListParams;
 use codex_app_server_protocol::ModelListResponse;
 use codex_app_server_protocol::NewConversationParams;
 use codex_app_server_protocol::NewConversationResponse;
+use codex_app_server_protocol::ReasoningSummaryPartAddedNotification;
+use codex_app_server_protocol::ReasoningSummaryTextDeltaNotification;
+use codex_app_server_protocol::ReasoningTextDeltaNotification;
 use codex_app_server_protocol::RemoveConversationListenerParams;
 use codex_app_server_protocol::RemoveConversationSubscriptionResponse;
 use codex_app_server_protocol::RequestId;
@@ -130,8 +142,10 @@ use codex_protocol::ConversationId;
 use codex_protocol::config_types::ForcedLoginMethod;
 use codex_protocol::items::TurnItem;
 use codex_protocol::models::ResponseItem;
+use codex_protocol::protocol::GitInfo;
 use codex_protocol::protocol::RateLimitSnapshot as CoreRateLimitSnapshot;
 use codex_protocol::protocol::RolloutItem;
+use codex_protocol::protocol::SessionMetaLine;
 use codex_protocol::protocol::USER_MESSAGE_BEGIN;
 use codex_protocol::user_input::UserInput as CoreInputItem;
 use codex_utils_json_to_toml::json_to_toml;
@@ -190,6 +204,30 @@ enum ApiVersion {
 }
 
 impl CodexMessageProcessor {
+    async fn conversation_from_thread_id(
+        &self,
+        thread_id: &str,
+    ) -> Result<(ConversationId, Arc<CodexConversation>), JSONRPCErrorError> {
+        // Resolve conversation id from v2 thread id string.
+        let conversation_id =
+            ConversationId::from_string(thread_id).map_err(|err| JSONRPCErrorError {
+                code: INVALID_REQUEST_ERROR_CODE,
+                message: format!("invalid thread id: {err}"),
+                data: None,
+            })?;
+
+        let conversation = self
+            .conversation_manager
+            .get_conversation(conversation_id)
+            .await
+            .map_err(|_| JSONRPCErrorError {
+                code: INVALID_REQUEST_ERROR_CODE,
+                message: format!("conversation not found: {conversation_id}"),
+                data: None,
+            })?;
+
+        Ok((conversation_id, conversation))
+    }
     pub fn new(
         auth_manager: Arc<AuthManager>,
         conversation_manager: Arc<ConversationManager>,
@@ -270,12 +308,8 @@ impl CodexMessageProcessor {
             ClientRequest::CancelLoginAccount { request_id, params } => {
                 self.cancel_login_v2(request_id, params).await;
             }
-            ClientRequest::GetAccount {
-                request_id,
-                params: _,
-            } => {
-                self.send_unimplemented_error(request_id, "account/read")
-                    .await;
+            ClientRequest::GetAccount { request_id, params } => {
+                self.get_account(request_id, params).await;
             }
             ClientRequest::ResumeConversation { request_id, params } => {
                 self.handle_resume_conversation(request_id, params).await;
@@ -798,13 +832,17 @@ impl CodexMessageProcessor {
         }
     }
 
+    async fn refresh_token_if_requested(&self, do_refresh: bool) {
+        if do_refresh && let Err(err) = self.auth_manager.refresh_token().await {
+            tracing::warn!("failed to refresh token whilte getting account: {err}");
+        }
+    }
+
     async fn get_auth_status(&self, request_id: RequestId, params: GetAuthStatusParams) {
         let include_token = params.include_token.unwrap_or(false);
         let do_refresh = params.refresh_token.unwrap_or(false);
 
-        if do_refresh && let Err(err) = self.auth_manager.refresh_token().await {
-            tracing::warn!("failed to refresh token while getting auth status: {err}");
-        }
+        self.refresh_token_if_requested(do_refresh).await;
 
         // Determine whether auth is required based on the active model provider.
         // If a custom provider is configured with `requires_openai_auth == false`,
@@ -849,6 +887,56 @@ impl CodexMessageProcessor {
         self.outgoing.send_response(request_id, response).await;
     }
 
+    async fn get_account(&self, request_id: RequestId, params: GetAccountParams) {
+        let do_refresh = params.refresh_token;
+
+        self.refresh_token_if_requested(do_refresh).await;
+
+        // Whether auth is required for the active model provider.
+        let requires_openai_auth = self.config.model_provider.requires_openai_auth;
+
+        if !requires_openai_auth {
+            let response = GetAccountResponse {
+                account: None,
+                requires_openai_auth,
+            };
+            self.outgoing.send_response(request_id, response).await;
+            return;
+        }
+
+        let account = match self.auth_manager.auth() {
+            Some(auth) => Some(match auth.mode {
+                AuthMode::ApiKey => Account::ApiKey {},
+                AuthMode::ChatGPT => {
+                    let email = auth.get_account_email();
+                    let plan_type = auth.account_plan_type();
+
+                    match (email, plan_type) {
+                        (Some(email), Some(plan_type)) => Account::Chatgpt { email, plan_type },
+                        _ => {
+                            let error = JSONRPCErrorError {
+                                code: INVALID_REQUEST_ERROR_CODE,
+                                message:
+                                    "email and plan type are required for chatgpt authentication"
+                                        .to_string(),
+                                data: None,
+                            };
+                            self.outgoing.send_error(request_id, error).await;
+                            return;
+                        }
+                    }
+                }
+            }),
+            None => None,
+        };
+
+        let response = GetAccountResponse {
+            account,
+            requires_openai_auth,
+        };
+        self.outgoing.send_response(request_id, response).await;
+    }
+
     async fn get_user_agent(&self, request_id: RequestId) {
         let user_agent = get_codex_user_agent();
         let response = GetUserAgentResponse { user_agent };
@@ -1117,21 +1205,17 @@ impl CodexMessageProcessor {
     }
 
     async fn thread_start(&mut self, request_id: RequestId, params: ThreadStartParams) {
-        // Build ConfigOverrides directly from ThreadStartParams for config derivation.
-        let cli_overrides = params.config;
-        let overrides = ConfigOverrides {
-            model: params.model,
-            cwd: params.cwd.map(PathBuf::from),
-            approval_policy: params.approval_policy.map(AskForApproval::to_core),
-            sandbox_mode: params.sandbox.map(SandboxMode::to_core),
-            model_provider: params.model_provider,
-            codex_linux_sandbox_exe: self.codex_linux_sandbox_exe.clone(),
-            base_instructions: params.base_instructions,
-            developer_instructions: params.developer_instructions,
-            ..Default::default()
-        };
+        let overrides = self.build_thread_config_overrides(
+            params.model,
+            params.model_provider,
+            params.cwd,
+            params.approval_policy,
+            params.sandbox,
+            params.base_instructions,
+            params.developer_instructions,
+        );
 
-        let config = match derive_config_from_params(overrides, cli_overrides).await {
+        let config = match derive_config_from_params(overrides, params.config).await {
             Ok(config) => config,
             Err(err) => {
                 let error = JSONRPCErrorError {
@@ -1146,8 +1230,30 @@ impl CodexMessageProcessor {
 
         match self.conversation_manager.new_conversation(config).await {
             Ok(new_conv) => {
-                let thread = Thread {
-                    id: new_conv.conversation_id.to_string(),
+                let conversation_id = new_conv.conversation_id;
+                let rollout_path = new_conv.session_configured.rollout_path.clone();
+                let fallback_provider = self.config.model_provider_id.as_str();
+
+                // A bit hacky, but the summary contains a lot of useful information for the thread
+                // that unfortunately does not get returned from conversation_manager.new_conversation().
+                let thread = match read_summary_from_rollout(
+                    rollout_path.as_path(),
+                    fallback_provider,
+                )
+                .await
+                {
+                    Ok(summary) => summary_to_thread(summary),
+                    Err(err) => {
+                        self.send_internal_error(
+                            request_id,
+                            format!(
+                                "failed to load rollout `{}` for conversation {conversation_id}: {err}",
+                                rollout_path.display()
+                            ),
+                        )
+                        .await;
+                        return;
+                    }
                 };
 
                 let response = ThreadStartResponse {
@@ -1157,12 +1263,12 @@ impl CodexMessageProcessor {
                 // Auto-attach a conversation listener when starting a thread.
                 // Use the same behavior as the v1 API with experimental_raw_events=false.
                 if let Err(err) = self
-                    .attach_conversation_listener(new_conv.conversation_id, false)
+                    .attach_conversation_listener(conversation_id, false)
                     .await
                 {
                     tracing::warn!(
                         "failed to attach listener for conversation {}: {}",
-                        new_conv.conversation_id,
+                        conversation_id,
                         err.message
                     );
                 }
@@ -1185,6 +1291,31 @@ impl CodexMessageProcessor {
         }
     }
 
+    #[allow(clippy::too_many_arguments)]
+    fn build_thread_config_overrides(
+        &self,
+        model: Option<String>,
+        model_provider: Option<String>,
+        cwd: Option<String>,
+        approval_policy: Option<codex_app_server_protocol::AskForApproval>,
+        sandbox: Option<SandboxMode>,
+        base_instructions: Option<String>,
+        developer_instructions: Option<String>,
+    ) -> ConfigOverrides {
+        ConfigOverrides {
+            model,
+            model_provider,
+            cwd: cwd.map(PathBuf::from),
+            approval_policy: approval_policy
+                .map(codex_app_server_protocol::AskForApproval::to_core),
+            sandbox_mode: sandbox.map(SandboxMode::to_core),
+            codex_linux_sandbox_exe: self.codex_linux_sandbox_exe.clone(),
+            base_instructions,
+            developer_instructions,
+            ..Default::default()
+        }
+    }
+
     async fn thread_archive(&mut self, request_id: RequestId, params: ThreadArchiveParams) {
         let conversation_id = match ConversationId::from_string(&params.thread_id) {
             Ok(id) => id,
@@ -1260,98 +1391,154 @@ impl CodexMessageProcessor {
             }
         };
 
-        let data = summaries
-            .into_iter()
-            .map(|s| Thread {
-                id: s.conversation_id.to_string(),
-            })
-            .collect();
+        let data = summaries.into_iter().map(summary_to_thread).collect();
 
         let response = ThreadListResponse { data, next_cursor };
         self.outgoing.send_response(request_id, response).await;
     }
 
     async fn thread_resume(&mut self, request_id: RequestId, params: ThreadResumeParams) {
-        let conversation_id = match ConversationId::from_string(&params.thread_id) {
-            Ok(id) => id,
-            Err(err) => {
-                let error = JSONRPCErrorError {
-                    code: INVALID_REQUEST_ERROR_CODE,
-                    message: format!("invalid thread id: {err}"),
-                    data: None,
-                };
-                self.outgoing.send_error(request_id, error).await;
+        let ThreadResumeParams {
+            thread_id,
+            history,
+            path,
+            model,
+            model_provider,
+            cwd,
+            approval_policy,
+            sandbox,
+            config: cli_overrides,
+            base_instructions,
+            developer_instructions,
+        } = params;
+
+        let overrides_requested = model.is_some()
+            || model_provider.is_some()
+            || cwd.is_some()
+            || approval_policy.is_some()
+            || sandbox.is_some()
+            || cli_overrides.is_some()
+            || base_instructions.is_some()
+            || developer_instructions.is_some();
+
+        let config = if overrides_requested {
+            let overrides = self.build_thread_config_overrides(
+                model,
+                model_provider,
+                cwd,
+                approval_policy,
+                sandbox,
+                base_instructions,
+                developer_instructions,
+            );
+            match derive_config_from_params(overrides, cli_overrides).await {
+                Ok(config) => config,
+                Err(err) => {
+                    let error = JSONRPCErrorError {
+                        code: INVALID_REQUEST_ERROR_CODE,
+                        message: format!("error deriving config: {err}"),
+                        data: None,
+                    };
+                    self.outgoing.send_error(request_id, error).await;
+                    return;
+                }
+            }
+        } else {
+            self.config.as_ref().clone()
+        };
+
+        let conversation_history = if let Some(history) = history {
+            if history.is_empty() {
+                self.send_invalid_request_error(
+                    request_id,
+                    "history must not be empty".to_string(),
+                )
+                .await;
                 return;
             }
+            InitialHistory::Forked(history.into_iter().map(RolloutItem::ResponseItem).collect())
+        } else if let Some(path) = path {
+            match RolloutRecorder::get_rollout_history(&path).await {
+                Ok(initial_history) => initial_history,
+                Err(err) => {
+                    self.send_invalid_request_error(
+                        request_id,
+                        format!("failed to load rollout `{}`: {err}", path.display()),
+                    )
+                    .await;
+                    return;
+                }
+            }
+        } else {
+            let existing_conversation_id = match ConversationId::from_string(&thread_id) {
+                Ok(id) => id,
+                Err(err) => {
+                    let error = JSONRPCErrorError {
+                        code: INVALID_REQUEST_ERROR_CODE,
+                        message: format!("invalid thread id: {err}"),
+                        data: None,
+                    };
+                    self.outgoing.send_error(request_id, error).await;
+                    return;
+                }
+            };
+
+            let path = match find_conversation_path_by_id_str(
+                &self.config.codex_home,
+                &existing_conversation_id.to_string(),
+            )
+            .await
+            {
+                Ok(Some(p)) => p,
+                Ok(None) => {
+                    self.send_invalid_request_error(
+                        request_id,
+                        format!("no rollout found for conversation id {existing_conversation_id}"),
+                    )
+                    .await;
+                    return;
+                }
+                Err(err) => {
+                    self.send_invalid_request_error(
+                        request_id,
+                        format!(
+                            "failed to locate conversation id {existing_conversation_id}: {err}"
+                        ),
+                    )
+                    .await;
+                    return;
+                }
+            };
+
+            match RolloutRecorder::get_rollout_history(&path).await {
+                Ok(initial_history) => initial_history,
+                Err(err) => {
+                    self.send_invalid_request_error(
+                        request_id,
+                        format!("failed to load rollout `{}`: {err}", path.display()),
+                    )
+                    .await;
+                    return;
+                }
+            }
         };
 
-        let path = match find_conversation_path_by_id_str(
-            &self.config.codex_home,
-            &conversation_id.to_string(),
-        )
-        .await
-        {
-            Ok(Some(p)) => p,
-            Ok(None) => {
-                let error = JSONRPCErrorError {
-                    code: INVALID_REQUEST_ERROR_CODE,
-                    message: format!("no rollout found for conversation id {conversation_id}"),
-                    data: None,
-                };
-                self.outgoing.send_error(request_id, error).await;
-                return;
-            }
-            Err(err) => {
-                let error = JSONRPCErrorError {
-                    code: INVALID_REQUEST_ERROR_CODE,
-                    message: format!("failed to locate conversation id {conversation_id}: {err}"),
-                    data: None,
-                };
-                self.outgoing.send_error(request_id, error).await;
-                return;
-            }
-        };
-
-        let fallback_provider = self.config.model_provider_id.as_str();
-        let summary = match read_summary_from_rollout(&path, fallback_provider).await {
-            Ok(s) => s,
-            Err(err) => {
-                let error = JSONRPCErrorError {
-                    code: INVALID_REQUEST_ERROR_CODE,
-                    message: format!("failed to load rollout `{}`: {err}", path.display()),
-                    data: None,
-                };
-                self.outgoing.send_error(request_id, error).await;
-                return;
-            }
-        };
-
-        let initial_history = match RolloutRecorder::get_rollout_history(&summary.path).await {
-            Ok(initial_history) => initial_history,
-            Err(err) => {
-                let error = JSONRPCErrorError {
-                    code: INVALID_REQUEST_ERROR_CODE,
-                    message: format!(
-                        "failed to load rollout `{}` for conversation {conversation_id}: {err}",
-                        summary.path.display()
-                    ),
-                    data: None,
-                };
-                self.outgoing.send_error(request_id, error).await;
-                return;
-            }
-        };
+        let fallback_model_provider = config.model_provider_id.clone();
 
         match self
             .conversation_manager
             .resume_conversation_with_history(
-                self.config.as_ref().clone(),
-                initial_history,
+                config,
+                conversation_history,
                 self.auth_manager.clone(),
             )
             .await
         {
-            Ok(_) => {
+            Ok(NewConversation {
+                conversation_id,
+                session_configured,
+                ..
+            }) => {
                 // Auto-attach a conversation listener when resuming a thread.
                 if let Err(err) = self
                     .attach_conversation_listener(conversation_id, false)
@@ -1364,11 +1551,26 @@ impl CodexMessageProcessor {
                     );
                 }
 
-                let response = ThreadResumeResponse {
-                    thread: Thread {
-                        id: conversation_id.to_string(),
-                    },
+                let thread = match read_summary_from_rollout(
+                    session_configured.rollout_path.as_path(),
+                    fallback_model_provider.as_str(),
+                )
+                .await
+                {
+                    Ok(summary) => summary_to_thread(summary),
+                    Err(err) => {
+                        self.send_internal_error(
+                            request_id,
+                            format!(
+                                "failed to load rollout `{}` for conversation {conversation_id}: {err}",
+                                session_configured.rollout_path.display()
+                            ),
+                        )
+                        .await;
+                        return;
+                    }
                 };
+                let response = ThreadResumeResponse { thread };
                 self.outgoing.send_response(request_id, response).await;
             }
             Err(err) => {
@@ -1510,7 +1712,18 @@ impl CodexMessageProcessor {
         let items = page
             .items
             .into_iter()
-            .filter_map(|it| extract_conversation_summary(it.path, &it.head, &fallback_provider))
+            .filter_map(|it| {
+                let session_meta_line = it.head.first().and_then(|first| {
+                    serde_json::from_value::<SessionMetaLine>(first.clone()).ok()
+                })?;
+                extract_conversation_summary(
+                    it.path,
+                    &it.head,
+                    &session_meta_line.meta,
+                    session_meta_line.git.as_ref(),
+                    fallback_provider.as_str(),
+                )
+            })
             .collect::<Vec<_>>();
 
         // Encode next_cursor as a plain string
@@ -1767,6 +1980,15 @@ impl CodexMessageProcessor {
         self.outgoing.send_error(request_id, error).await;
     }
 
+    async fn send_internal_error(&self, request_id: RequestId, message: String) {
+        let error = JSONRPCErrorError {
+            code: INTERNAL_ERROR_CODE,
+            message,
+            data: None,
+        };
+        self.outgoing.send_error(request_id, error).await;
+    }
+
     async fn archive_conversation(
         &mut self,
         request_id: RequestId,
@@ -2060,34 +2282,14 @@ impl CodexMessageProcessor {
     }
 
     async fn turn_start(&self, request_id: RequestId, params: TurnStartParams) {
-        // Resolve conversation id from v2 thread id string.
-        let conversation_id = match ConversationId::from_string(&params.thread_id) {
-            Ok(id) => id,
-            Err(err) => {
-                let error = JSONRPCErrorError {
-                    code: INVALID_REQUEST_ERROR_CODE,
-                    message: format!("invalid thread id: {err}"),
-                    data: None,
-                };
+        let (_, conversation) = match self.conversation_from_thread_id(&params.thread_id).await {
+            Ok(v) => v,
+            Err(error) => {
                 self.outgoing.send_error(request_id, error).await;
                 return;
             }
         };
 
-        let Ok(conversation) = self
-            .conversation_manager
-            .get_conversation(conversation_id)
-            .await
-        else {
-            let error = JSONRPCErrorError {
-                code: INVALID_REQUEST_ERROR_CODE,
-                message: format!("conversation not found: {conversation_id}"),
-                data: None,
-            };
-            self.outgoing.send_error(request_id, error).await;
-            return;
-        };
-
         // Keep a copy of v2 inputs for the notification payload.
         let v2_inputs_for_notif = params.input.clone();
 
@@ -2161,33 +2363,14 @@ impl CodexMessageProcessor {
     async fn turn_interrupt(&mut self, request_id: RequestId, params: TurnInterruptParams) {
         let TurnInterruptParams { thread_id, .. } = params;
 
-        // Resolve conversation id from v2 thread id string.
-        let conversation_id = match ConversationId::from_string(&thread_id) {
-            Ok(id) => id,
-            Err(err) => {
-                let error = JSONRPCErrorError {
-                    code: INVALID_REQUEST_ERROR_CODE,
-                    message: format!("invalid thread id: {err}"),
-                    data: None,
-                };
-                self.outgoing.send_error(request_id, error).await;
-                return;
-            }
-        };
-
-        let Ok(conversation) = self
-            .conversation_manager
-            .get_conversation(conversation_id)
-            .await
-        else {
-            let error = JSONRPCErrorError {
-                code: INVALID_REQUEST_ERROR_CODE,
-                message: format!("conversation not found: {conversation_id}"),
-                data: None,
+        let (conversation_id, conversation) =
+            match self.conversation_from_thread_id(&thread_id).await {
+                Ok(v) => v,
+                Err(error) => {
+                    self.outgoing.send_error(request_id, error).await;
+                    return;
+                }
             };
-            self.outgoing.send_error(request_id, error).await;
-            return;
-        };
 
         // Record the pending interrupt so we can reply when TurnAborted arrives.
         {
@@ -2502,6 +2685,48 @@ async fn apply_bespoke_event_handling(
                 on_patch_approval_response(event_id, rx, conversation).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,
@@ -2539,6 +2764,20 @@ async fn apply_bespoke_event_handling(
                     .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 = {
@@ -2671,16 +2910,25 @@ async fn read_summary_from_rollout(
         )));
     };
 
-    let session_meta = serde_json::from_value::<SessionMeta>(first.clone()).map_err(|_| {
-        IoError::other(format!(
-            "rollout at {} does not start with session metadata",
-            path.display()
-        ))
-    })?;
+    let session_meta_line =
+        serde_json::from_value::<SessionMetaLine>(first.clone()).map_err(|_| {
+            IoError::other(format!(
+                "rollout at {} does not start with session metadata",
+                path.display()
+            ))
+        })?;
+    let SessionMetaLine {
+        meta: session_meta,
+        git,
+    } = session_meta_line;
 
-    if let Some(summary) =
-        extract_conversation_summary(path.to_path_buf(), &head, fallback_provider)
-    {
+    if let Some(summary) = extract_conversation_summary(
+        path.to_path_buf(),
+        &head,
+        &session_meta,
+        git.as_ref(),
+        fallback_provider,
+    ) {
         return Ok(summary);
     }
 
@@ -2691,7 +2939,9 @@ async fn read_summary_from_rollout(
     };
     let model_provider = session_meta
         .model_provider
+        .clone()
         .unwrap_or_else(|| fallback_provider.to_string());
+    let git_info = git.as_ref().map(map_git_info);
 
     Ok(ConversationSummary {
         conversation_id: session_meta.id,
@@ -2699,19 +2949,20 @@ async fn read_summary_from_rollout(
         path: path.to_path_buf(),
         preview: String::new(),
         model_provider,
+        cwd: session_meta.cwd,
+        cli_version: session_meta.cli_version,
+        source: session_meta.source,
+        git_info,
     })
 }
 
 fn extract_conversation_summary(
     path: PathBuf,
     head: &[serde_json::Value],
+    session_meta: &SessionMeta,
+    git: Option<&GitInfo>,
     fallback_provider: &str,
 ) -> Option<ConversationSummary> {
-    let session_meta = match head.first() {
-        Some(first_line) => serde_json::from_value::<SessionMeta>(first_line.clone()).ok()?,
-        None => return None,
-    };
-
     let preview = head
         .iter()
         .filter_map(|value| serde_json::from_value::<ResponseItem>(value.clone()).ok())
@@ -2733,7 +2984,9 @@ fn extract_conversation_summary(
     let conversation_id = session_meta.id;
     let model_provider = session_meta
         .model_provider
+        .clone()
         .unwrap_or_else(|| fallback_provider.to_string());
+    let git_info = git.map(map_git_info);
 
     Some(ConversationSummary {
         conversation_id,
@@ -2741,13 +2994,55 @@ fn extract_conversation_summary(
         path,
         preview: preview.to_string(),
         model_provider,
+        cwd: session_meta.cwd.clone(),
+        cli_version: session_meta.cli_version.clone(),
+        source: session_meta.source.clone(),
+        git_info,
     })
 }
 
+fn map_git_info(git_info: &GitInfo) -> ConversationGitInfo {
+    ConversationGitInfo {
+        sha: git_info.commit_hash.clone(),
+        branch: git_info.branch.clone(),
+        origin_url: git_info.repository_url.clone(),
+    }
+}
+
+fn parse_datetime(timestamp: Option<&str>) -> Option<DateTime<Utc>> {
+    timestamp.and_then(|ts| {
+        chrono::DateTime::parse_from_rfc3339(ts)
+            .ok()
+            .map(|dt| dt.with_timezone(&chrono::Utc))
+    })
+}
+
+fn summary_to_thread(summary: ConversationSummary) -> Thread {
+    let ConversationSummary {
+        conversation_id,
+        path,
+        preview,
+        timestamp,
+        model_provider,
+        ..
+    } = summary;
+
+    let created_at = parse_datetime(timestamp.as_deref());
+
+    Thread {
+        id: conversation_id.to_string(),
+        preview,
+        model_provider,
+        created_at: created_at.map(|dt| dt.timestamp()).unwrap_or(0),
+        path,
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
     use anyhow::Result;
+    use codex_protocol::protocol::SessionSource;
     use pretty_assertions::assert_eq;
     use serde_json::json;
     use tempfile::TempDir;
@@ -2786,8 +3081,11 @@ mod tests {
             }),
         ];
 
+        let session_meta = serde_json::from_value::<SessionMeta>(head[0].clone())?;
+
         let summary =
-            extract_conversation_summary(path.clone(), &head, "test-provider").expect("summary");
+            extract_conversation_summary(path.clone(), &head, &session_meta, None, "test-provider")
+                .expect("summary");
 
         let expected = ConversationSummary {
             conversation_id,
@@ -2795,6 +3093,10 @@ mod tests {
             path,
             preview: "Count to 5".to_string(),
             model_provider: "test-provider".to_string(),
+            cwd: PathBuf::from("/"),
+            cli_version: "0.0.0".to_string(),
+            source: SessionSource::VSCode,
+            git_info: None,
         };
 
         assert_eq!(summary, expected);
@@ -2839,6 +3141,10 @@ mod tests {
             path: path.clone(),
             preview: String::new(),
             model_provider: "fallback".to_string(),
+            cwd: PathBuf::new(),
+            cli_version: String::new(),
+            source: SessionSource::VSCode,
+            git_info: None,
         };
 
         assert_eq!(summary, expected);

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

@@ -19,6 +19,7 @@ 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;
@@ -249,6 +250,15 @@ impl McpProcess {
         self.send_request("account/rateLimits/read", None).await
     }
 
+    /// Send an `account/read` JSON-RPC request.
+    pub async fn send_get_account_request(
+        &mut self,
+        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,

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

@@ -7,8 +7,6 @@ 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;

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

@@ -2,11 +2,15 @@ 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::GetAuthStatusParams;
-use codex_app_server_protocol::GetAuthStatusResponse;
+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;
@@ -15,6 +19,7 @@ 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;
@@ -25,22 +30,30 @@ 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
-fn create_config_toml(
-    codex_home: &Path,
-    forced_method: Option<&str>,
-    forced_workspace_id: Option<&str>,
-) -> std::io::Result<()> {
+#[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) = forced_method {
+    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) = forced_workspace_id {
+    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"
@@ -57,6 +70,7 @@ 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)
@@ -65,7 +79,7 @@ stream_max_retries = 0
 #[tokio::test]
 async fn logout_account_removes_auth_and_notifies() -> Result<()> {
     let codex_home = TempDir::new()?;
-    create_config_toml(codex_home.path(), None, None)?;
+    create_config_toml(codex_home.path(), CreateConfigTomlParams::default())?;
 
     login_with_api_key(
         codex_home.path(),
@@ -104,27 +118,25 @@ async fn logout_account_removes_auth_and_notifies() -> Result<()> {
         "auth.json should be deleted"
     );
 
-    let status_id = mcp
-        .send_get_auth_status_request(GetAuthStatusParams {
-            include_token: Some(true),
-            refresh_token: Some(false),
+    let get_id = mcp
+        .send_get_account_request(GetAccountParams {
+            refresh_token: false,
         })
         .await?;
-    let status_resp: JSONRPCResponse = timeout(
+    let get_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
-        mcp.read_stream_until_response_message(RequestId::Integer(status_id)),
+        mcp.read_stream_until_response_message(RequestId::Integer(get_id)),
     )
     .await??;
-    let status: GetAuthStatusResponse = to_response(status_resp)?;
-    assert_eq!(status.auth_method, None);
-    assert_eq!(status.auth_token, None);
+    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(), None, None)?;
+    create_config_toml(codex_home.path(), CreateConfigTomlParams::default())?;
 
     let mut mcp = McpProcess::new(codex_home.path()).await?;
     timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
@@ -171,7 +183,13 @@ async fn login_account_api_key_succeeds_and_notifies() -> Result<()> {
 #[tokio::test]
 async fn login_account_api_key_rejected_when_forced_chatgpt() -> Result<()> {
     let codex_home = TempDir::new()?;
-    create_config_toml(codex_home.path(), Some("chatgpt"), None)?;
+    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??;
@@ -195,7 +213,13 @@ async fn login_account_api_key_rejected_when_forced_chatgpt() -> Result<()> {
 #[tokio::test]
 async fn login_account_chatgpt_rejected_when_forced_api() -> Result<()> {
     let codex_home = TempDir::new()?;
-    create_config_toml(codex_home.path(), Some("api"), None)?;
+    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??;
@@ -219,7 +243,7 @@ async fn login_account_chatgpt_rejected_when_forced_api() -> Result<()> {
 #[serial(login_port)]
 async fn login_account_chatgpt_start() -> Result<()> {
     let codex_home = TempDir::new()?;
-    create_config_toml(codex_home.path(), None, None)?;
+    create_config_toml(codex_home.path(), CreateConfigTomlParams::default())?;
 
     let mut mcp = McpProcess::new(codex_home.path()).await?;
     timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
@@ -285,7 +309,13 @@ async fn login_account_chatgpt_start() -> Result<()> {
 #[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(), None, Some("ws-forced"))?;
+    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??;
@@ -307,3 +337,156 @@ async fn login_account_chatgpt_includes_forced_workspace_query_param() -> Result
     );
     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

@@ -1,4 +1,6 @@
 mod account;
+mod model_list;
+mod rate_limits;
 mod thread_archive;
 mod thread_list;
 mod thread_resume;

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

@@ -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?;
@@ -46,9 +46,9 @@ async fn list_models_returns_all_models_with_large_limit() -> Result<()> {
 
     let expected_models = vec![
         Model {
-            id: "gpt-5-codex".to_string(),
-            model: "gpt-5-codex".to_string(),
-            display_name: "gpt-5-codex".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 {
@@ -69,15 +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 \
@@ -106,7 +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?;
@@ -132,7 +147,7 @@ async fn list_models_pagination_works() -> Result<()> {
     } = 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
@@ -154,12 +169,34 @@ async fn list_models_pagination_works() -> Result<()> {
     } = 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?;

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

@@ -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(

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

@@ -6,10 +6,8 @@ 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 serde_json::json;
 use tempfile::TempDir;
 use tokio::time::timeout;
-use uuid::Uuid;
 
 const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
 
@@ -102,6 +100,11 @@ async fn thread_list_pagination_next_cursor_none_on_last_page() -> Result<()> {
         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.
@@ -122,6 +125,11 @@ async fn thread_list_pagination_next_cursor_none_on_last_page() -> Result<()> {
         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(())
@@ -140,46 +148,13 @@ async fn thread_list_respects_provider_filter() -> Result<()> {
         "X",
         Some("mock_provider"),
     )?; // mock_provider
-    // one with a different provider
-    let uuid = Uuid::new_v4();
-    let dir = codex_home
-        .path()
-        .join("sessions")
-        .join("2025")
-        .join("01")
-        .join("02");
-    std::fs::create_dir_all(&dir)?;
-    let file_path = dir.join(format!("rollout-2025-01-02T11-00-00-{uuid}.jsonl"));
-    let lines = [
-        json!({
-            "timestamp": "2025-01-02T11:00:00Z",
-            "type": "session_meta",
-            "payload": {
-                "id": uuid,
-                "timestamp": "2025-01-02T11:00:00Z",
-                "cwd": "/",
-                "originator": "codex",
-                "cli_version": "0.0.0",
-                "instructions": null,
-                "source": "vscode",
-                "model_provider": "other_provider"
-            }
-        })
-        .to_string(),
-        json!({
-            "timestamp": "2025-01-02T11:00:00Z",
-            "type":"response_item",
-            "payload": {"type":"message","role":"user","content":[{"type":"input_text","text":"X"}]}
-        })
-        .to_string(),
-        json!({
-            "timestamp": "2025-01-02T11:00:00Z",
-            "type":"event_msg",
-            "payload": {"type":"user_message","message":"X","kind":"plain"}
-        })
-        .to_string(),
-    ];
-    std::fs::write(file_path, lines.join("\n") + "\n")?;
+    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??;
@@ -200,6 +175,11 @@ async fn thread_list_respects_provider_filter() -> Result<()> {
     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

@@ -8,13 +8,15 @@ 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_existing_thread() -> Result<()> {
+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())?;
@@ -40,6 +42,7 @@ async fn thread_resume_returns_existing_thread() -> Result<()> {
     let resume_id = mcp
         .send_thread_resume_request(ThreadResumeParams {
             thread_id: thread.id.clone(),
+            ..Default::default()
         })
         .await?;
     let resume_resp: JSONRPCResponse = timeout(
@@ -49,7 +52,106 @@ async fn thread_resume_returns_existing_thread() -> Result<()> {
     .await??;
     let ThreadResumeResponse { thread: resumed } =
         to_response::<ThreadResumeResponse>(resume_resp)?;
-    assert_eq!(resumed.id, thread.id);
+    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(())
 }

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

@@ -42,6 +42,15 @@ async fn thread_start_creates_thread_and_emits_started() -> Result<()> {
     .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(
@@ -51,7 +60,7 @@ async fn thread_start_creates_thread_and_emits_started() -> Result<()> {
     .await??;
     let started: ThreadStartedNotification =
         serde_json::from_value(notif.params.expect("params must be present"))?;
-    assert_eq!(started.thread.id, thread.id);
+    assert_eq!(started.thread, thread);
 
     Ok(())
 }

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

@@ -30,13 +30,14 @@ 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 }
@@ -47,6 +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(
@@ -125,6 +149,8 @@ async fn run_command_under_sandbox(
             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,
@@ -167,7 +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,
@@ -199,8 +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,
@@ -504,9 +559,6 @@ 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`.

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!(

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,
@@ -27,86 +36,175 @@ 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 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,
-    },
-    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,
-    },
-    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> {
-    let allow_codex_mini = matches!(auth_mode, Some(AuthMode::ChatGPT));
+pub fn builtin_model_presets(_auth_mode: Option<AuthMode>) -> Vec<ModelPreset> {
+    // leave auth mode for later use
     PRESETS
         .iter()
-        .filter(|preset| allow_codex_mini || preset.id != "gpt-5-codex-mini")
-        .copied()
+        .filter(|preset| preset.upgrade.is_none())
+        .cloned()
         .collect()
 }
 
+pub fn all_model_presets() -> &'static Vec<ModelPreset> {
+    &PRESETS
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;

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 = [
@@ -83,7 +85,6 @@ tree-sitter-bash = { workspace = true }
 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

@@ -26,10 +26,12 @@ use crate::config::Config;
 use crate::default_client::CodexHttpClient;
 use crate::error::RefreshTokenFailedError;
 use crate::error::RefreshTokenFailedReason;
-use crate::token_data::PlanType;
+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;
 
@@ -202,7 +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)
     }
@@ -306,7 +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,
@@ -609,8 +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;
@@ -727,7 +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,
                     },
@@ -981,6 +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

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

@@ -35,10 +35,10 @@ 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;
@@ -199,12 +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()]
@@ -215,7 +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!(
@@ -294,10 +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
@@ -447,6 +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));
                         }
                     }
                 }
@@ -551,6 +560,8 @@ struct SseEvent {
     response: Option<Value>,
     item: Option<Value>,
     delta: Option<String>,
+    summary_index: Option<i64>,
+    content_index: Option<i64>,
 }
 
 #[derive(Debug, Deserialize)]
@@ -810,16 +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;
                     }
@@ -844,6 +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();
@@ -894,10 +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" => {}
@@ -975,6 +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::*;
@@ -1307,6 +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`
     // ────────────────────────────

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;
@@ -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;
@@ -129,10 +134,6 @@ 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;
-
 /// The high-level interface to the Codex system.
 /// It operates as a queue pair where you send submissions and receive events.
 pub struct Codex {
@@ -675,6 +676,34 @@ 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);
@@ -968,7 +997,7 @@ 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 +1049,7 @@ 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()
@@ -1037,7 +1066,7 @@ impl Session {
         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 +1083,37 @@ 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 +1134,7 @@ 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 +1177,11 @@ 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 +1706,7 @@ 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,
@@ -1928,6 +1990,7 @@ 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.
@@ -2169,13 +2232,17 @@ 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;
@@ -2183,18 +2250,29 @@ 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;
@@ -2320,6 +2398,7 @@ 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;
@@ -3059,6 +3138,48 @@ 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/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;
@@ -25,7 +25,7 @@ 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;
 
 pub(crate) async fn run_inline_auto_compact_task(
@@ -153,6 +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(),
     });
@@ -164,7 +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;
 }

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

@@ -23,8 +23,14 @@ 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>"]`,
@@ -239,6 +245,23 @@ 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"],
@@ -473,6 +496,27 @@ 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));
@@ -720,6 +764,63 @@ 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

@@ -241,8 +241,6 @@ 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,
 
@@ -655,7 +653,6 @@ 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>,
@@ -999,7 +996,6 @@ 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 +1152,6 @@ 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,
@@ -1715,7 +1710,6 @@ 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 +1723,11 @@ 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 +2895,6 @@ 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 +2966,6 @@ 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 +3052,6 @@ 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 +3124,6 @@ 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(),

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

@@ -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

@@ -1,12 +1,20 @@
+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;
 
-use crate::context_manager::normalize;
-use crate::context_manager::truncate::format_output_for_model_body;
-use crate::context_manager::truncate::globally_truncate_function_output_items;
+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)]
@@ -28,6 +36,10 @@ impl ContextManager {
         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),
@@ -68,6 +80,28 @@ impl ContextManager {
         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
@@ -119,7 +153,11 @@ impl ContextManager {
     fn process_item(item: &ResponseItem) -> ResponseItem {
         match item {
             ResponseItem::FunctionCallOutput { call_id, output } => {
-                let truncated = format_output_for_model_body(output.content.as_str());
+                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()
@@ -134,7 +172,11 @@ impl ContextManager {
                 }
             }
             ResponseItem::CustomToolCallOutput { call_id, output } => {
-                let truncated = format_output_for_model_body(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,

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

@@ -1,4 +1,5 @@
 use super::*;
+use crate::context_manager::MODEL_FORMAT_MAX_LINES;
 use crate::context_manager::truncate;
 use codex_git::GhostCommit;
 use codex_protocol::models::ContentItem;
@@ -308,8 +309,10 @@ fn assert_truncated_message_matches(message: &str, line: &str, total_lines: usiz
 }
 
 fn truncated_message_pattern(line: &str, total_lines: usize) -> String {
-    let head_take = truncate::MODEL_FORMAT_HEAD_LINES.min(total_lines);
-    let tail_take = truncate::MODEL_FORMAT_TAIL_LINES.min(total_lines.saturating_sub(head_take));
+    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 {
@@ -328,7 +331,11 @@ 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);
+    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);
@@ -338,7 +345,11 @@ fn format_exec_output_truncates_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);
+    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!(
@@ -359,7 +370,14 @@ fn format_exec_output_marks_byte_truncation_without_omitted_lines() {
 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), content);
+    assert_eq!(
+        truncate::format_output_for_model_body(
+            &content,
+            truncate::MODEL_FORMAT_MAX_BYTES,
+            truncate::MODEL_FORMAT_MAX_LINES
+        ),
+        content
+    );
 }
 
 #[test]
@@ -369,7 +387,11 @@ fn format_exec_output_reports_omitted_lines_and_keeps_head_and_tail() {
         .map(|idx| format!("line-{idx}\n"))
         .collect();
 
-    let truncated = truncate::format_output_for_model_body(&content);
+    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 ...]");
 
@@ -397,7 +419,11 @@ fn format_exec_output_prefers_line_marker_when_both_limits_exceeded() {
         .map(|idx| format!("line-{idx}-{long_line}\n"))
         .collect();
 
-    let truncated = truncate::format_output_for_model_body(&content);
+    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 ...]"),

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

@@ -3,4 +3,6 @@ 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/truncate.rs

@@ -2,12 +2,11 @@ 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(crate) const MODEL_FORMAT_MAX_BYTES: usize = 10 * 1024; // 10 KiB
-pub(crate) const MODEL_FORMAT_MAX_LINES: usize = 256; // lines
-pub(crate) const MODEL_FORMAT_HEAD_LINES: usize = MODEL_FORMAT_MAX_LINES / 2;
-pub(crate) const MODEL_FORMAT_TAIL_LINES: usize = MODEL_FORMAT_MAX_LINES - MODEL_FORMAT_HEAD_LINES; // 128
-pub(crate) const MODEL_FORMAT_HEAD_BYTES: usize = MODEL_FORMAT_MAX_BYTES / 2;
+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],
@@ -56,21 +55,34 @@ pub(crate) fn globally_truncate_function_output_items(
     out
 }
 
-pub(crate) fn format_output_for_model_body(content: &str) -> String {
+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() <= MODEL_FORMAT_MAX_BYTES && total_lines <= MODEL_FORMAT_MAX_LINES {
+    if content.len() <= limit_bytes && total_lines <= limit_lines {
         return content.to_string();
     }
-    let output = truncate_formatted_exec_output(content, total_lines);
+    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) -> String {
+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 = MODEL_FORMAT_HEAD_LINES.min(segments.len());
-    let tail_take = MODEL_FORMAT_TAIL_LINES.min(segments.len().saturating_sub(head_take));
+    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
@@ -91,7 +103,7 @@ fn truncate_formatted_exec_output(content: &str, total_lines: usize) -> String {
     };
     let head_slice = &content[..head_slice_end];
     let tail_slice = &content[tail_slice_start..];
-    let truncated_by_bytes = content.len() > MODEL_FORMAT_MAX_BYTES;
+    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!(
@@ -99,24 +111,24 @@ fn truncate_formatted_exec_output(content: &str, total_lines: usize) -> String {
         ))
     } else if truncated_by_bytes {
         Some(format!(
-            "\n[... output truncated to fit {MODEL_FORMAT_MAX_BYTES} bytes ...]\n\n"
+            "\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 = MODEL_FORMAT_HEAD_BYTES.min(MODEL_FORMAT_MAX_BYTES);
-    let head_budget = base_head_budget.min(MODEL_FORMAT_MAX_BYTES.saturating_sub(marker_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(MODEL_FORMAT_MAX_BYTES.min(content.len()));
+    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 = MODEL_FORMAT_MAX_BYTES.saturating_sub(result.len());
+    let remaining = limit_bytes.saturating_sub(result.len());
     if remaining == 0 {
         return result;
     }
@@ -126,3 +138,11 @@ fn truncate_formatted_exec_output(content: &str, total_lines: usize) -> String {
 
     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/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."
     )]
@@ -235,18 +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()
+            )
+        }
     }
 }
 
@@ -662,6 +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();

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

@@ -313,6 +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(),
+                ),
             }
         }
     }
@@ -514,6 +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)
@@ -521,13 +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);
 
@@ -617,6 +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::*;
@@ -689,4 +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,

codex-rs/core/src/lib.rs

@@ -81,6 +81,7 @@ 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;
@@ -99,11 +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))
         }
 

codex-rs/core/src/parse_command.rs

@@ -1,3 +1,4 @@
+use crate::bash::extract_bash_command;
 use crate::bash::try_parse_shell;
 use crate::bash::try_parse_word_only_commands_sequence;
 use codex_protocol::parse_command::ParsedCommand;
@@ -853,6 +854,29 @@ mod tests {
             }],
         );
     }
+
+    #[test]
+    fn bin_bash_lc_sed() {
+        assert_parsed(
+            &shlex_split_safe("/bin/bash -lc 'sed -n '1,10p' Cargo.toml'"),
+            vec![ParsedCommand::Read {
+                cmd: "sed -n '1,10p' Cargo.toml".to_string(),
+                name: "Cargo.toml".to_string(),
+                path: PathBuf::from("Cargo.toml"),
+            }],
+        );
+    }
+    #[test]
+    fn bin_zsh_lc_sed() {
+        assert_parsed(
+            &shlex_split_safe("/bin/zsh -lc 'sed -n '1,10p' Cargo.toml'"),
+            vec![ParsedCommand::Read {
+                cmd: "sed -n '1,10p' Cargo.toml".to_string(),
+                name: "Cargo.toml".to_string(),
+                path: PathBuf::from("Cargo.toml"),
+            }],
+        );
+    }
 }
 
 pub fn parse_command_impl(command: &[String]) -> Vec<ParsedCommand> {
@@ -1166,18 +1190,13 @@ fn parse_find_query_and_path(tail: &[String]) -> (Option<String>, Option<String>
 }
 
 fn parse_shell_lc_commands(original: &[String]) -> Option<Vec<ParsedCommand>> {
-    let [shell, flag, script] = original else {
-        return None;
-    };
-    if flag != "-lc" || !(shell == "bash" || shell == "zsh") {
-        return None;
-    }
+    let (_, script) = extract_bash_command(original)?;
+
     if let Some(tree) = try_parse_shell(script)
         && let Some(all_commands) = try_parse_word_only_commands_sequence(&tree, script)
         && !all_commands.is_empty()
     {
-        let script_tokens = shlex_split(script)
-            .unwrap_or_else(|| vec![shell.clone(), flag.clone(), script.clone()]);
+        let script_tokens = shlex_split(script).unwrap_or_else(|| vec![script.to_string()]);
         // Strip small formatting helpers (e.g., head/tail/awk/wc/etc) so we
         // bias toward the primary command when pipelines are present.
         // First, drop obvious small formatting helpers (e.g., wc/awk/etc).
@@ -1186,7 +1205,7 @@ fn parse_shell_lc_commands(original: &[String]) -> Option<Vec<ParsedCommand>> {
         let filtered_commands = drop_small_formatting_commands(all_commands);
         if filtered_commands.is_empty() {
             return Some(vec![ParsedCommand::Unknown {
-                cmd: script.clone(),
+                cmd: script.to_string(),
             }]);
         }
         // Build parsed commands, tracking `cd` segments to compute effective file paths.
@@ -1250,7 +1269,7 @@ fn parse_shell_lc_commands(original: &[String]) -> Option<Vec<ParsedCommand>> {
                             });
                             if has_pipe && has_sed_n {
                                 ParsedCommand::Read {
-                                    cmd: script.clone(),
+                                    cmd: script.to_string(),
                                     name,
                                     path,
                                 }
@@ -1295,7 +1314,7 @@ fn parse_shell_lc_commands(original: &[String]) -> Option<Vec<ParsedCommand>> {
         return Some(commands);
     }
     Some(vec![ParsedCommand::Unknown {
-        cmd: script.clone(),
+        cmd: script.to_string(),
     }])
 }
 

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

@@ -14,8 +14,11 @@ use crate::exec::StdoutStream;
 use crate::exec::execute_exec_env;
 use crate::landlock::create_linux_sandbox_command_args;
 use crate::protocol::SandboxPolicy;
+#[cfg(target_os = "macos")]
 use crate::seatbelt::MACOS_PATH_TO_SEATBELT_EXECUTABLE;
+#[cfg(target_os = "macos")]
 use crate::seatbelt::create_seatbelt_command_args;
+#[cfg(target_os = "macos")]
 use crate::spawn::CODEX_SANDBOX_ENV_VAR;
 use crate::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
 use crate::tools::sandboxing::SandboxablePreference;
@@ -56,6 +59,9 @@ pub enum SandboxPreference {
 pub(crate) enum SandboxTransformError {
     #[error("missing codex-linux-sandbox executable path")]
     MissingLinuxSandboxExecutable,
+    #[cfg(not(target_os = "macos"))]
+    #[error("seatbelt sandbox is only available on macOS")]
+    SeatbeltUnavailable,
 }
 
 #[derive(Default)]
@@ -107,6 +113,7 @@ impl SandboxManager {
 
         let (command, sandbox_env, arg0_override) = match sandbox {
             SandboxType::None => (command, HashMap::new(), None),
+            #[cfg(target_os = "macos")]
             SandboxType::MacosSeatbelt => {
                 let mut seatbelt_env = HashMap::new();
                 seatbelt_env.insert(CODEX_SANDBOX_ENV_VAR.to_string(), "seatbelt".to_string());
@@ -117,6 +124,8 @@ impl SandboxManager {
                 full_command.append(&mut args);
                 (full_command, seatbelt_env, None)
             }
+            #[cfg(not(target_os = "macos"))]
+            SandboxType::MacosSeatbelt => return Err(SandboxTransformError::SeatbeltUnavailable),
             SandboxType::LinuxSeccomp => {
                 let exe = codex_linux_sandbox_exe
                     .ok_or(SandboxTransformError::MissingLinuxSandboxExecutable)?;

codex-rs/core/src/seatbelt.rs

@@ -1,4 +1,7 @@
+#![cfg(target_os = "macos")]
+
 use std::collections::HashMap;
+use std::ffi::CStr;
 use std::path::Path;
 use std::path::PathBuf;
 use tokio::process::Child;
@@ -9,6 +12,7 @@ use crate::spawn::StdioPolicy;
 use crate::spawn::spawn_child_async;
 
 const MACOS_SEATBELT_BASE_POLICY: &str = include_str!("seatbelt_base_policy.sbpl");
+const MACOS_SEATBELT_NETWORK_POLICY: &str = include_str!("seatbelt_network_policy.sbpl");
 
 /// When working with `sandbox-exec`, only consider `sandbox-exec` in `/usr/bin`
 /// to defend against an attacker trying to inject a malicious version on the
@@ -44,27 +48,24 @@ pub(crate) fn create_seatbelt_command_args(
     sandbox_policy: &SandboxPolicy,
     sandbox_policy_cwd: &Path,
 ) -> Vec<String> {
-    let (file_write_policy, extra_cli_args) = {
+    let (file_write_policy, file_write_dir_params) = {
         if sandbox_policy.has_full_disk_write_access() {
             // Allegedly, this is more permissive than `(allow file-write*)`.
             (
                 r#"(allow file-write* (regex #"^/"))"#.to_string(),
-                Vec::<String>::new(),
+                Vec::new(),
             )
         } else {
             let writable_roots = sandbox_policy.get_writable_roots_with_cwd(sandbox_policy_cwd);
 
             let mut writable_folder_policies: Vec<String> = Vec::new();
-            let mut cli_args: Vec<String> = Vec::new();
+            let mut file_write_params = Vec::new();
 
             for (index, wr) in writable_roots.iter().enumerate() {
                 // Canonicalize to avoid mismatches like /var vs /private/var on macOS.
                 let canonical_root = wr.root.canonicalize().unwrap_or_else(|_| wr.root.clone());
                 let root_param = format!("WRITABLE_ROOT_{index}");
-                cli_args.push(format!(
-                    "-D{root_param}={}",
-                    canonical_root.to_string_lossy()
-                ));
+                file_write_params.push((root_param.clone(), canonical_root));
 
                 if wr.read_only_subpaths.is_empty() {
                     writable_folder_policies.push(format!("(subpath (param \"{root_param}\"))"));
@@ -76,9 +77,9 @@ pub(crate) fn create_seatbelt_command_args(
                     for (subpath_index, ro) in wr.read_only_subpaths.iter().enumerate() {
                         let canonical_ro = ro.canonicalize().unwrap_or_else(|_| ro.clone());
                         let ro_param = format!("WRITABLE_ROOT_{index}_RO_{subpath_index}");
-                        cli_args.push(format!("-D{ro_param}={}", canonical_ro.to_string_lossy()));
                         require_parts
                             .push(format!("(require-not (subpath (param \"{ro_param}\")))"));
+                        file_write_params.push((ro_param, canonical_ro));
                     }
                     let policy_component = format!("(require-all {} )", require_parts.join(" "));
                     writable_folder_policies.push(policy_component);
@@ -86,13 +87,13 @@ pub(crate) fn create_seatbelt_command_args(
             }
 
             if writable_folder_policies.is_empty() {
-                ("".to_string(), Vec::<String>::new())
+                ("".to_string(), Vec::new())
             } else {
                 let file_write_policy = format!(
                     "(allow file-write*\n{}\n)",
                     writable_folder_policies.join(" ")
                 );
-                (file_write_policy, cli_args)
+                (file_write_policy, file_write_params)
             }
         }
     };
@@ -105,7 +106,7 @@ pub(crate) fn create_seatbelt_command_args(
 
     // TODO(mbolin): apply_patch calls must also honor the SandboxPolicy.
     let network_policy = if sandbox_policy.has_full_network_access() {
-        "(allow network-outbound)\n(allow network-inbound)\n(allow system-socket)"
+        MACOS_SEATBELT_NETWORK_POLICY
     } else {
         ""
     };
@@ -114,17 +115,49 @@ pub(crate) fn create_seatbelt_command_args(
         "{MACOS_SEATBELT_BASE_POLICY}\n{file_read_policy}\n{file_write_policy}\n{network_policy}"
     );
 
+    let dir_params = [file_write_dir_params, macos_dir_params()].concat();
+
     let mut seatbelt_args: Vec<String> = vec!["-p".to_string(), full_policy];
-    seatbelt_args.extend(extra_cli_args);
+    let definition_args = dir_params
+        .into_iter()
+        .map(|(key, value)| format!("-D{key}={value}", value = value.to_string_lossy()));
+    seatbelt_args.extend(definition_args);
     seatbelt_args.push("--".to_string());
     seatbelt_args.extend(command);
     seatbelt_args
 }
 
+/// Wraps libc::confstr to return a String.
+fn confstr(name: libc::c_int) -> Option<String> {
+    let mut buf = vec![0_i8; (libc::PATH_MAX as usize) + 1];
+    let len = unsafe { libc::confstr(name, buf.as_mut_ptr(), buf.len()) };
+    if len == 0 {
+        return None;
+    }
+    // confstr guarantees NUL-termination when len > 0.
+    let cstr = unsafe { CStr::from_ptr(buf.as_ptr()) };
+    cstr.to_str().ok().map(ToString::to_string)
+}
+
+/// Wraps confstr to return a canonicalized PathBuf.
+fn confstr_path(name: libc::c_int) -> Option<PathBuf> {
+    let s = confstr(name)?;
+    let path = PathBuf::from(s);
+    path.canonicalize().ok().or(Some(path))
+}
+
+fn macos_dir_params() -> Vec<(String, PathBuf)> {
+    if let Some(p) = confstr_path(libc::_CS_DARWIN_USER_CACHE_DIR) {
+        return vec![("DARWIN_USER_CACHE_DIR".to_string(), p)];
+    }
+    vec![]
+}
+
 #[cfg(test)]
 mod tests {
     use super::MACOS_SEATBELT_BASE_POLICY;
     use super::create_seatbelt_command_args;
+    use super::macos_dir_params;
     use crate::protocol::SandboxPolicy;
     use pretty_assertions::assert_eq;
     use std::fs;
@@ -134,11 +167,6 @@ mod tests {
 
     #[test]
     fn create_seatbelt_args_with_read_only_git_subpath() {
-        if cfg!(target_os = "windows") {
-            // /tmp does not exist on Windows, so skip this test.
-            return;
-        }
-
         // Create a temporary workspace with two writable roots: one containing
         // a top-level .git directory and one without it.
         let tmp = TempDir::new().expect("tempdir");
@@ -199,6 +227,12 @@ mod tests {
             format!("-DWRITABLE_ROOT_2={}", cwd.to_string_lossy()),
         ];
 
+        expected_args.extend(
+            macos_dir_params()
+                .into_iter()
+                .map(|(key, value)| format!("-D{key}={value}", value = value.to_string_lossy())),
+        );
+
         expected_args.extend(vec![
             "--".to_string(),
             "/bin/echo".to_string(),
@@ -210,11 +244,6 @@ mod tests {
 
     #[test]
     fn create_seatbelt_args_for_cwd_as_git_repo() {
-        if cfg!(target_os = "windows") {
-            // /tmp does not exist on Windows, so skip this test.
-            return;
-        }
-
         // Create a temporary workspace with two writable roots: one containing
         // a top-level .git directory and one without it.
         let tmp = TempDir::new().expect("tempdir");
@@ -292,6 +321,12 @@ mod tests {
             expected_args.push(format!("-DWRITABLE_ROOT_2={p}"));
         }
 
+        expected_args.extend(
+            macos_dir_params()
+                .into_iter()
+                .map(|(key, value)| format!("-D{key}={value}", value = value.to_string_lossy())),
+        );
+
         expected_args.extend(vec![
             "--".to_string(),
             "/bin/echo".to_string(),

codex-rs/core/src/seatbelt_base_policy.sbpl

@@ -49,6 +49,7 @@
   (sysctl-name "hw.packages")
   (sysctl-name "hw.pagesize_compat")
   (sysctl-name "hw.pagesize")
+  (sysctl-name "hw.physicalcpu")
   (sysctl-name "hw.physicalcpu_max")
   (sysctl-name "hw.tbfrequency_compat")
   (sysctl-name "hw.vectorunit")

codex-rs/core/src/seatbelt_network_policy.sbpl

@@ -0,0 +1,30 @@
+; when network access is enabled, these policies are added after those in seatbelt_base_policy.sbpl
+; Ref https://source.chromium.org/chromium/chromium/src/+/main:sandbox/policy/mac/network.sb;drc=f8f264d5e4e7509c913f4c60c2639d15905a07e4
+
+(allow network-outbound)
+(allow network-inbound)
+(allow system-socket)
+
+(allow mach-lookup
+    ; Used to look up the _CS_DARWIN_USER_CACHE_DIR in the sandbox.
+    (global-name "com.apple.bsd.dirhelper")
+    (global-name "com.apple.system.opendirectoryd.membership")
+
+    ; Communicate with the security server for TLS certificate information.
+    (global-name "com.apple.SecurityServer")
+    (global-name "com.apple.networkd")
+    (global-name "com.apple.ocspd")
+    (global-name "com.apple.trustd.agent")
+
+    ; Read network configuration.
+    (global-name "com.apple.SystemConfiguration.DNSConfiguration")
+    (global-name "com.apple.SystemConfiguration.configd")
+)
+
+(allow sysctl-read
+  (sysctl-name-regex #"^net.routetable")
+)
+
+(allow file-write*
+  (subpath (param "DARWIN_USER_CACHE_DIR"))
+)

codex-rs/core/src/shell.rs

@@ -2,22 +2,26 @@ use serde::Deserialize;
 use serde::Serialize;
 use std::path::PathBuf;
 
+#[derive(Debug, PartialEq, Eq, Clone, Serialize, Deserialize)]
+pub enum ShellType {
+    Zsh,
+    Bash,
+    PowerShell,
+}
+
 #[derive(Debug, PartialEq, Eq, Clone, Serialize, Deserialize)]
 pub struct ZshShell {
-    pub(crate) shell_path: String,
-    pub(crate) zshrc_path: String,
+    pub(crate) shell_path: PathBuf,
 }
 
 #[derive(Debug, PartialEq, Eq, Clone, Serialize, Deserialize)]
 pub struct BashShell {
-    pub(crate) shell_path: String,
-    pub(crate) bashrc_path: String,
+    pub(crate) shell_path: PathBuf,
 }
 
 #[derive(Debug, PartialEq, Eq, Clone, Serialize, Deserialize)]
 pub struct PowerShellConfig {
-    pub(crate) exe: String, // Executable name or path, e.g. "pwsh" or "powershell.exe".
-    pub(crate) bash_exe_fallback: Option<PathBuf>, // In case the model generates a bash command.
+    pub(crate) shell_path: PathBuf, // Executable name or path, e.g. "pwsh" or "powershell.exe".
 }
 
 #[derive(Debug, PartialEq, Eq, Clone, Serialize, Deserialize)]
@@ -31,20 +35,51 @@ pub enum Shell {
 impl Shell {
     pub fn name(&self) -> Option<String> {
         match self {
-            Shell::Zsh(zsh) => std::path::Path::new(&zsh.shell_path)
-                .file_name()
+            Shell::Zsh(ZshShell { shell_path, .. }) | Shell::Bash(BashShell { shell_path, .. }) => {
+                std::path::Path::new(shell_path)
+                    .file_name()
+                    .map(|s| s.to_string_lossy().to_string())
+            }
+            Shell::PowerShell(ps) => ps
+                .shell_path
+                .file_stem()
                 .map(|s| s.to_string_lossy().to_string()),
-            Shell::Bash(bash) => std::path::Path::new(&bash.shell_path)
-                .file_name()
-                .map(|s| s.to_string_lossy().to_string()),
-            Shell::PowerShell(ps) => Some(ps.exe.clone()),
             Shell::Unknown => None,
         }
     }
+
+    /// Takes a string of shell and returns the full list of command args to
+    /// use with `exec()` to run the shell command.
+    pub fn derive_exec_args(&self, command: &str, use_login_shell: bool) -> Vec<String> {
+        match self {
+            Shell::Zsh(ZshShell { shell_path, .. }) | Shell::Bash(BashShell { shell_path, .. }) => {
+                let arg = if use_login_shell { "-lc" } else { "-c" };
+                vec![
+                    shell_path.to_string_lossy().to_string(),
+                    arg.to_string(),
+                    command.to_string(),
+                ]
+            }
+            Shell::PowerShell(ps) => {
+                let mut args = vec![
+                    ps.shell_path.to_string_lossy().to_string(),
+                    "-NoLogo".to_string(),
+                ];
+                if !use_login_shell {
+                    args.push("-NoProfile".to_string());
+                }
+
+                args.push("-Command".to_string());
+                args.push(command.to_string());
+                args
+            }
+            Shell::Unknown => shlex::split(command).unwrap_or_else(|| vec![command.to_string()]),
+        }
+    }
 }
 
 #[cfg(unix)]
-fn detect_default_user_shell() -> Shell {
+fn get_user_shell_path() -> Option<PathBuf> {
     use libc::getpwuid;
     use libc::getuid;
     use std::ffi::CStr;
@@ -57,75 +92,174 @@ fn detect_default_user_shell() -> Shell {
             let shell_path = CStr::from_ptr((*pw).pw_shell)
                 .to_string_lossy()
                 .into_owned();
-            let home_path = CStr::from_ptr((*pw).pw_dir).to_string_lossy().into_owned();
+            Some(PathBuf::from(shell_path))
+        } else {
+            None
+        }
+    }
+}
 
-            if shell_path.ends_with("/zsh") {
-                return Shell::Zsh(ZshShell {
-                    shell_path,
-                    zshrc_path: format!("{home_path}/.zshrc"),
-                });
-            }
+#[cfg(not(unix))]
+fn get_user_shell_path() -> Option<PathBuf> {
+    None
+}
 
-            if shell_path.ends_with("/bash") {
-                return Shell::Bash(BashShell {
-                    shell_path,
-                    bashrc_path: format!("{home_path}/.bashrc"),
-                });
+fn file_exists(path: &PathBuf) -> Option<PathBuf> {
+    if std::fs::metadata(path).is_ok_and(|metadata| metadata.is_file()) {
+        Some(PathBuf::from(path))
+    } else {
+        None
+    }
+}
+
+fn get_shell_path(
+    shell_type: ShellType,
+    provided_path: Option<&PathBuf>,
+    binary_name: &str,
+    fallback_paths: Vec<&str>,
+) -> Option<PathBuf> {
+    // If exact provided path exists, use it
+    if provided_path.and_then(file_exists).is_some() {
+        return provided_path.cloned();
+    }
+
+    // Check if the shell we are trying to load is user's default shell
+    // if just use it
+    let default_shell_path = get_user_shell_path();
+    if let Some(default_shell_path) = default_shell_path
+        && detect_shell_type(&default_shell_path) == Some(shell_type)
+    {
+        return Some(default_shell_path);
+    }
+
+    if let Ok(path) = which::which(binary_name) {
+        return Some(path);
+    }
+
+    for path in fallback_paths {
+        //check exists
+        if let Some(path) = file_exists(&PathBuf::from(path)) {
+            return Some(path);
+        }
+    }
+
+    None
+}
+
+fn get_zsh_shell(path: Option<&PathBuf>) -> Option<ZshShell> {
+    let shell_path = get_shell_path(ShellType::Zsh, path, "zsh", vec!["/bin/zsh"]);
+
+    shell_path.map(|shell_path| ZshShell { shell_path })
+}
+
+fn get_bash_shell(path: Option<&PathBuf>) -> Option<BashShell> {
+    let shell_path = get_shell_path(ShellType::Bash, path, "bash", vec!["/bin/bash"]);
+
+    shell_path.map(|shell_path| BashShell { shell_path })
+}
+
+fn get_powershell_shell(path: Option<&PathBuf>) -> Option<PowerShellConfig> {
+    let shell_path = get_shell_path(
+        ShellType::PowerShell,
+        path,
+        "pwsh",
+        vec!["/usr/local/bin/pwsh"],
+    )
+    .or_else(|| get_shell_path(ShellType::PowerShell, path, "powershell", vec![]));
+
+    shell_path.map(|shell_path| PowerShellConfig { shell_path })
+}
+
+pub fn get_shell_by_model_provided_path(shell_path: &PathBuf) -> Shell {
+    detect_shell_type(shell_path)
+        .and_then(|shell_type| get_shell(shell_type, Some(shell_path)))
+        .unwrap_or(Shell::Unknown)
+}
+
+pub fn get_shell(shell_type: ShellType, path: Option<&PathBuf>) -> Option<Shell> {
+    match shell_type {
+        ShellType::Zsh => get_zsh_shell(path).map(Shell::Zsh),
+        ShellType::Bash => get_bash_shell(path).map(Shell::Bash),
+        ShellType::PowerShell => get_powershell_shell(path).map(Shell::PowerShell),
+    }
+}
+
+pub fn detect_shell_type(shell_path: &PathBuf) -> Option<ShellType> {
+    match shell_path.as_os_str().to_str() {
+        Some("zsh") => Some(ShellType::Zsh),
+        Some("bash") => Some(ShellType::Bash),
+        Some("pwsh") => Some(ShellType::PowerShell),
+        Some("powershell") => Some(ShellType::PowerShell),
+        _ => {
+            let shell_name = shell_path.file_stem();
+
+            if let Some(shell_name) = shell_name
+                && shell_name != shell_path
+            {
+                detect_shell_type(&PathBuf::from(shell_name))
+            } else {
+                None
             }
         }
     }
-    Shell::Unknown
 }
 
-#[cfg(unix)]
 pub async fn default_user_shell() -> Shell {
-    detect_default_user_shell()
-}
-
-#[cfg(target_os = "windows")]
-pub async fn default_user_shell() -> Shell {
-    use tokio::process::Command;
-
-    // Prefer PowerShell 7+ (`pwsh`) if available, otherwise fall back to Windows PowerShell.
-    let has_pwsh = Command::new("pwsh")
-        .arg("-NoLogo")
-        .arg("-NoProfile")
-        .arg("-Command")
-        .arg("$PSVersionTable.PSVersion.Major")
-        .output()
-        .await
-        .map(|o| o.status.success())
-        .unwrap_or(false);
-    let bash_exe = if Command::new("bash.exe")
-        .arg("--version")
-        .stdin(std::process::Stdio::null())
-        .output()
-        .await
-        .ok()
-        .map(|o| o.status.success())
-        .unwrap_or(false)
-    {
-        which::which("bash.exe").ok()
+    if cfg!(windows) {
+        get_shell(ShellType::PowerShell, None).unwrap_or(Shell::Unknown)
     } else {
-        None
-    };
-
-    if has_pwsh {
-        Shell::PowerShell(PowerShellConfig {
-            exe: "pwsh.exe".to_string(),
-            bash_exe_fallback: bash_exe,
-        })
-    } else {
-        Shell::PowerShell(PowerShellConfig {
-            exe: "powershell.exe".to_string(),
-            bash_exe_fallback: bash_exe,
-        })
+        get_user_shell_path()
+            .and_then(|shell| detect_shell_type(&shell))
+            .and_then(|shell_type| get_shell(shell_type, None))
+            .unwrap_or(Shell::Unknown)
     }
 }
 
-#[cfg(all(not(target_os = "windows"), not(unix)))]
-pub async fn default_user_shell() -> Shell {
-    Shell::Unknown
+#[cfg(test)]
+mod detect_shell_type_tests {
+    use super::*;
+
+    #[test]
+    fn test_detect_shell_type() {
+        assert_eq!(
+            detect_shell_type(&PathBuf::from("zsh")),
+            Some(ShellType::Zsh)
+        );
+        assert_eq!(
+            detect_shell_type(&PathBuf::from("bash")),
+            Some(ShellType::Bash)
+        );
+        assert_eq!(
+            detect_shell_type(&PathBuf::from("pwsh")),
+            Some(ShellType::PowerShell)
+        );
+        assert_eq!(
+            detect_shell_type(&PathBuf::from("powershell")),
+            Some(ShellType::PowerShell)
+        );
+        assert_eq!(detect_shell_type(&PathBuf::from("fish")), None);
+        assert_eq!(detect_shell_type(&PathBuf::from("other")), None);
+        assert_eq!(
+            detect_shell_type(&PathBuf::from("/bin/zsh")),
+            Some(ShellType::Zsh)
+        );
+        assert_eq!(
+            detect_shell_type(&PathBuf::from("/bin/bash")),
+            Some(ShellType::Bash)
+        );
+        assert_eq!(
+            detect_shell_type(&PathBuf::from("powershell.exe")),
+            Some(ShellType::PowerShell)
+        );
+        assert_eq!(
+            detect_shell_type(&PathBuf::from("pwsh.exe")),
+            Some(ShellType::PowerShell)
+        );
+        assert_eq!(
+            detect_shell_type(&PathBuf::from("/usr/local/bin/pwsh")),
+            Some(ShellType::PowerShell)
+        );
+    }
 }
 
 #[cfg(test)]
@@ -135,6 +269,34 @@ mod tests {
     use std::path::PathBuf;
     use std::process::Command;
 
+    #[test]
+    #[cfg(target_os = "macos")]
+    fn detects_zsh() {
+        let zsh_shell = get_shell(ShellType::Zsh, None).unwrap();
+
+        let ZshShell { shell_path } = match zsh_shell {
+            Shell::Zsh(zsh_shell) => zsh_shell,
+            _ => panic!("expected zsh shell"),
+        };
+
+        assert_eq!(shell_path, PathBuf::from("/bin/zsh"));
+    }
+
+    #[test]
+    fn detects_bash() {
+        let bash_shell = get_shell(ShellType::Bash, None).unwrap();
+        let BashShell { shell_path } = match bash_shell {
+            Shell::Bash(bash_shell) => bash_shell,
+            _ => panic!("expected bash shell"),
+        };
+
+        assert!(
+            shell_path == PathBuf::from("/bin/bash")
+                || shell_path == PathBuf::from("/usr/bin/bash"),
+            "shell path: {shell_path:?}",
+        );
+    }
+
     #[tokio::test]
     async fn test_current_shell_detects_zsh() {
         let shell = Command::new("sh")
@@ -143,292 +305,44 @@ mod tests {
             .output()
             .unwrap();
 
-        let home = std::env::var("HOME").unwrap();
         let shell_path = String::from_utf8_lossy(&shell.stdout).trim().to_string();
         if shell_path.ends_with("/zsh") {
             assert_eq!(
                 default_user_shell().await,
                 Shell::Zsh(ZshShell {
-                    shell_path: shell_path.to_string(),
-                    zshrc_path: format!("{home}/.zshrc",),
+                    shell_path: PathBuf::from(shell_path),
                 })
             );
         }
     }
 
     #[tokio::test]
-    async fn test_run_with_profile_bash_escaping_and_execution() {
-        let shell_path = "/bin/bash";
-
-        let cases = vec![
-            (
-                vec!["myecho"],
-                vec![shell_path, "-lc", "source BASHRC_PATH && (myecho)"],
-                Some("It works!\n"),
-            ),
-            (
-                vec!["bash", "-lc", "echo 'single' \"double\""],
-                vec![
-                    shell_path,
-                    "-lc",
-                    "source BASHRC_PATH && (echo 'single' \"double\")",
-                ],
-                Some("single double\n"),
-            ),
-        ];
-
-        for (input, expected_cmd, expected_output) in cases {
-            use std::collections::HashMap;
-
-            use crate::exec::ExecParams;
-            use crate::exec::SandboxType;
-            use crate::exec::process_exec_tool_call;
-            use crate::protocol::SandboxPolicy;
-
-            let temp_home = tempfile::tempdir().unwrap();
-            let bashrc_path = temp_home.path().join(".bashrc");
-            std::fs::write(
-                &bashrc_path,
-                r#"
-                set -x
-                function myecho {
-                    echo 'It works!'
-                }
-                "#,
-            )
-            .unwrap();
-            let command = expected_cmd
-                .iter()
-                .map(|s| s.replace("BASHRC_PATH", bashrc_path.to_str().unwrap()))
-                .collect::<Vec<_>>();
-
-            let output = process_exec_tool_call(
-                ExecParams {
-                    command: command.clone(),
-                    cwd: PathBuf::from(temp_home.path()),
-                    timeout_ms: None,
-                    env: HashMap::from([(
-                        "HOME".to_string(),
-                        temp_home.path().to_str().unwrap().to_string(),
-                    )]),
-                    with_escalated_permissions: None,
-                    justification: None,
-                    arg0: None,
-                },
-                SandboxType::None,
-                &SandboxPolicy::DangerFullAccess,
-                temp_home.path(),
-                &None,
-                None,
-            )
-            .await
-            .unwrap();
-
-            assert_eq!(output.exit_code, 0, "input: {input:?} output: {output:?}");
-            if let Some(expected) = expected_output {
-                assert_eq!(
-                    output.stdout.text, expected,
-                    "input: {input:?} output: {output:?}"
-                );
-            }
+    async fn detects_powershell_as_default() {
+        if !cfg!(windows) {
+            return;
         }
+
+        let powershell_shell = default_user_shell().await;
+        let PowerShellConfig { shell_path } = match powershell_shell {
+            Shell::PowerShell(powershell_shell) => powershell_shell,
+            _ => panic!("expected powershell shell"),
+        };
+
+        assert!(shell_path.ends_with("pwsh.exe") || shell_path.ends_with("powershell.exe"));
     }
-}
-
-#[cfg(test)]
-#[cfg(target_os = "macos")]
-mod macos_tests {
-    use std::path::PathBuf;
-
-    #[tokio::test]
-    async fn test_run_with_profile_escaping_and_execution() {
-        let shell_path = "/bin/zsh";
-
-        let cases = vec![
-            (
-                vec!["myecho"],
-                vec![shell_path, "-lc", "source ZSHRC_PATH && (myecho)"],
-                Some("It works!\n"),
-            ),
-            (
-                vec!["myecho"],
-                vec![shell_path, "-lc", "source ZSHRC_PATH && (myecho)"],
-                Some("It works!\n"),
-            ),
-            (
-                vec!["bash", "-c", "echo 'single' \"double\""],
-                vec![
-                    shell_path,
-                    "-lc",
-                    "source ZSHRC_PATH && (bash -c \"echo 'single' \\\"double\\\"\")",
-                ],
-                Some("single double\n"),
-            ),
-            (
-                vec!["bash", "-lc", "echo 'single' \"double\""],
-                vec![
-                    shell_path,
-                    "-lc",
-                    "source ZSHRC_PATH && (echo 'single' \"double\")",
-                ],
-                Some("single double\n"),
-            ),
-        ];
-        for (input, expected_cmd, expected_output) in cases {
-            use std::collections::HashMap;
-
-            use crate::exec::ExecParams;
-            use crate::exec::SandboxType;
-            use crate::exec::process_exec_tool_call;
-            use crate::protocol::SandboxPolicy;
-
-            let temp_home = tempfile::tempdir().unwrap();
-            let zshrc_path = temp_home.path().join(".zshrc");
-            std::fs::write(
-                &zshrc_path,
-                r#"
-                set -x
-                function myecho {
-                    echo 'It works!'
-                }
-                "#,
-            )
-            .unwrap();
-            let command = expected_cmd
-                .iter()
-                .map(|s| s.replace("ZSHRC_PATH", zshrc_path.to_str().unwrap()))
-                .collect::<Vec<_>>();
-
-            let output = process_exec_tool_call(
-                ExecParams {
-                    command: command.clone(),
-                    cwd: PathBuf::from(temp_home.path()),
-                    timeout_ms: None,
-                    env: HashMap::from([(
-                        "HOME".to_string(),
-                        temp_home.path().to_str().unwrap().to_string(),
-                    )]),
-                    with_escalated_permissions: None,
-                    justification: None,
-                    arg0: None,
-                },
-                SandboxType::None,
-                &SandboxPolicy::DangerFullAccess,
-                temp_home.path(),
-                &None,
-                None,
-            )
-            .await
-            .unwrap();
-
-            assert_eq!(output.exit_code, 0, "input: {input:?} output: {output:?}");
-            if let Some(expected) = expected_output {
-                assert_eq!(
-                    output.stdout.text, expected,
-                    "input: {input:?} output: {output:?}"
-                );
-            }
-        }
-    }
-}
-
-#[cfg(test)]
-#[cfg(target_os = "windows")]
-mod tests_windows {
-    use super::*;
 
     #[test]
-    fn test_format_default_shell_invocation_powershell() {
-        use std::path::PathBuf;
-
-        let cases = vec![
-            (
-                PowerShellConfig {
-                    exe: "pwsh.exe".to_string(),
-                    bash_exe_fallback: None,
-                },
-                vec!["bash", "-lc", "echo hello"],
-                vec!["pwsh.exe", "-NoProfile", "-Command", "echo hello"],
-            ),
-            (
-                PowerShellConfig {
-                    exe: "powershell.exe".to_string(),
-                    bash_exe_fallback: None,
-                },
-                vec!["bash", "-lc", "echo hello"],
-                vec!["powershell.exe", "-NoProfile", "-Command", "echo hello"],
-            ),
-            (
-                PowerShellConfig {
-                    exe: "pwsh.exe".to_string(),
-                    bash_exe_fallback: Some(PathBuf::from("bash.exe")),
-                },
-                vec!["bash", "-lc", "echo hello"],
-                vec!["bash.exe", "-lc", "echo hello"],
-            ),
-            (
-                PowerShellConfig {
-                    exe: "pwsh.exe".to_string(),
-                    bash_exe_fallback: Some(PathBuf::from("bash.exe")),
-                },
-                vec![
-                    "bash",
-                    "-lc",
-                    "apply_patch <<'EOF'\n*** Begin Patch\n*** Update File: destination_file.txt\n-original content\n+modified content\n*** End Patch\nEOF",
-                ],
-                vec![
-                    "bash.exe",
-                    "-lc",
-                    "apply_patch <<'EOF'\n*** Begin Patch\n*** Update File: destination_file.txt\n-original content\n+modified content\n*** End Patch\nEOF",
-                ],
-            ),
-            (
-                PowerShellConfig {
-                    exe: "pwsh.exe".to_string(),
-                    bash_exe_fallback: Some(PathBuf::from("bash.exe")),
-                },
-                vec!["echo", "hello"],
-                vec!["pwsh.exe", "-NoProfile", "-Command", "echo hello"],
-            ),
-            (
-                PowerShellConfig {
-                    exe: "pwsh.exe".to_string(),
-                    bash_exe_fallback: Some(PathBuf::from("bash.exe")),
-                },
-                vec!["pwsh.exe", "-NoProfile", "-Command", "echo hello"],
-                vec!["pwsh.exe", "-NoProfile", "-Command", "echo hello"],
-            ),
-            (
-                PowerShellConfig {
-                    exe: "powershell.exe".to_string(),
-                    bash_exe_fallback: Some(PathBuf::from("bash.exe")),
-                },
-                vec![
-                    "codex-mcp-server.exe",
-                    "--codex-run-as-apply-patch",
-                    "*** Begin Patch\n*** Update File: C:\\Users\\person\\destination_file.txt\n-original content\n+modified content\n*** End Patch",
-                ],
-                vec![
-                    "codex-mcp-server.exe",
-                    "--codex-run-as-apply-patch",
-                    "*** Begin Patch\n*** Update File: C:\\Users\\person\\destination_file.txt\n-original content\n+modified content\n*** End Patch",
-                ],
-            ),
-        ];
-
-        for (config, input, expected_cmd) in cases {
-            let command = expected_cmd
-                .iter()
-                .map(|s| (*s).to_string())
-                .collect::<Vec<_>>();
-
-            // These tests assert the final command for each scenario now that the helper
-            // has been removed. The inputs remain to document the original coverage.
-            let expected = expected_cmd
-                .iter()
-                .map(|s| (*s).to_string())
-                .collect::<Vec<_>>();
-            assert_eq!(command, expected, "input: {input:?} config: {config:?}");
+    fn finds_poweshell() {
+        if !cfg!(windows) {
+            return;
         }
+
+        let powershell_shell = get_shell(ShellType::PowerShell, None).unwrap();
+        let PowerShellConfig { shell_path } = match powershell_shell {
+            Shell::PowerShell(powershell_shell) => powershell_shell,
+            _ => panic!("expected powershell shell"),
+        };
+
+        assert!(shell_path.ends_with("pwsh.exe") || shell_path.ends_with("powershell.exe"));
     }
 }

codex-rs/core/src/spawn.rs

@@ -64,24 +64,32 @@ pub(crate) async fn spawn_child_async(
     // any child processes that were spawned as part of a `"shell"` tool call
     // to also be terminated.
 
-    // This relies on prctl(2), so it only works on Linux.
-    #[cfg(target_os = "linux")]
+    #[cfg(unix)]
     unsafe {
+        #[cfg(target_os = "linux")]
         let parent_pid = libc::getpid();
         cmd.pre_exec(move || {
-            // This prctl call effectively requests, "deliver SIGTERM when my
-            // current parent dies."
-            if libc::prctl(libc::PR_SET_PDEATHSIG, libc::SIGTERM) == -1 {
+            if libc::setpgid(0, 0) == -1 {
                 return Err(std::io::Error::last_os_error());
             }
 
-            // Though if there was a race condition and this pre_exec() block is
-            // run _after_ the parent (i.e., the Codex process) has already
-            // exited, then parent will be the closest configured "subreaper"
-            // ancestor process, or PID 1 (init). If the Codex process has exited
-            // already, so should the child process.
-            if libc::getppid() != parent_pid {
-                libc::raise(libc::SIGTERM);
+            // This relies on prctl(2), so it only works on Linux.
+            #[cfg(target_os = "linux")]
+            {
+                // This prctl call effectively requests, "deliver SIGTERM when my
+                // current parent dies."
+                if libc::prctl(libc::PR_SET_PDEATHSIG, libc::SIGTERM) == -1 {
+                    return Err(std::io::Error::last_os_error());
+                }
+
+                // Though if there was a race condition and this pre_exec() block is
+                // run _after_ the parent (i.e., the Codex process) has already
+                // exited, then parent will be the closest configured "subreaper"
+                // ancestor process, or PID 1 (init). If the Codex process has exited
+                // already, so should the child process.
+                if libc::getppid() != parent_pid {
+                    libc::raise(libc::SIGTERM);
+                }
             }
             Ok(())
         });

codex-rs/core/src/state/session.rs

@@ -42,6 +42,10 @@ impl SessionState {
         self.history.replace(items);
     }
 
+    pub(crate) fn set_token_info(&mut self, info: Option<TokenUsageInfo>) {
+        self.history.set_token_info(info);
+    }
+
     // Token/rate limit helpers
     pub(crate) fn update_token_info_from_usage(
         &mut self,

codex-rs/core/src/tasks/compact.rs

@@ -4,7 +4,7 @@ use async_trait::async_trait;
 use tokio_util::sync::CancellationToken;
 
 use crate::codex::TurnContext;
-use crate::codex::compact;
+use crate::compact;
 use crate::state::TaskKind;
 use codex_protocol::user_input::UserInput;
 

codex-rs/core/src/tasks/review.rs

@@ -75,12 +75,12 @@ async fn start_review_conversation(
     // Avoid loading project docs; reviewer only needs findings
     sub_agent_config.project_doc_max_bytes = 0;
     // Carry over review-only feature restrictions so the delegate cannot
-    // re-enable blocked tools (web search, view image, streamable shell).
+    // re-enable blocked tools (web search, view image).
     sub_agent_config
         .features
         .disable(crate::features::Feature::WebSearchRequest)
-        .disable(crate::features::Feature::ViewImageTool)
-        .disable(crate::features::Feature::StreamableShell);
+        .disable(crate::features::Feature::ViewImageTool);
+
     // Set explicit review rubric for the sub-agent
     sub_agent_config.base_instructions = Some(crate::REVIEW_PROMPT.to_string());
     (run_codex_conversation_one_shot(

codex-rs/core/src/tasks/user_shell.rs

@@ -1,28 +1,35 @@
 use std::sync::Arc;
+use std::time::Duration;
 
 use async_trait::async_trait;
-use codex_protocol::models::ShellToolCallParams;
+use codex_async_utils::CancelErr;
+use codex_async_utils::OrCancelExt;
 use codex_protocol::user_input::UserInput;
-use tokio::sync::Mutex;
 use tokio_util::sync::CancellationToken;
 use tracing::error;
 use uuid::Uuid;
 
 use crate::codex::TurnContext;
+use crate::exec::ExecToolCallOutput;
+use crate::exec::SandboxType;
+use crate::exec::StdoutStream;
+use crate::exec::StreamOutput;
+use crate::exec::execute_exec_env;
+use crate::exec_env::create_env;
+use crate::parse_command::parse_command;
 use crate::protocol::EventMsg;
+use crate::protocol::ExecCommandBeginEvent;
+use crate::protocol::ExecCommandEndEvent;
+use crate::protocol::SandboxPolicy;
 use crate::protocol::TaskStartedEvent;
+use crate::sandboxing::ExecEnv;
 use crate::state::TaskKind;
-use crate::tools::context::ToolPayload;
-use crate::tools::parallel::ToolCallRuntime;
-use crate::tools::router::ToolCall;
-use crate::tools::router::ToolRouter;
-use crate::turn_diff_tracker::TurnDiffTracker;
+use crate::tools::format_exec_output_str;
+use crate::user_shell_command::user_shell_command_record_item;
 
 use super::SessionTask;
 use super::SessionTaskContext;
 
-const USER_SHELL_TOOL_NAME: &str = "local_shell";
-
 #[derive(Clone)]
 pub(crate) struct UserShellCommandTask {
     command: String,
@@ -56,56 +63,131 @@ impl SessionTask for UserShellCommandTask {
         // Execute the user's script under their default shell when known; this
         // allows commands that use shell features (pipes, &&, redirects, etc.).
         // We do not source rc files or otherwise reformat the script.
-        let shell_invocation = match session.user_shell() {
-            crate::shell::Shell::Zsh(zsh) => vec![
-                zsh.shell_path.clone(),
-                "-lc".to_string(),
-                self.command.clone(),
-            ],
-            crate::shell::Shell::Bash(bash) => vec![
-                bash.shell_path.clone(),
-                "-lc".to_string(),
-                self.command.clone(),
-            ],
-            crate::shell::Shell::PowerShell(ps) => vec![
-                ps.exe.clone(),
-                "-NoProfile".to_string(),
-                "-Command".to_string(),
-                self.command.clone(),
-            ],
-            crate::shell::Shell::Unknown => {
-                shlex::split(&self.command).unwrap_or_else(|| vec![self.command.clone()])
-            }
-        };
+        let use_login_shell = true;
+        let shell_invocation = session
+            .user_shell()
+            .derive_exec_args(&self.command, use_login_shell);
 
-        let params = ShellToolCallParams {
+        let call_id = Uuid::new_v4().to_string();
+        let raw_command = self.command.clone();
+
+        let parsed_cmd = parse_command(&shell_invocation);
+        session
+            .send_event(
+                turn_context.as_ref(),
+                EventMsg::ExecCommandBegin(ExecCommandBeginEvent {
+                    call_id: call_id.clone(),
+                    command: shell_invocation.clone(),
+                    cwd: turn_context.cwd.clone(),
+                    parsed_cmd,
+                    is_user_shell_command: true,
+                }),
+            )
+            .await;
+
+        let exec_env = ExecEnv {
             command: shell_invocation,
-            workdir: None,
+            cwd: turn_context.cwd.clone(),
+            env: create_env(&turn_context.shell_environment_policy),
             timeout_ms: None,
+            sandbox: SandboxType::None,
             with_escalated_permissions: None,
             justification: None,
+            arg0: None,
         };
 
-        let tool_call = ToolCall {
-            tool_name: USER_SHELL_TOOL_NAME.to_string(),
-            call_id: Uuid::new_v4().to_string(),
-            payload: ToolPayload::LocalShell { params },
-        };
+        let stdout_stream = Some(StdoutStream {
+            sub_id: turn_context.sub_id.clone(),
+            call_id: call_id.clone(),
+            tx_event: session.get_tx_event(),
+        });
 
-        let router = Arc::new(ToolRouter::from_config(&turn_context.tools_config, None));
-        let tracker = Arc::new(Mutex::new(TurnDiffTracker::new()));
-        let runtime = ToolCallRuntime::new(
-            Arc::clone(&router),
-            Arc::clone(&session),
-            Arc::clone(&turn_context),
-            Arc::clone(&tracker),
-        );
+        let sandbox_policy = SandboxPolicy::DangerFullAccess;
+        let exec_result = execute_exec_env(exec_env, &sandbox_policy, stdout_stream)
+            .or_cancel(&cancellation_token)
+            .await;
 
-        if let Err(err) = runtime
-            .handle_tool_call(tool_call, cancellation_token)
-            .await
-        {
-            error!("user shell command failed: {err:?}");
+        match exec_result {
+            Err(CancelErr::Cancelled) => {
+                let aborted_message = "command aborted by user".to_string();
+                let exec_output = ExecToolCallOutput {
+                    exit_code: -1,
+                    stdout: StreamOutput::new(String::new()),
+                    stderr: StreamOutput::new(aborted_message.clone()),
+                    aggregated_output: StreamOutput::new(aborted_message.clone()),
+                    duration: Duration::ZERO,
+                    timed_out: false,
+                };
+                let output_items = [user_shell_command_record_item(&raw_command, &exec_output)];
+                session
+                    .record_conversation_items(turn_context.as_ref(), &output_items)
+                    .await;
+                session
+                    .send_event(
+                        turn_context.as_ref(),
+                        EventMsg::ExecCommandEnd(ExecCommandEndEvent {
+                            call_id,
+                            stdout: String::new(),
+                            stderr: aborted_message.clone(),
+                            aggregated_output: aborted_message.clone(),
+                            exit_code: -1,
+                            duration: Duration::ZERO,
+                            formatted_output: aborted_message,
+                        }),
+                    )
+                    .await;
+            }
+            Ok(Ok(output)) => {
+                session
+                    .send_event(
+                        turn_context.as_ref(),
+                        EventMsg::ExecCommandEnd(ExecCommandEndEvent {
+                            call_id: call_id.clone(),
+                            stdout: output.stdout.text.clone(),
+                            stderr: output.stderr.text.clone(),
+                            aggregated_output: output.aggregated_output.text.clone(),
+                            exit_code: output.exit_code,
+                            duration: output.duration,
+                            formatted_output: format_exec_output_str(&output),
+                        }),
+                    )
+                    .await;
+
+                let output_items = [user_shell_command_record_item(&raw_command, &output)];
+                session
+                    .record_conversation_items(turn_context.as_ref(), &output_items)
+                    .await;
+            }
+            Ok(Err(err)) => {
+                error!("user shell command failed: {err:?}");
+                let message = format!("execution error: {err:?}");
+                let exec_output = ExecToolCallOutput {
+                    exit_code: -1,
+                    stdout: StreamOutput::new(String::new()),
+                    stderr: StreamOutput::new(message.clone()),
+                    aggregated_output: StreamOutput::new(message.clone()),
+                    duration: Duration::ZERO,
+                    timed_out: false,
+                };
+                session
+                    .send_event(
+                        turn_context.as_ref(),
+                        EventMsg::ExecCommandEnd(ExecCommandEndEvent {
+                            call_id,
+                            stdout: exec_output.stdout.text.clone(),
+                            stderr: exec_output.stderr.text.clone(),
+                            aggregated_output: exec_output.aggregated_output.text.clone(),
+                            exit_code: exec_output.exit_code,
+                            duration: exec_output.duration,
+                            formatted_output: format_exec_output_str(&exec_output),
+                        }),
+                    )
+                    .await;
+                let output_items = [user_shell_command_record_item(&raw_command, &exec_output)];
+                session
+                    .record_conversation_items(turn_context.as_ref(), &output_items)
+                    .await;
+            }
         }
         None
     }

codex-rs/core/src/tools/events.rs

@@ -87,7 +87,7 @@ pub(crate) enum ToolEmitter {
         auto_approved: bool,
     },
     UnifiedExec {
-        command: String,
+        command: Vec<String>,
         cwd: PathBuf,
         // True for `exec_command` and false for `write_stdin`.
         #[allow(dead_code)]
@@ -111,9 +111,9 @@ impl ToolEmitter {
         }
     }
 
-    pub fn unified_exec(command: String, cwd: PathBuf, is_startup_command: bool) -> Self {
+    pub fn unified_exec(command: &[String], cwd: PathBuf, is_startup_command: bool) -> Self {
         Self::UnifiedExec {
-            command,
+            command: command.to_vec(),
             cwd,
             is_startup_command,
         }
@@ -218,7 +218,7 @@ impl ToolEmitter {
                 emit_patch_end(ctx, String::new(), (*message).to_string(), false).await;
             }
             (Self::UnifiedExec { command, cwd, .. }, ToolEventStage::Begin) => {
-                emit_exec_command_begin(ctx, &[command.to_string()], cwd.as_path(), false).await;
+                emit_exec_command_begin(ctx, command, cwd.as_path(), false).await;
             }
             (Self::UnifiedExec { .. }, ToolEventStage::Success(output)) => {
                 emit_exec_end(

codex-rs/core/src/tools/handlers/apply_patch.rs

@@ -42,6 +42,10 @@ impl ToolHandler for ApplyPatchHandler {
         )
     }
 
+    fn is_mutating(&self, _invocation: &ToolInvocation) -> bool {
+        true
+    }
+
     async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError> {
         let ToolInvocation {
             session,

codex-rs/core/src/tools/handlers/mod.rs

@@ -19,6 +19,7 @@ pub use mcp::McpHandler;
 pub use mcp_resource::McpResourceHandler;
 pub use plan::PlanHandler;
 pub use read_file::ReadFileHandler;
+pub use shell::ShellCommandHandler;
 pub use shell::ShellHandler;
 pub use test_sync::TestSyncHandler;
 pub use unified_exec::UnifiedExecHandler;

codex-rs/core/src/tools/handlers/shell.rs

@@ -1,4 +1,5 @@
 use async_trait::async_trait;
+use codex_protocol::models::ShellCommandToolCallParams;
 use codex_protocol::models::ShellToolCallParams;
 use std::sync::Arc;
 
@@ -9,6 +10,7 @@ use crate::codex::TurnContext;
 use crate::exec::ExecParams;
 use crate::exec_env::create_env;
 use crate::function_tool::FunctionCallError;
+use crate::is_safe_command::is_known_safe_command;
 use crate::tools::context::ToolInvocation;
 use crate::tools::context::ToolOutput;
 use crate::tools::context::ToolPayload;
@@ -25,6 +27,8 @@ use crate::tools::sandboxing::ToolCtx;
 
 pub struct ShellHandler;
 
+pub struct ShellCommandHandler;
+
 impl ShellHandler {
     fn to_exec_params(params: ShellToolCallParams, turn_context: &TurnContext) -> ExecParams {
         ExecParams {
@@ -39,6 +43,28 @@ impl ShellHandler {
     }
 }
 
+impl ShellCommandHandler {
+    fn to_exec_params(
+        params: ShellCommandToolCallParams,
+        session: &crate::codex::Session,
+        turn_context: &TurnContext,
+    ) -> ExecParams {
+        let shell = session.user_shell();
+        let use_login_shell = true;
+        let command = shell.derive_exec_args(&params.command, use_login_shell);
+
+        ExecParams {
+            command,
+            cwd: turn_context.resolve_path(params.workdir.clone()),
+            timeout_ms: params.timeout_ms,
+            env: create_env(&turn_context.shell_environment_policy),
+            with_escalated_permissions: params.with_escalated_permissions,
+            justification: params.justification,
+            arg0: None,
+        }
+    }
+}
+
 #[async_trait]
 impl ToolHandler for ShellHandler {
     fn kind(&self) -> ToolKind {
@@ -52,6 +78,18 @@ impl ToolHandler for ShellHandler {
         )
     }
 
+    fn is_mutating(&self, invocation: &ToolInvocation) -> bool {
+        match &invocation.payload {
+            ToolPayload::Function { arguments } => {
+                serde_json::from_str::<ShellToolCallParams>(arguments)
+                    .map(|params| !is_known_safe_command(&params.command))
+                    .unwrap_or(true)
+            }
+            ToolPayload::LocalShell { params } => !is_known_safe_command(&params.command),
+            _ => true, // unknown payloads => assume mutating
+        }
+    }
+
     async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError> {
         let ToolInvocation {
             session,
@@ -102,6 +140,49 @@ impl ToolHandler for ShellHandler {
     }
 }
 
+#[async_trait]
+impl ToolHandler for ShellCommandHandler {
+    fn kind(&self) -> ToolKind {
+        ToolKind::Function
+    }
+
+    fn matches_kind(&self, payload: &ToolPayload) -> bool {
+        matches!(payload, ToolPayload::Function { .. })
+    }
+
+    async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError> {
+        let ToolInvocation {
+            session,
+            turn,
+            tracker,
+            call_id,
+            tool_name,
+            payload,
+        } = invocation;
+
+        let ToolPayload::Function { arguments } = payload else {
+            return Err(FunctionCallError::RespondToModel(format!(
+                "unsupported payload for shell_command handler: {tool_name}"
+            )));
+        };
+
+        let params: ShellCommandToolCallParams = serde_json::from_str(&arguments).map_err(|e| {
+            FunctionCallError::RespondToModel(format!("failed to parse function arguments: {e:?}"))
+        })?;
+        let exec_params = Self::to_exec_params(params, session.as_ref(), turn.as_ref());
+        ShellHandler::run_exec_like(
+            tool_name.as_str(),
+            exec_params,
+            session,
+            turn,
+            tracker,
+            call_id,
+            false,
+        )
+        .await
+    }
+}
+
 impl ShellHandler {
     async fn run_exec_like(
         tool_name: &str,
@@ -240,3 +321,44 @@ impl ShellHandler {
         })
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use std::path::PathBuf;
+
+    use crate::is_safe_command::is_known_safe_command;
+    use crate::shell::BashShell;
+    use crate::shell::PowerShellConfig;
+    use crate::shell::Shell;
+    use crate::shell::ZshShell;
+
+    /// The logic for is_known_safe_command() has heuristics for known shells,
+    /// so we must ensure the commands generated by [ShellCommandHandler] can be
+    /// recognized as safe if the `command` is safe.
+    #[test]
+    fn commands_generated_by_shell_command_handler_can_be_matched_by_is_known_safe_command() {
+        let bash_shell = Shell::Bash(BashShell {
+            shell_path: PathBuf::from("/bin/bash"),
+        });
+        assert_safe(&bash_shell, "ls -la");
+
+        let zsh_shell = Shell::Zsh(ZshShell {
+            shell_path: PathBuf::from("/bin/zsh"),
+        });
+        assert_safe(&zsh_shell, "ls -la");
+
+        let powershell = Shell::PowerShell(PowerShellConfig {
+            shell_path: PathBuf::from("pwsh.exe"),
+        });
+        assert_safe(&powershell, "ls -Name");
+    }
+
+    fn assert_safe(shell: &Shell, command: &str) {
+        assert!(is_known_safe_command(
+            &shell.derive_exec_args(command, /* use_login_shell */ true)
+        ));
+        assert!(is_known_safe_command(
+            &shell.derive_exec_args(command, /* use_login_shell */ false)
+        ));
+    }
+}

codex-rs/core/src/tools/handlers/unified_exec.rs

@@ -1,13 +1,11 @@
-use std::time::Duration;
-
-use async_trait::async_trait;
-use serde::Deserialize;
-use serde::Serialize;
+use std::path::PathBuf;
 
 use crate::function_tool::FunctionCallError;
+use crate::is_safe_command::is_known_safe_command;
 use crate::protocol::EventMsg;
 use crate::protocol::ExecCommandOutputDeltaEvent;
 use crate::protocol::ExecOutputStream;
+use crate::shell::get_shell_by_model_provided_path;
 use crate::tools::context::ToolInvocation;
 use crate::tools::context::ToolOutput;
 use crate::tools::context::ToolPayload;
@@ -21,20 +19,28 @@ use crate::unified_exec::UnifiedExecContext;
 use crate::unified_exec::UnifiedExecResponse;
 use crate::unified_exec::UnifiedExecSessionManager;
 use crate::unified_exec::WriteStdinRequest;
+use async_trait::async_trait;
+use serde::Deserialize;
 
 pub struct UnifiedExecHandler;
 
 #[derive(Debug, Deserialize)]
 struct ExecCommandArgs {
     cmd: String,
+    #[serde(default)]
+    workdir: Option<String>,
     #[serde(default = "default_shell")]
     shell: String,
     #[serde(default = "default_login")]
     login: bool,
-    #[serde(default)]
-    yield_time_ms: Option<u64>,
+    #[serde(default = "default_exec_yield_time_ms")]
+    yield_time_ms: u64,
     #[serde(default)]
     max_output_tokens: Option<usize>,
+    #[serde(default)]
+    with_escalated_permissions: Option<bool>,
+    #[serde(default)]
+    justification: Option<String>,
 }
 
 #[derive(Debug, Deserialize)]
@@ -42,12 +48,20 @@ struct WriteStdinArgs {
     session_id: i32,
     #[serde(default)]
     chars: String,
-    #[serde(default)]
-    yield_time_ms: Option<u64>,
+    #[serde(default = "default_write_stdin_yield_time_ms")]
+    yield_time_ms: u64,
     #[serde(default)]
     max_output_tokens: Option<usize>,
 }
 
+fn default_exec_yield_time_ms() -> u64 {
+    10000
+}
+
+fn default_write_stdin_yield_time_ms() -> u64 {
+    250
+}
+
 fn default_shell() -> String {
     "/bin/bash".to_string()
 }
@@ -69,6 +83,20 @@ impl ToolHandler for UnifiedExecHandler {
         )
     }
 
+    fn is_mutating(&self, invocation: &ToolInvocation) -> bool {
+        let (ToolPayload::Function { arguments } | ToolPayload::UnifiedExec { arguments }) =
+            &invocation.payload
+        else {
+            return true;
+        };
+
+        let Ok(params) = serde_json::from_str::<ExecCommandArgs>(arguments) else {
+            return true;
+        };
+        let command = get_command(&params);
+        !is_known_safe_command(&command)
+    }
+
     async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError> {
         let ToolInvocation {
             session,
@@ -100,24 +128,53 @@ impl ToolHandler for UnifiedExecHandler {
                     ))
                 })?;
 
+                let command = get_command(&args);
+                let ExecCommandArgs {
+                    workdir,
+                    yield_time_ms,
+                    max_output_tokens,
+                    with_escalated_permissions,
+                    justification,
+                    ..
+                } = args;
+
+                if with_escalated_permissions.unwrap_or(false)
+                    && !matches!(
+                        context.turn.approval_policy,
+                        codex_protocol::protocol::AskForApproval::OnRequest
+                    )
+                {
+                    return Err(FunctionCallError::RespondToModel(format!(
+                        "approval policy is {policy:?}; reject command — you cannot ask for escalated permissions if the approval policy is {policy:?}",
+                        policy = context.turn.approval_policy
+                    )));
+                }
+
+                let workdir = workdir
+                    .as_deref()
+                    .filter(|value| !value.is_empty())
+                    .map(PathBuf::from);
+                let cwd = workdir.clone().unwrap_or_else(|| context.turn.cwd.clone());
+
                 let event_ctx = ToolEventCtx::new(
                     context.session.as_ref(),
                     context.turn.as_ref(),
                     &context.call_id,
                     None,
                 );
-                let emitter =
-                    ToolEmitter::unified_exec(args.cmd.clone(), context.turn.cwd.clone(), true);
+
+                let emitter = ToolEmitter::unified_exec(&command, cwd.clone(), true);
                 emitter.emit(event_ctx, ToolEventStage::Begin).await;
 
                 manager
                     .exec_command(
                         ExecCommandRequest {
-                            command: &args.cmd,
-                            shell: &args.shell,
-                            login: args.login,
-                            yield_time_ms: args.yield_time_ms,
-                            max_output_tokens: args.max_output_tokens,
+                            command,
+                            yield_time_ms,
+                            max_output_tokens,
+                            workdir,
+                            with_escalated_permissions,
+                            justification,
                         },
                         &context,
                     )
@@ -163,11 +220,7 @@ impl ToolHandler for UnifiedExecHandler {
                 .await;
         }
 
-        let content = serialize_response(&response).map_err(|err| {
-            FunctionCallError::RespondToModel(format!(
-                "failed to serialize unified exec output: {err:?}"
-            ))
-        })?;
+        let content = format_response(&response);
 
         Ok(ToolOutput::Function {
             content,
@@ -177,32 +230,35 @@ impl ToolHandler for UnifiedExecHandler {
     }
 }
 
-#[derive(Serialize)]
-struct SerializedUnifiedExecResponse<'a> {
-    chunk_id: &'a str,
-    wall_time_seconds: f64,
-    output: &'a str,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    session_id: Option<i32>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    exit_code: Option<i32>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    original_token_count: Option<usize>,
+fn get_command(args: &ExecCommandArgs) -> Vec<String> {
+    let shell = get_shell_by_model_provided_path(&PathBuf::from(args.shell.clone()));
+    shell.derive_exec_args(&args.cmd, args.login)
 }
 
-fn serialize_response(response: &UnifiedExecResponse) -> Result<String, serde_json::Error> {
-    let payload = SerializedUnifiedExecResponse {
-        chunk_id: &response.chunk_id,
-        wall_time_seconds: duration_to_seconds(response.wall_time),
-        output: &response.output,
-        session_id: response.session_id,
-        exit_code: response.exit_code,
-        original_token_count: response.original_token_count,
-    };
+fn format_response(response: &UnifiedExecResponse) -> String {
+    let mut sections = Vec::new();
 
-    serde_json::to_string(&payload)
-}
+    if !response.chunk_id.is_empty() {
+        sections.push(format!("Chunk ID: {}", response.chunk_id));
+    }
 
-fn duration_to_seconds(duration: Duration) -> f64 {
-    duration.as_secs_f64()
+    let wall_time_seconds = response.wall_time.as_secs_f64();
+    sections.push(format!("Wall time: {wall_time_seconds:.4} seconds"));
+
+    if let Some(exit_code) = response.exit_code {
+        sections.push(format!("Process exited with code {exit_code}"));
+    }
+
+    if let Some(session_id) = response.session_id {
+        sections.push(format!("Process running with session ID {session_id}"));
+    }
+
+    if let Some(original_token_count) = response.original_token_count {
+        sections.push(format!("Original token count: {original_token_count}"));
+    }
+
+    sections.push("Output:".to_string());
+    sections.push(response.output.clone());
+
+    sections.join("\n")
 }

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

@@ -9,6 +9,8 @@ pub mod runtimes;
 pub mod sandboxing;
 pub mod spec;
 
+use crate::context_manager::MODEL_FORMAT_MAX_BYTES;
+use crate::context_manager::MODEL_FORMAT_MAX_LINES;
 use crate::context_manager::format_output_for_model_body;
 use crate::exec::ExecToolCallOutput;
 pub use router::ToolRouter;
@@ -75,5 +77,5 @@ pub fn format_exec_output_str(exec_output: &ExecToolCallOutput) -> String {
     };
 
     // Truncate for model consumption before serialization.
-    format_output_for_model_body(&body)
+    format_output_for_model_body(&body, MODEL_FORMAT_MAX_BYTES, MODEL_FORMAT_MAX_LINES)
 }

codex-rs/core/src/tools/parallel.rs

@@ -16,7 +16,6 @@ use crate::tools::router::ToolCall;
 use crate::tools::router::ToolRouter;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::ResponseInputItem;
-use codex_utils_readiness::Readiness;
 
 pub(crate) struct ToolCallRuntime {
     router: Arc<ToolRouter>,
@@ -55,7 +54,6 @@ impl ToolCallRuntime {
         let tracker = Arc::clone(&self.tracker);
         let lock = Arc::clone(&self.parallel_execution);
         let started = Instant::now();
-        let readiness = self.turn_context.tool_call_gate.clone();
 
         let handle: AbortOnDropHandle<Result<ResponseInputItem, FunctionCallError>> =
             AbortOnDropHandle::new(tokio::spawn(async move {
@@ -65,9 +63,6 @@ impl ToolCallRuntime {
                         Ok(Self::aborted_response(&call, secs))
                     },
                     res = async {
-                        tracing::info!("waiting for tool gate");
-                        readiness.wait_ready().await;
-                        tracing::info!("tool gate released");
                         let _guard = if supports_parallel {
                             Either::Left(lock.read().await)
                         } else {

codex-rs/core/src/tools/registry.rs

@@ -2,15 +2,15 @@ use std::collections::HashMap;
 use std::sync::Arc;
 use std::time::Duration;
 
-use async_trait::async_trait;
-use codex_protocol::models::ResponseInputItem;
-use tracing::warn;
-
 use crate::client_common::tools::ToolSpec;
 use crate::function_tool::FunctionCallError;
 use crate::tools::context::ToolInvocation;
 use crate::tools::context::ToolOutput;
 use crate::tools::context::ToolPayload;
+use async_trait::async_trait;
+use codex_protocol::models::ResponseInputItem;
+use codex_utils_readiness::Readiness;
+use tracing::warn;
 
 #[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
 pub enum ToolKind {
@@ -30,6 +30,10 @@ pub trait ToolHandler: Send + Sync {
         )
     }
 
+    fn is_mutating(&self, _invocation: &ToolInvocation) -> bool {
+        false
+    }
+
     async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError>;
 }
 
@@ -106,6 +110,11 @@ impl ToolRegistry {
                     let output_cell = &output_cell;
                     let invocation = invocation;
                     async move {
+                        if handler.is_mutating(&invocation) {
+                            tracing::trace!("waiting for tool gate");
+                            invocation.turn.tool_call_gate.wait_ready().await;
+                            tracing::trace!("tool gate released");
+                        }
                         match handler.handle(invocation).await {
                             Ok(output) => {
                                 let preview = output.log_preview();

codex-rs/core/src/tools/runtimes/shell.rs

@@ -4,8 +4,7 @@ Runtime: shell
 Executes shell requests under the orchestrator: asks for approval when needed,
 builds a CommandSpec, and runs it under the current SandboxAttempt.
 */
-use crate::command_safety::is_dangerous_command::command_might_be_dangerous;
-use crate::command_safety::is_safe_command::is_known_safe_command;
+use crate::command_safety::is_dangerous_command::requires_initial_appoval;
 use crate::exec::ExecToolCallOutput;
 use crate::protocol::SandboxPolicy;
 use crate::sandboxing::execute_env;
@@ -121,28 +120,12 @@ impl Approvable<ShellRequest> for ShellRuntime {
         policy: AskForApproval,
         sandbox_policy: &SandboxPolicy,
     ) -> bool {
-        if is_known_safe_command(&req.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(&req.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 = req.with_escalated_permissions.unwrap_or(false);
-                if wants_escalation {
-                    return true;
-                }
-                command_might_be_dangerous(&req.command)
-            }
-            AskForApproval::UnlessTrusted => !is_known_safe_command(&req.command),
-        }
+        requires_initial_appoval(
+            policy,
+            sandbox_policy,
+            &req.command,
+            req.with_escalated_permissions.unwrap_or(false),
+        )
     }
 
     fn wants_escalated_first_attempt(&self, req: &ShellRequest) -> bool {

codex-rs/core/src/tools/runtimes/unified_exec.rs

@@ -1,3 +1,4 @@
+use crate::command_safety::is_dangerous_command::requires_initial_appoval;
 /*
 Runtime: unified exec
 
@@ -21,7 +22,9 @@ use crate::tools::sandboxing::with_cached_approval;
 use crate::unified_exec::UnifiedExecError;
 use crate::unified_exec::UnifiedExecSession;
 use crate::unified_exec::UnifiedExecSessionManager;
+use codex_protocol::protocol::AskForApproval;
 use codex_protocol::protocol::ReviewDecision;
+use codex_protocol::protocol::SandboxPolicy;
 use futures::future::BoxFuture;
 use std::collections::HashMap;
 use std::path::PathBuf;
@@ -31,6 +34,8 @@ pub struct UnifiedExecRequest {
     pub command: Vec<String>,
     pub cwd: PathBuf,
     pub env: HashMap<String, String>,
+    pub with_escalated_permissions: Option<bool>,
+    pub justification: Option<String>,
 }
 
 impl ProvidesSandboxRetryData for UnifiedExecRequest {
@@ -46,6 +51,7 @@ impl ProvidesSandboxRetryData for UnifiedExecRequest {
 pub struct UnifiedExecApprovalKey {
     pub command: Vec<String>,
     pub cwd: PathBuf,
+    pub escalated: bool,
 }
 
 pub struct UnifiedExecRuntime<'a> {
@@ -53,8 +59,20 @@ pub struct UnifiedExecRuntime<'a> {
 }
 
 impl UnifiedExecRequest {
-    pub fn new(command: Vec<String>, cwd: PathBuf, env: HashMap<String, String>) -> Self {
-        Self { command, cwd, env }
+    pub fn new(
+        command: Vec<String>,
+        cwd: PathBuf,
+        env: HashMap<String, String>,
+        with_escalated_permissions: Option<bool>,
+        justification: Option<String>,
+    ) -> Self {
+        Self {
+            command,
+            cwd,
+            env,
+            with_escalated_permissions,
+            justification,
+        }
     }
 }
 
@@ -81,6 +99,7 @@ impl Approvable<UnifiedExecRequest> for UnifiedExecRuntime<'_> {
         UnifiedExecApprovalKey {
             command: req.command.clone(),
             cwd: req.cwd.clone(),
+            escalated: req.with_escalated_permissions.unwrap_or(false),
         }
     }
 
@@ -95,7 +114,10 @@ impl Approvable<UnifiedExecRequest> for UnifiedExecRuntime<'_> {
         let call_id = ctx.call_id.to_string();
         let command = req.command.clone();
         let cwd = req.cwd.clone();
-        let reason = ctx.retry_reason.clone();
+        let reason = ctx
+            .retry_reason
+            .clone()
+            .or_else(|| req.justification.clone());
         let risk = ctx.risk.clone();
         Box::pin(async move {
             with_cached_approval(&session.services, key, || async move {
@@ -106,6 +128,24 @@ impl Approvable<UnifiedExecRequest> for UnifiedExecRuntime<'_> {
             .await
         })
     }
+
+    fn wants_initial_approval(
+        &self,
+        req: &UnifiedExecRequest,
+        policy: AskForApproval,
+        sandbox_policy: &SandboxPolicy,
+    ) -> bool {
+        requires_initial_appoval(
+            policy,
+            sandbox_policy,
+            &req.command,
+            req.with_escalated_permissions.unwrap_or(false),
+        )
+    }
+
+    fn wants_escalated_first_attempt(&self, req: &UnifiedExecRequest) -> bool {
+        req.with_escalated_permissions.unwrap_or(false)
+    }
 }
 
 impl<'a> ToolRuntime<UnifiedExecRequest, UnifiedExecSession> for UnifiedExecRuntime<'a> {
@@ -115,8 +155,15 @@ impl<'a> ToolRuntime<UnifiedExecRequest, UnifiedExecSession> for UnifiedExecRunt
         attempt: &SandboxAttempt<'_>,
         _ctx: &ToolCtx<'_>,
     ) -> Result<UnifiedExecSession, ToolError> {
-        let spec = build_command_spec(&req.command, &req.cwd, &req.env, None, None, None)
-            .map_err(|_| ToolError::Rejected("missing command line for PTY".to_string()))?;
+        let spec = build_command_spec(
+            &req.command,
+            &req.cwd,
+            &req.env,
+            None,
+            req.with_escalated_permissions,
+            req.justification.clone(),
+        )
+        .map_err(|_| ToolError::Rejected("missing command line for PTY".to_string()))?;
         let exec_env = attempt
             .env_for(&spec)
             .map_err(|err| ToolError::Codex(err.into()))?;

codex-rs/core/src/tools/spec.rs

@@ -15,11 +15,13 @@ use serde_json::json;
 use std::collections::BTreeMap;
 use std::collections::HashMap;
 
-#[derive(Debug, Clone)]
+#[derive(Debug, Clone, PartialEq, Eq, Hash)]
 pub enum ConfigShellToolType {
     Default,
     Local,
-    Streamable,
+    UnifiedExec,
+    /// Takes a command as a single string to be run in the user's default shell.
+    ShellCommand,
 }
 
 #[derive(Debug, Clone)]
@@ -28,7 +30,6 @@ pub(crate) struct ToolsConfig {
     pub apply_patch_tool_type: Option<ApplyPatchToolType>,
     pub web_search_request: bool,
     pub include_view_image_tool: bool,
-    pub experimental_unified_exec_tool: bool,
     pub experimental_supported_tools: Vec<String>,
 }
 
@@ -43,18 +44,16 @@ impl ToolsConfig {
             model_family,
             features,
         } = params;
-        let use_streamable_shell_tool = features.enabled(Feature::StreamableShell);
-        let experimental_unified_exec_tool = features.enabled(Feature::UnifiedExec);
         let include_apply_patch_tool = features.enabled(Feature::ApplyPatchFreeform);
         let include_web_search_request = features.enabled(Feature::WebSearchRequest);
         let include_view_image_tool = features.enabled(Feature::ViewImageTool);
 
-        let shell_type = if use_streamable_shell_tool {
-            ConfigShellToolType::Streamable
-        } else if model_family.uses_local_shell_tool {
-            ConfigShellToolType::Local
+        let shell_type = if features.enabled(Feature::UnifiedExec) {
+            ConfigShellToolType::UnifiedExec
+        } else if features.enabled(Feature::ShellCommandTool) {
+            ConfigShellToolType::ShellCommand
         } else {
-            ConfigShellToolType::Default
+            model_family.shell_type.clone()
         };
 
         let apply_patch_tool_type = match model_family.apply_patch_tool_type {
@@ -74,7 +73,6 @@ impl ToolsConfig {
             apply_patch_tool_type,
             web_search_request: include_web_search_request,
             include_view_image_tool,
-            experimental_unified_exec_tool,
             experimental_supported_tools: model_family.experimental_supported_tools.clone(),
         }
     }
@@ -144,6 +142,15 @@ fn create_exec_command_tool() -> ToolSpec {
             description: Some("Shell command to execute.".to_string()),
         },
     );
+    properties.insert(
+        "workdir".to_string(),
+        JsonSchema::String {
+            description: Some(
+                "Optional working directory to run the command in; defaults to the turn cwd."
+                    .to_string(),
+            ),
+        },
+    );
     properties.insert(
         "shell".to_string(),
         JsonSchema::String {
@@ -174,6 +181,24 @@ fn create_exec_command_tool() -> ToolSpec {
             ),
         },
     );
+    properties.insert(
+        "with_escalated_permissions".to_string(),
+        JsonSchema::Boolean {
+            description: Some(
+                "Whether to request escalated permissions. Set to true if command needs to be run without sandbox restrictions"
+                    .to_string(),
+            ),
+        },
+    );
+    properties.insert(
+        "justification".to_string(),
+        JsonSchema::String {
+            description: Some(
+                "Only set if with_escalated_permissions is true. 1-sentence explanation of why we want to run this command."
+                    .to_string(),
+            ),
+        },
+    );
 
     ToolSpec::Function(ResponsesApiTool {
         name: "exec_command".to_string(),
@@ -281,6 +306,53 @@ fn create_shell_tool() -> ToolSpec {
     })
 }
 
+fn create_shell_command_tool() -> ToolSpec {
+    let mut properties = BTreeMap::new();
+    properties.insert(
+        "command".to_string(),
+        JsonSchema::String {
+            description: Some(
+                "The shell script to execute in the user's default shell".to_string(),
+            ),
+        },
+    );
+    properties.insert(
+        "workdir".to_string(),
+        JsonSchema::String {
+            description: Some("The working directory to execute the command in".to_string()),
+        },
+    );
+    properties.insert(
+        "timeout_ms".to_string(),
+        JsonSchema::Number {
+            description: Some("The timeout for the command in milliseconds".to_string()),
+        },
+    );
+    properties.insert(
+        "with_escalated_permissions".to_string(),
+        JsonSchema::Boolean {
+            description: Some("Whether to request escalated permissions. Set to true if command needs to be run without sandbox restrictions".to_string()),
+        },
+    );
+    properties.insert(
+        "justification".to_string(),
+        JsonSchema::String {
+            description: Some("Only set if with_escalated_permissions is true. 1-sentence explanation of why we want to run this command.".to_string()),
+        },
+    );
+
+    ToolSpec::Function(ResponsesApiTool {
+        name: "shell_command".to_string(),
+        description: "Runs a shell command string and returns its output.".to_string(),
+        strict: false,
+        parameters: JsonSchema::Object {
+            properties,
+            required: Some(vec!["command".to_string()]),
+            additional_properties: Some(false.into()),
+        },
+    })
+}
+
 fn create_view_image_tool() -> ToolSpec {
     // Support only local filesystem path.
     let mut properties = BTreeMap::new();
@@ -870,6 +942,7 @@ pub(crate) fn build_specs(
     use crate::tools::handlers::McpResourceHandler;
     use crate::tools::handlers::PlanHandler;
     use crate::tools::handlers::ReadFileHandler;
+    use crate::tools::handlers::ShellCommandHandler;
     use crate::tools::handlers::ShellHandler;
     use crate::tools::handlers::TestSyncHandler;
     use crate::tools::handlers::UnifiedExecHandler;
@@ -885,16 +958,8 @@ pub(crate) fn build_specs(
     let view_image_handler = Arc::new(ViewImageHandler);
     let mcp_handler = Arc::new(McpHandler);
     let mcp_resource_handler = Arc::new(McpResourceHandler);
+    let shell_command_handler = Arc::new(ShellCommandHandler);
 
-    let use_unified_exec = config.experimental_unified_exec_tool
-        || matches!(config.shell_type, ConfigShellToolType::Streamable);
-
-    if use_unified_exec {
-        builder.push_spec(create_exec_command_tool());
-        builder.push_spec(create_write_stdin_tool());
-        builder.register_handler("exec_command", unified_exec_handler.clone());
-        builder.register_handler("write_stdin", unified_exec_handler);
-    }
     match &config.shell_type {
         ConfigShellToolType::Default => {
             builder.push_spec(create_shell_tool());
@@ -902,8 +967,14 @@ pub(crate) fn build_specs(
         ConfigShellToolType::Local => {
             builder.push_spec(ToolSpec::LocalShell {});
         }
-        ConfigShellToolType::Streamable => {
-            // Already handled by use_unified_exec.
+        ConfigShellToolType::UnifiedExec => {
+            builder.push_spec(create_exec_command_tool());
+            builder.push_spec(create_write_stdin_tool());
+            builder.register_handler("exec_command", unified_exec_handler.clone());
+            builder.register_handler("write_stdin", unified_exec_handler);
+        }
+        ConfigShellToolType::ShellCommand => {
+            builder.push_spec(create_shell_command_tool());
         }
     }
 
@@ -911,6 +982,7 @@ pub(crate) fn build_specs(
     builder.register_handler("shell", shell_handler.clone());
     builder.register_handler("container.exec", shell_handler.clone());
     builder.register_handler("local_shell", shell_handler);
+    builder.register_handler("shell_command", shell_command_handler);
 
     builder.push_spec_with_parallel_support(create_list_mcp_resources_tool(), true);
     builder.push_spec_with_parallel_support(create_list_mcp_resource_templates_tool(), true);
@@ -1045,7 +1117,8 @@ mod tests {
         match config.shell_type {
             ConfigShellToolType::Default => Some("shell"),
             ConfigShellToolType::Local => Some("local_shell"),
-            ConfigShellToolType::Streamable => None,
+            ConfigShellToolType::UnifiedExec => None,
+            ConfigShellToolType::ShellCommand => Some("shell_command"),
         }
     }
 
@@ -1095,7 +1168,7 @@ mod tests {
     }
 
     #[test]
-    fn test_full_toolset_specs_for_gpt5_codex() {
+    fn test_full_toolset_specs_for_gpt5_codex_unified_exec_web_search() {
         let model_family = find_family_for_model("gpt-5-codex")
             .expect("gpt-5-codex should be a valid model family");
         let mut features = Features::with_defaults();
@@ -1129,7 +1202,6 @@ mod tests {
         for spec in [
             create_exec_command_tool(),
             create_write_stdin_tool(),
-            create_shell_tool(),
             create_list_mcp_resources_tool(),
             create_list_mcp_resource_templates_tool(),
             create_read_mcp_resource_tool(),
@@ -1156,32 +1228,161 @@ mod tests {
         }
     }
 
-    #[test]
-    fn test_build_specs_contains_expected_basics() {
-        let model_family = find_family_for_model("codex-mini-latest")
-            .expect("codex-mini-latest should be a valid model family");
-        let mut features = Features::with_defaults();
-        features.enable(Feature::WebSearchRequest);
-        features.enable(Feature::UnifiedExec);
+    fn assert_model_tools(model_family: &str, features: &Features, expected_tools: &[&str]) {
+        let model_family = find_family_for_model(model_family)
+            .unwrap_or_else(|| panic!("{model_family} should be a valid model family"));
         let config = ToolsConfig::new(&ToolsConfigParams {
             model_family: &model_family,
-            features: &features,
+            features,
         });
         let (tools, _) = build_specs(&config, Some(HashMap::new())).build();
         let tool_names = tools.iter().map(|t| t.spec.name()).collect::<Vec<_>>();
-        assert_eq!(
-            &tool_names,
+        assert_eq!(&tool_names, &expected_tools,);
+    }
+
+    #[test]
+    fn test_build_specs_gpt5_codex_default() {
+        assert_model_tools(
+            "gpt-5-codex",
+            &Features::with_defaults(),
+            &[
+                "shell",
+                "list_mcp_resources",
+                "list_mcp_resource_templates",
+                "read_mcp_resource",
+                "update_plan",
+                "apply_patch",
+                "view_image",
+            ],
+        );
+    }
+
+    #[test]
+    fn test_build_specs_gpt51_codex_default() {
+        assert_model_tools(
+            "gpt-5.1-codex",
+            &Features::with_defaults(),
+            &[
+                "shell",
+                "list_mcp_resources",
+                "list_mcp_resource_templates",
+                "read_mcp_resource",
+                "update_plan",
+                "apply_patch",
+                "view_image",
+            ],
+        );
+    }
+
+    #[test]
+    fn test_build_specs_gpt5_codex_unified_exec_web_search() {
+        assert_model_tools(
+            "gpt-5-codex",
+            Features::with_defaults()
+                .enable(Feature::UnifiedExec)
+                .enable(Feature::WebSearchRequest),
             &[
                 "exec_command",
                 "write_stdin",
+                "list_mcp_resources",
+                "list_mcp_resource_templates",
+                "read_mcp_resource",
+                "update_plan",
+                "apply_patch",
+                "web_search",
+                "view_image",
+            ],
+        );
+    }
+
+    #[test]
+    fn test_build_specs_gpt51_codex_unified_exec_web_search() {
+        assert_model_tools(
+            "gpt-5.1-codex",
+            Features::with_defaults()
+                .enable(Feature::UnifiedExec)
+                .enable(Feature::WebSearchRequest),
+            &[
+                "exec_command",
+                "write_stdin",
+                "list_mcp_resources",
+                "list_mcp_resource_templates",
+                "read_mcp_resource",
+                "update_plan",
+                "apply_patch",
+                "web_search",
+                "view_image",
+            ],
+        );
+    }
+
+    #[test]
+    fn test_codex_mini_defaults() {
+        assert_model_tools(
+            "codex-mini-latest",
+            &Features::with_defaults(),
+            &[
                 "local_shell",
                 "list_mcp_resources",
                 "list_mcp_resource_templates",
                 "read_mcp_resource",
                 "update_plan",
+                "view_image",
+            ],
+        );
+    }
+
+    #[test]
+    fn test_codex_5_1_mini_defaults() {
+        assert_model_tools(
+            "gpt-5.1-codex-mini",
+            &Features::with_defaults(),
+            &[
+                "shell",
+                "list_mcp_resources",
+                "list_mcp_resource_templates",
+                "read_mcp_resource",
+                "update_plan",
+                "apply_patch",
+                "view_image",
+            ],
+        );
+    }
+
+    #[test]
+    fn test_gpt_5_1_defaults() {
+        assert_model_tools(
+            "gpt-5.1",
+            &Features::with_defaults(),
+            &[
+                "shell",
+                "list_mcp_resources",
+                "list_mcp_resource_templates",
+                "read_mcp_resource",
+                "update_plan",
+                "apply_patch",
+                "view_image",
+            ],
+        );
+    }
+
+    #[test]
+    fn test_codex_mini_unified_exec_web_search() {
+        assert_model_tools(
+            "codex-mini-latest",
+            Features::with_defaults()
+                .enable(Feature::UnifiedExec)
+                .enable(Feature::WebSearchRequest),
+            &[
+                "exec_command",
+                "write_stdin",
+                "list_mcp_resources",
+                "list_mcp_resource_templates",
+                "read_mcp_resource",
+                "update_plan",
                 "web_search",
                 "view_image",
-            ]
+            ],
         );
     }
 
@@ -1205,6 +1406,22 @@ mod tests {
         assert_contains_tool_names(&tools, &subset);
     }
 
+    #[test]
+    fn test_build_specs_shell_command_present() {
+        assert_model_tools(
+            "codex-mini-latest",
+            Features::with_defaults().enable(Feature::ShellCommandTool),
+            &[
+                "shell_command",
+                "list_mcp_resources",
+                "list_mcp_resource_templates",
+                "read_mcp_resource",
+                "update_plan",
+                "view_image",
+            ],
+        );
+    }
+
     #[test]
     #[ignore]
     fn test_parallel_support_flags() {
@@ -1660,6 +1877,21 @@ mod tests {
         assert_eq!(description, expected);
     }
 
+    #[test]
+    fn test_shell_command_tool() {
+        let tool = super::create_shell_command_tool();
+        let ToolSpec::Function(ResponsesApiTool {
+            description, name, ..
+        }) = &tool
+        else {
+            panic!("expected function tool");
+        };
+        assert_eq!(name, "shell_command");
+
+        let expected = "Runs a shell command string and returns its output.";
+        assert_eq!(description, expected);
+    }
+
     #[test]
     fn test_get_openai_tools_mcp_tools_with_additional_properties_schema() {
         let model_family = find_family_for_model("gpt-5-codex")

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

@@ -41,7 +41,6 @@ mod session_manager;
 pub(crate) use errors::UnifiedExecError;
 pub(crate) use session::UnifiedExecSession;
 
-pub(crate) const DEFAULT_YIELD_TIME_MS: u64 = 10_000;
 pub(crate) const MIN_YIELD_TIME_MS: u64 = 250;
 pub(crate) const MAX_YIELD_TIME_MS: u64 = 30_000;
 pub(crate) const DEFAULT_MAX_OUTPUT_TOKENS: usize = 10_000;
@@ -64,19 +63,20 @@ impl UnifiedExecContext {
 }
 
 #[derive(Debug)]
-pub(crate) struct ExecCommandRequest<'a> {
-    pub command: &'a str,
-    pub shell: &'a str,
-    pub login: bool,
-    pub yield_time_ms: Option<u64>,
+pub(crate) struct ExecCommandRequest {
+    pub command: Vec<String>,
+    pub yield_time_ms: u64,
     pub max_output_tokens: Option<usize>,
+    pub workdir: Option<PathBuf>,
+    pub with_escalated_permissions: Option<bool>,
+    pub justification: Option<String>,
 }
 
 #[derive(Debug)]
 pub(crate) struct WriteStdinRequest<'a> {
     pub session_id: i32,
     pub input: &'a str,
-    pub yield_time_ms: Option<u64>,
+    pub yield_time_ms: u64,
     pub max_output_tokens: Option<usize>,
 }
 
@@ -102,16 +102,13 @@ struct SessionEntry {
     session_ref: Arc<Session>,
     turn_ref: Arc<TurnContext>,
     call_id: String,
-    command: String,
+    command: Vec<String>,
     cwd: PathBuf,
     started_at: tokio::time::Instant,
 }
 
-pub(crate) fn clamp_yield_time(yield_time_ms: Option<u64>) -> u64 {
-    match yield_time_ms {
-        Some(value) => value.clamp(MIN_YIELD_TIME_MS, MAX_YIELD_TIME_MS),
-        None => DEFAULT_YIELD_TIME_MS,
-    }
+pub(crate) fn clamp_yield_time(yield_time_ms: u64) -> u64 {
+    yield_time_ms.clamp(MIN_YIELD_TIME_MS, MAX_YIELD_TIME_MS)
 }
 
 pub(crate) fn resolve_max_tokens(max_tokens: Option<usize>) -> usize {
@@ -184,7 +181,7 @@ mod tests {
         session: &Arc<Session>,
         turn: &Arc<TurnContext>,
         cmd: &str,
-        yield_time_ms: Option<u64>,
+        yield_time_ms: u64,
     ) -> Result<UnifiedExecResponse, UnifiedExecError> {
         let context =
             UnifiedExecContext::new(Arc::clone(session), Arc::clone(turn), "call".to_string());
@@ -194,11 +191,12 @@ mod tests {
             .unified_exec_manager
             .exec_command(
                 ExecCommandRequest {
-                    command: cmd,
-                    shell: "/bin/bash",
-                    login: true,
+                    command: vec!["bash".to_string(), "-lc".to_string(), cmd.to_string()],
                     yield_time_ms,
                     max_output_tokens: None,
+                    workdir: None,
+                    with_escalated_permissions: None,
+                    justification: None,
                 },
                 &context,
             )
@@ -209,7 +207,7 @@ mod tests {
         session: &Arc<Session>,
         session_id: i32,
         input: &str,
-        yield_time_ms: Option<u64>,
+        yield_time_ms: u64,
     ) -> Result<UnifiedExecResponse, UnifiedExecError> {
         session
             .services
@@ -247,14 +245,14 @@ mod tests {
 
         let (session, turn) = test_session_and_turn();
 
-        let open_shell = exec_command(&session, &turn, "bash -i", Some(2_500)).await?;
+        let open_shell = exec_command(&session, &turn, "bash -i", 2_500).await?;
         let session_id = open_shell.session_id.expect("expected session_id");
 
         write_stdin(
             &session,
             session_id,
             "export CODEX_INTERACTIVE_SHELL_VAR=codex\n",
-            Some(2_500),
+            2_500,
         )
         .await?;
 
@@ -262,7 +260,7 @@ mod tests {
             &session,
             session_id,
             "echo $CODEX_INTERACTIVE_SHELL_VAR\n",
-            Some(2_500),
+            2_500,
         )
         .await?;
         assert!(
@@ -279,24 +277,19 @@ mod tests {
 
         let (session, turn) = test_session_and_turn();
 
-        let shell_a = exec_command(&session, &turn, "bash -i", Some(2_500)).await?;
+        let shell_a = exec_command(&session, &turn, "bash -i", 2_500).await?;
         let session_a = shell_a.session_id.expect("expected session id");
 
         write_stdin(
             &session,
             session_a,
             "export CODEX_INTERACTIVE_SHELL_VAR=codex\n",
-            Some(2_500),
+            2_500,
         )
         .await?;
 
-        let out_2 = exec_command(
-            &session,
-            &turn,
-            "echo $CODEX_INTERACTIVE_SHELL_VAR",
-            Some(2_500),
-        )
-        .await?;
+        let out_2 =
+            exec_command(&session, &turn, "echo $CODEX_INTERACTIVE_SHELL_VAR", 2_500).await?;
         assert!(
             out_2.session_id.is_none(),
             "short command should not retain a session"
@@ -310,7 +303,7 @@ mod tests {
             &session,
             session_a,
             "echo $CODEX_INTERACTIVE_SHELL_VAR\n",
-            Some(2_500),
+            2_500,
         )
         .await?;
         assert!(
@@ -327,14 +320,14 @@ mod tests {
 
         let (session, turn) = test_session_and_turn();
 
-        let open_shell = exec_command(&session, &turn, "bash -i", Some(2_500)).await?;
+        let open_shell = exec_command(&session, &turn, "bash -i", 2_500).await?;
         let session_id = open_shell.session_id.expect("expected session id");
 
         write_stdin(
             &session,
             session_id,
             "export CODEX_INTERACTIVE_SHELL_VAR=codex\n",
-            Some(2_500),
+            2_500,
         )
         .await?;
 
@@ -342,7 +335,7 @@ mod tests {
             &session,
             session_id,
             "sleep 5 && echo $CODEX_INTERACTIVE_SHELL_VAR\n",
-            Some(10),
+            10,
         )
         .await?;
         assert!(
@@ -352,7 +345,7 @@ mod tests {
 
         tokio::time::sleep(Duration::from_secs(7)).await;
 
-        let out_3 = write_stdin(&session, session_id, "", Some(100)).await?;
+        let out_3 = write_stdin(&session, session_id, "", 100).await?;
 
         assert!(
             out_3.output.contains("codex"),
@@ -367,7 +360,7 @@ mod tests {
     async fn requests_with_large_timeout_are_capped() -> anyhow::Result<()> {
         let (session, turn) = test_session_and_turn();
 
-        let result = exec_command(&session, &turn, "echo codex", Some(120_000)).await?;
+        let result = exec_command(&session, &turn, "echo codex", 120_000).await?;
 
         assert!(result.session_id.is_none());
         assert!(result.output.contains("codex"));
@@ -379,7 +372,7 @@ mod tests {
     #[ignore] // Ignored while we have a better way to test this.
     async fn completed_commands_do_not_persist_sessions() -> anyhow::Result<()> {
         let (session, turn) = test_session_and_turn();
-        let result = exec_command(&session, &turn, "echo codex", Some(2_500)).await?;
+        let result = exec_command(&session, &turn, "echo codex", 2_500).await?;
 
         assert!(
             result.session_id.is_none(),
@@ -406,14 +399,14 @@ mod tests {
 
         let (session, turn) = test_session_and_turn();
 
-        let open_shell = exec_command(&session, &turn, "bash -i", Some(2_500)).await?;
+        let open_shell = exec_command(&session, &turn, "bash -i", 2_500).await?;
         let session_id = open_shell.session_id.expect("expected session id");
 
-        write_stdin(&session, session_id, "exit\n", Some(2_500)).await?;
+        write_stdin(&session, session_id, "exit\n", 2_500).await?;
 
         tokio::time::sleep(Duration::from_millis(200)).await;
 
-        let err = write_stdin(&session, session_id, "", Some(100))
+        let err = write_stdin(&session, session_id, "", 100)
             .await
             .expect_err("expected unknown session error");
 

codex-rs/core/src/unified_exec/session_manager.rs

@@ -1,3 +1,4 @@
+use std::path::PathBuf;
 use std::sync::Arc;
 
 use tokio::sync::Notify;
@@ -18,7 +19,6 @@ use crate::tools::runtimes::unified_exec::UnifiedExecRuntime;
 use crate::tools::sandboxing::ToolCtx;
 
 use super::ExecCommandRequest;
-use super::MIN_YIELD_TIME_MS;
 use super::SessionEntry;
 use super::UnifiedExecContext;
 use super::UnifiedExecError;
@@ -35,21 +35,26 @@ use super::truncate_output_to_tokens;
 impl UnifiedExecSessionManager {
     pub(crate) async fn exec_command(
         &self,
-        request: ExecCommandRequest<'_>,
+        request: ExecCommandRequest,
         context: &UnifiedExecContext,
     ) -> Result<UnifiedExecResponse, UnifiedExecError> {
-        let shell_flag = if request.login { "-lc" } else { "-c" };
-        let command = vec![
-            request.shell.to_string(),
-            shell_flag.to_string(),
-            request.command.to_string(),
-        ];
+        let cwd = request
+            .workdir
+            .clone()
+            .unwrap_or_else(|| context.turn.cwd.clone());
 
-        let session = self.open_session_with_sandbox(command, context).await?;
+        let session = self
+            .open_session_with_sandbox(
+                &request.command,
+                cwd.clone(),
+                request.with_escalated_permissions,
+                request.justification,
+                context,
+            )
+            .await?;
 
         let max_tokens = resolve_max_tokens(request.max_output_tokens);
-        let yield_time_ms =
-            clamp_yield_time(Some(request.yield_time_ms.unwrap_or(MIN_YIELD_TIME_MS)));
+        let yield_time_ms = clamp_yield_time(request.yield_time_ms);
 
         let start = Instant::now();
         let (output_buffer, output_notify) = session.output_handles();
@@ -66,7 +71,7 @@ impl UnifiedExecSessionManager {
             None
         } else {
             Some(
-                self.store_session(session, context, request.command, start)
+                self.store_session(session, context, &request.command, cwd.clone(), start)
                     .await,
             )
         };
@@ -86,7 +91,8 @@ impl UnifiedExecSessionManager {
             let exit = response.exit_code.unwrap_or(-1);
             Self::emit_exec_end_from_context(
                 context,
-                request.command.to_string(),
+                &request.command,
+                cwd,
                 response.output.clone(),
                 exit,
                 response.wall_time,
@@ -210,7 +216,8 @@ impl UnifiedExecSessionManager {
         &self,
         session: UnifiedExecSession,
         context: &UnifiedExecContext,
-        command: &str,
+        command: &[String],
+        cwd: PathBuf,
         started_at: Instant,
     ) -> i32 {
         let session_id = self
@@ -221,8 +228,8 @@ impl UnifiedExecSessionManager {
             session_ref: Arc::clone(&context.session),
             turn_ref: Arc::clone(&context.turn),
             call_id: context.call_id.clone(),
-            command: command.to_string(),
-            cwd: context.turn.cwd.clone(),
+            command: command.to_vec(),
+            cwd,
             started_at,
         };
         self.sessions.lock().await.insert(session_id, entry);
@@ -249,7 +256,7 @@ impl UnifiedExecSessionManager {
             &entry.call_id,
             None,
         );
-        let emitter = ToolEmitter::unified_exec(entry.command, entry.cwd, true);
+        let emitter = ToolEmitter::unified_exec(&entry.command, entry.cwd, true);
         emitter
             .emit(event_ctx, ToolEventStage::Success(output))
             .await;
@@ -257,7 +264,8 @@ impl UnifiedExecSessionManager {
 
     async fn emit_exec_end_from_context(
         context: &UnifiedExecContext,
-        command: String,
+        command: &[String],
+        cwd: PathBuf,
         aggregated_output: String,
         exit_code: i32,
         duration: Duration,
@@ -276,7 +284,7 @@ impl UnifiedExecSessionManager {
             &context.call_id,
             None,
         );
-        let emitter = ToolEmitter::unified_exec(command, context.turn.cwd.clone(), true);
+        let emitter = ToolEmitter::unified_exec(command, cwd, true);
         emitter
             .emit(event_ctx, ToolEventStage::Success(output))
             .await;
@@ -290,24 +298,35 @@ impl UnifiedExecSessionManager {
             .command
             .split_first()
             .ok_or(UnifiedExecError::MissingCommandLine)?;
-        let spawned =
-            codex_utils_pty::spawn_pty_process(program, args, env.cwd.as_path(), &env.env)
-                .await
-                .map_err(|err| UnifiedExecError::create_session(err.to_string()))?;
+
+        let spawned = codex_utils_pty::spawn_pty_process(
+            program,
+            args,
+            env.cwd.as_path(),
+            &env.env,
+            &env.arg0,
+        )
+        .await
+        .map_err(|err| UnifiedExecError::create_session(err.to_string()))?;
         UnifiedExecSession::from_spawned(spawned, env.sandbox).await
     }
 
     pub(super) async fn open_session_with_sandbox(
         &self,
-        command: Vec<String>,
+        command: &[String],
+        cwd: PathBuf,
+        with_escalated_permissions: Option<bool>,
+        justification: Option<String>,
         context: &UnifiedExecContext,
     ) -> Result<UnifiedExecSession, UnifiedExecError> {
         let mut orchestrator = ToolOrchestrator::new();
         let mut runtime = UnifiedExecRuntime::new(self);
         let req = UnifiedExecToolRequest::new(
-            command,
-            context.turn.cwd.clone(),
+            command.to_vec(),
+            cwd,
             create_env(&context.turn.shell_environment_policy),
+            with_escalated_permissions,
+            justification,
         );
         let tool_ctx = ToolCtx {
             session: context.session.as_ref(),

codex-rs/core/src/user_shell_command.rs

@@ -0,0 +1,108 @@
+use std::time::Duration;
+
+use codex_protocol::models::ContentItem;
+use codex_protocol::models::ResponseItem;
+
+use crate::exec::ExecToolCallOutput;
+use crate::tools::format_exec_output_str;
+
+pub const USER_SHELL_COMMAND_OPEN: &str = "<user_shell_command>";
+pub const USER_SHELL_COMMAND_CLOSE: &str = "</user_shell_command>";
+
+pub fn is_user_shell_command_text(text: &str) -> bool {
+    let trimmed = text.trim_start();
+    let lowered = trimmed.to_ascii_lowercase();
+    lowered.starts_with(USER_SHELL_COMMAND_OPEN)
+}
+
+fn format_duration_line(duration: Duration) -> String {
+    let duration_seconds = duration.as_secs_f64();
+    format!("Duration: {duration_seconds:.4} seconds")
+}
+
+fn format_user_shell_command_body(command: &str, exec_output: &ExecToolCallOutput) -> String {
+    let mut sections = Vec::new();
+    sections.push("<command>".to_string());
+    sections.push(command.to_string());
+    sections.push("</command>".to_string());
+    sections.push("<result>".to_string());
+    sections.push(format!("Exit code: {}", exec_output.exit_code));
+    sections.push(format_duration_line(exec_output.duration));
+    sections.push("Output:".to_string());
+    sections.push(format_exec_output_str(exec_output));
+    sections.push("</result>".to_string());
+    sections.join("\n")
+}
+
+pub fn format_user_shell_command_record(command: &str, exec_output: &ExecToolCallOutput) -> String {
+    let body = format_user_shell_command_body(command, exec_output);
+    format!("{USER_SHELL_COMMAND_OPEN}\n{body}\n{USER_SHELL_COMMAND_CLOSE}")
+}
+
+pub fn user_shell_command_record_item(
+    command: &str,
+    exec_output: &ExecToolCallOutput,
+) -> ResponseItem {
+    ResponseItem::Message {
+        id: None,
+        role: "user".to_string(),
+        content: vec![ContentItem::InputText {
+            text: format_user_shell_command_record(command, exec_output),
+        }],
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::exec::StreamOutput;
+    use pretty_assertions::assert_eq;
+
+    #[test]
+    fn detects_user_shell_command_text_variants() {
+        assert!(is_user_shell_command_text(
+            "<user_shell_command>\necho hi\n</user_shell_command>"
+        ));
+        assert!(!is_user_shell_command_text("echo hi"));
+    }
+
+    #[test]
+    fn formats_basic_record() {
+        let exec_output = ExecToolCallOutput {
+            exit_code: 0,
+            stdout: StreamOutput::new("hi".to_string()),
+            stderr: StreamOutput::new(String::new()),
+            aggregated_output: StreamOutput::new("hi".to_string()),
+            duration: Duration::from_secs(1),
+            timed_out: false,
+        };
+        let item = user_shell_command_record_item("echo hi", &exec_output);
+        let ResponseItem::Message { content, .. } = item else {
+            panic!("expected message");
+        };
+        let [ContentItem::InputText { text }] = content.as_slice() else {
+            panic!("expected input text");
+        };
+        assert_eq!(
+            text,
+            "<user_shell_command>\n<command>\necho hi\n</command>\n<result>\nExit code: 0\nDuration: 1.0000 seconds\nOutput:\nhi\n</result>\n</user_shell_command>"
+        );
+    }
+
+    #[test]
+    fn uses_aggregated_output_over_streams() {
+        let exec_output = ExecToolCallOutput {
+            exit_code: 42,
+            stdout: StreamOutput::new("stdout-only".to_string()),
+            stderr: StreamOutput::new("stderr-only".to_string()),
+            aggregated_output: StreamOutput::new("combined output wins".to_string()),
+            duration: Duration::from_millis(120),
+            timed_out: false,
+        };
+        let record = format_user_shell_command_record("false", &exec_output);
+        assert_eq!(
+            record,
+            "<user_shell_command>\n<command>\nfalse\n</command>\n<result>\nExit code: 42\nDuration: 0.1200 seconds\nOutput:\ncombined output wins\n</result>\n</user_shell_command>"
+        );
+    }
+}

codex-rs/core/tests/chat_completions_payload.rs

@@ -10,11 +10,11 @@ use codex_core::ModelProviderInfo;
 use codex_core::Prompt;
 use codex_core::ResponseItem;
 use codex_core::WireApi;
-use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
 use codex_otel::otel_event_manager::OtelEventManager;
 use codex_protocol::ConversationId;
 use codex_protocol::models::ReasoningItemContent;
 use core_test_support::load_default_config_for_test;
+use core_test_support::skip_if_no_network;
 use futures::StreamExt;
 use serde_json::Value;
 use tempfile::TempDir;
@@ -24,10 +24,6 @@ use wiremock::ResponseTemplate;
 use wiremock::matchers::method;
 use wiremock::matchers::path;
 
-fn network_disabled() -> bool {
-    std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok()
-}
-
 async fn run_request(input: Vec<ResponseItem>) -> Value {
     let server = MockServer::start().await;
 
@@ -191,12 +187,7 @@ fn first_assistant(messages: &[Value]) -> &Value {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn omits_reasoning_when_none_present() {
-    if network_disabled() {
-        println!(
-            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
-        );
-        return;
-    }
+    skip_if_no_network!();
 
     let body = run_request(vec![user_message("u1"), assistant_message("a1")]).await;
     let messages = messages_from(&body);
@@ -208,12 +199,7 @@ async fn omits_reasoning_when_none_present() {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn attaches_reasoning_to_previous_assistant() {
-    if network_disabled() {
-        println!(
-            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
-        );
-        return;
-    }
+    skip_if_no_network!();
 
     let body = run_request(vec![
         user_message("u1"),
@@ -230,12 +216,7 @@ async fn attaches_reasoning_to_previous_assistant() {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn attaches_reasoning_to_function_call_anchor() {
-    if network_disabled() {
-        println!(
-            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
-        );
-        return;
-    }
+    skip_if_no_network!();
 
     let body = run_request(vec![
         user_message("u1"),
@@ -257,12 +238,7 @@ async fn attaches_reasoning_to_function_call_anchor() {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn attaches_reasoning_to_local_shell_call() {
-    if network_disabled() {
-        println!(
-            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
-        );
-        return;
-    }
+    skip_if_no_network!();
 
     let body = run_request(vec![
         user_message("u1"),
@@ -282,12 +258,7 @@ async fn attaches_reasoning_to_local_shell_call() {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn drops_reasoning_when_last_role_is_user() {
-    if network_disabled() {
-        println!(
-            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
-        );
-        return;
-    }
+    skip_if_no_network!();
 
     let body = run_request(vec![
         assistant_message("aPrev"),
@@ -301,12 +272,7 @@ async fn drops_reasoning_when_last_role_is_user() {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn ignores_reasoning_before_last_user() {
-    if network_disabled() {
-        println!(
-            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
-        );
-        return;
-    }
+    skip_if_no_network!();
 
     let body = run_request(vec![
         user_message("u1"),
@@ -321,12 +287,7 @@ async fn ignores_reasoning_before_last_user() {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn skips_empty_reasoning_segments() {
-    if network_disabled() {
-        println!(
-            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
-        );
-        return;
-    }
+    skip_if_no_network!();
 
     let body = run_request(vec![
         user_message("u1"),
@@ -342,12 +303,7 @@ async fn skips_empty_reasoning_segments() {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn suppresses_duplicate_assistant_messages() {
-    if network_disabled() {
-        println!(
-            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
-        );
-        return;
-    }
+    skip_if_no_network!();
 
     let body = run_request(vec![assistant_message("dup"), assistant_message("dup")]).await;
     let messages = messages_from(&body);

codex-rs/core/tests/chat_completions_sse.rs

@@ -10,11 +10,11 @@ use codex_core::Prompt;
 use codex_core::ResponseEvent;
 use codex_core::ResponseItem;
 use codex_core::WireApi;
-use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
 use codex_otel::otel_event_manager::OtelEventManager;
 use codex_protocol::ConversationId;
 use codex_protocol::models::ReasoningItemContent;
 use core_test_support::load_default_config_for_test;
+use core_test_support::skip_if_no_network;
 use futures::StreamExt;
 use tempfile::TempDir;
 use wiremock::Mock;
@@ -23,10 +23,6 @@ use wiremock::ResponseTemplate;
 use wiremock::matchers::method;
 use wiremock::matchers::path;
 
-fn network_disabled() -> bool {
-    std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok()
-}
-
 async fn run_stream(sse_body: &str) -> Vec<ResponseEvent> {
     run_stream_with_bytes(sse_body.as_bytes()).await
 }
@@ -157,12 +153,7 @@ fn assert_reasoning(item: &ResponseItem, expected: &str) {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn streams_text_without_reasoning() {
-    if network_disabled() {
-        println!(
-            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
-        );
-        return;
-    }
+    skip_if_no_network!();
 
     let sse = concat!(
         "data: {\"choices\":[{\"delta\":{\"content\":\"hi\"}}]}\n\n",
@@ -193,12 +184,7 @@ async fn streams_text_without_reasoning() {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn streams_reasoning_from_string_delta() {
-    if network_disabled() {
-        println!(
-            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
-        );
-        return;
-    }
+    skip_if_no_network!();
 
     let sse = concat!(
         "data: {\"choices\":[{\"delta\":{\"reasoning\":\"think1\"}}]}\n\n",
@@ -215,7 +201,13 @@ async fn streams_reasoning_from_string_delta() {
     }
 
     match &events[1] {
-        ResponseEvent::ReasoningContentDelta(text) => assert_eq!(text, "think1"),
+        ResponseEvent::ReasoningContentDelta {
+            delta,
+            content_index,
+        } => {
+            assert_eq!(delta, "think1");
+            assert_eq!(content_index, &0);
+        }
         other => panic!("expected reasoning delta, got {other:?}"),
     }
 
@@ -244,12 +236,7 @@ async fn streams_reasoning_from_string_delta() {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn streams_reasoning_from_object_delta() {
-    if network_disabled() {
-        println!(
-            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
-        );
-        return;
-    }
+    skip_if_no_network!();
 
     let sse = concat!(
         "data: {\"choices\":[{\"delta\":{\"reasoning\":{\"text\":\"partA\"}}}]}\n\n",
@@ -267,12 +254,24 @@ async fn streams_reasoning_from_object_delta() {
     }
 
     match &events[1] {
-        ResponseEvent::ReasoningContentDelta(text) => assert_eq!(text, "partA"),
+        ResponseEvent::ReasoningContentDelta {
+            delta,
+            content_index,
+        } => {
+            assert_eq!(delta, "partA");
+            assert_eq!(content_index, &0);
+        }
         other => panic!("expected reasoning delta, got {other:?}"),
     }
 
     match &events[2] {
-        ResponseEvent::ReasoningContentDelta(text) => assert_eq!(text, "partB"),
+        ResponseEvent::ReasoningContentDelta {
+            delta,
+            content_index,
+        } => {
+            assert_eq!(delta, "partB");
+            assert_eq!(content_index, &1);
+        }
         other => panic!("expected reasoning delta, got {other:?}"),
     }
 
@@ -301,12 +300,7 @@ async fn streams_reasoning_from_object_delta() {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn streams_reasoning_from_final_message() {
-    if network_disabled() {
-        println!(
-            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
-        );
-        return;
-    }
+    skip_if_no_network!();
 
     let sse = "data: {\"choices\":[{\"message\":{\"reasoning\":\"final-cot\"},\"finish_reason\":\"stop\"}]}\n\n";
 
@@ -319,7 +313,13 @@ async fn streams_reasoning_from_final_message() {
     }
 
     match &events[1] {
-        ResponseEvent::ReasoningContentDelta(text) => assert_eq!(text, "final-cot"),
+        ResponseEvent::ReasoningContentDelta {
+            delta,
+            content_index,
+        } => {
+            assert_eq!(delta, "final-cot");
+            assert_eq!(content_index, &0);
+        }
         other => panic!("expected reasoning delta, got {other:?}"),
     }
 
@@ -333,12 +333,7 @@ async fn streams_reasoning_from_final_message() {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn streams_reasoning_before_tool_call() {
-    if network_disabled() {
-        println!(
-            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
-        );
-        return;
-    }
+    skip_if_no_network!();
 
     let sse = concat!(
         "data: {\"choices\":[{\"delta\":{\"reasoning\":\"pre-tool\"}}]}\n\n",
@@ -354,7 +349,13 @@ async fn streams_reasoning_before_tool_call() {
     }
 
     match &events[1] {
-        ResponseEvent::ReasoningContentDelta(text) => assert_eq!(text, "pre-tool"),
+        ResponseEvent::ReasoningContentDelta {
+            delta,
+            content_index,
+        } => {
+            assert_eq!(delta, "pre-tool");
+            assert_eq!(content_index, &0);
+        }
         other => panic!("expected reasoning delta, got {other:?}"),
     }
 
@@ -383,12 +384,7 @@ async fn streams_reasoning_before_tool_call() {
 #[tokio::test]
 #[traced_test]
 async fn chat_sse_emits_failed_on_parse_error() {
-    if network_disabled() {
-        println!(
-            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
-        );
-        return;
-    }
+    skip_if_no_network!();
 
     let sse_body = concat!("data: not-json\n\n", "data: [DONE]\n\n");
 
@@ -420,12 +416,7 @@ async fn chat_sse_emits_failed_on_parse_error() {
 #[tokio::test]
 #[traced_test]
 async fn chat_sse_done_chunk_emits_event() {
-    if network_disabled() {
-        println!(
-            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
-        );
-        return;
-    }
+    skip_if_no_network!();
 
     let sse_body = "data: [DONE]\n\n";
 
@@ -443,12 +434,7 @@ async fn chat_sse_done_chunk_emits_event() {
 #[tokio::test]
 #[traced_test]
 async fn chat_sse_emits_error_on_invalid_utf8() {
-    if network_disabled() {
-        println!(
-            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
-        );
-        return;
-    }
+    skip_if_no_network!();
 
     let _ = run_stream_with_bytes(b"data: \x80\x80\n\n").await;
 

codex-rs/core/tests/common/responses.rs

@@ -1,6 +1,7 @@
 use std::sync::Arc;
 use std::sync::Mutex;
 
+use anyhow::Result;
 use serde_json::Value;
 use wiremock::BodyPrintLimit;
 use wiremock::Match;
@@ -12,6 +13,8 @@ use wiremock::ResponseTemplate;
 use wiremock::matchers::method;
 use wiremock::matchers::path_regex;
 
+use crate::test_codex::ApplyPatchModelOutput;
+
 #[derive(Debug, Clone)]
 pub struct ResponseMock {
     requests: Arc<Mutex<Vec<ResponsesRequest>>>,
@@ -36,6 +39,10 @@ impl ResponseMock {
         self.requests.lock().unwrap().clone()
     }
 
+    pub fn last_request(&self) -> Option<ResponsesRequest> {
+        self.requests.lock().unwrap().last().cloned()
+    }
+
     /// Returns true if any captured request contains a `function_call` with the
     /// provided `call_id`.
     pub fn saw_function_call(&self, call_id: &str) -> bool {
@@ -61,6 +68,18 @@ impl ResponsesRequest {
         self.0.body_json().unwrap()
     }
 
+    /// Returns all `input_text` spans from `message` inputs for the provided role.
+    pub fn message_input_texts(&self, role: &str) -> Vec<String> {
+        self.inputs_of_type("message")
+            .into_iter()
+            .filter(|item| item.get("role").and_then(Value::as_str) == Some(role))
+            .filter_map(|item| item.get("content").and_then(Value::as_array).cloned())
+            .flatten()
+            .filter(|span| span.get("type").and_then(Value::as_str) == Some("input_text"))
+            .filter_map(|span| span.get("text").and_then(Value::as_str).map(str::to_owned))
+            .collect()
+    }
+
     pub fn input(&self) -> Vec<Value> {
         self.0.body_json::<Value>().unwrap()["input"]
             .as_array()
@@ -116,6 +135,42 @@ impl ResponsesRequest {
             .map(str::to_string)
     }
 
+    pub fn function_call_output_content_and_success(
+        &self,
+        call_id: &str,
+    ) -> Option<(Option<String>, Option<bool>)> {
+        self.call_output_content_and_success(call_id, "function_call_output")
+    }
+
+    pub fn custom_tool_call_output_content_and_success(
+        &self,
+        call_id: &str,
+    ) -> Option<(Option<String>, Option<bool>)> {
+        self.call_output_content_and_success(call_id, "custom_tool_call_output")
+    }
+
+    fn call_output_content_and_success(
+        &self,
+        call_id: &str,
+        call_type: &str,
+    ) -> Option<(Option<String>, Option<bool>)> {
+        let output = self
+            .call_output(call_id, call_type)
+            .get("output")
+            .cloned()
+            .unwrap_or(Value::Null);
+        match output {
+            Value::String(text) => Some((Some(text), None)),
+            Value::Object(obj) => Some((
+                obj.get("content")
+                    .and_then(Value::as_str)
+                    .map(str::to_string),
+                obj.get("success").and_then(Value::as_bool),
+            )),
+            _ => Some((None, None)),
+        }
+    }
+
     pub fn header(&self, name: &str) -> Option<String> {
         self.0
             .headers
@@ -282,6 +337,7 @@ pub fn ev_reasoning_summary_text_delta(delta: &str) -> Value {
     serde_json::json!({
         "type": "response.reasoning_summary_text.delta",
         "delta": delta,
+        "summary_index": 0,
     })
 }
 
@@ -289,6 +345,7 @@ pub fn ev_reasoning_text_delta(delta: &str) -> Value {
     serde_json::json!({
         "type": "response.reasoning_text.delta",
         "delta": delta,
+        "content_index": 0,
     })
 }
 
@@ -355,6 +412,21 @@ pub fn ev_local_shell_call(call_id: &str, status: &str, command: Vec<&str>) -> V
     })
 }
 
+pub fn ev_apply_patch_call(
+    call_id: &str,
+    patch: &str,
+    output_type: ApplyPatchModelOutput,
+) -> Value {
+    match output_type {
+        ApplyPatchModelOutput::Freeform => ev_apply_patch_custom_tool_call(call_id, patch),
+        ApplyPatchModelOutput::Function => ev_apply_patch_function_call(call_id, patch),
+        ApplyPatchModelOutput::Shell => ev_apply_patch_shell_call(call_id, patch),
+        ApplyPatchModelOutput::ShellViaHeredoc => {
+            ev_apply_patch_shell_call_via_heredoc(call_id, patch)
+        }
+    }
+}
+
 /// Convenience: SSE event for an `apply_patch` custom tool call with raw patch
 /// text. This mirrors the payload produced by the Responses API when the model
 /// invokes `apply_patch` directly (before we convert it to a function call).
@@ -388,6 +460,21 @@ pub fn ev_apply_patch_function_call(call_id: &str, patch: &str) -> Value {
     })
 }
 
+pub fn ev_apply_patch_shell_call(call_id: &str, patch: &str) -> Value {
+    let args = serde_json::json!({ "command": ["apply_patch", patch] });
+    let arguments = serde_json::to_string(&args).expect("serialize apply_patch arguments");
+
+    ev_function_call(call_id, "shell", &arguments)
+}
+
+pub fn ev_apply_patch_shell_call_via_heredoc(call_id: &str, patch: &str) -> Value {
+    let script = format!("apply_patch <<'EOF'\n{patch}\nEOF\n");
+    let args = serde_json::json!({ "command": ["bash", "-lc", script] });
+    let arguments = serde_json::to_string(&args).expect("serialize apply_patch arguments");
+
+    ev_function_call(call_id, "shell", &arguments)
+}
+
 pub fn sse_failed(id: &str, code: &str, message: &str) -> String {
     sse(vec![serde_json::json!({
         "type": "response.failed",
@@ -434,12 +521,6 @@ pub async fn mount_sse_once(server: &MockServer, body: String) -> ResponseMock {
     response_mock
 }
 
-pub async fn mount_sse(server: &MockServer, body: String) -> ResponseMock {
-    let (mock, response_mock) = base_mock();
-    mock.respond_with(sse_response(body)).mount(server).await;
-    response_mock
-}
-
 pub async fn start_mock_server() -> MockServer {
     MockServer::builder()
         .body_print_limit(BodyPrintLimit::Limited(80_000))
@@ -447,6 +528,37 @@ pub async fn start_mock_server() -> MockServer {
         .await
 }
 
+#[derive(Clone)]
+pub struct FunctionCallResponseMocks {
+    pub function_call: ResponseMock,
+    pub completion: ResponseMock,
+}
+
+pub async fn mount_function_call_agent_response(
+    server: &MockServer,
+    call_id: &str,
+    arguments: &str,
+    tool_name: &str,
+) -> FunctionCallResponseMocks {
+    let first_response = sse(vec![
+        ev_response_created("resp-1"),
+        ev_function_call(call_id, tool_name, arguments),
+        ev_completed("resp-1"),
+    ]);
+    let function_call = mount_sse_once(server, first_response).await;
+
+    let second_response = sse(vec![
+        ev_assistant_message("msg-1", "done"),
+        ev_completed("resp-2"),
+    ]);
+    let completion = mount_sse_once(server, second_response).await;
+
+    FunctionCallResponseMocks {
+        function_call,
+        completion,
+    }
+}
+
 /// Mounts a sequence of SSE response bodies and serves them in order for each
 /// POST to `/v1/responses`. Panics if more requests are received than bodies
 /// provided. Also asserts the exact number of expected calls.

codex-rs/core/tests/common/test_codex.rs

@@ -11,6 +11,7 @@ use codex_core::ModelProviderInfo;
 use codex_core::built_in_model_providers;
 use codex_core::config::Config;
 use codex_core::features::Feature;
+use codex_core::model_family::find_family_for_model;
 use codex_core::protocol::AskForApproval;
 use codex_core::protocol::EventMsg;
 use codex_core::protocol::Op;
@@ -28,6 +29,15 @@ use crate::wait_for_event;
 
 type ConfigMutator = dyn FnOnce(&mut Config) + Send;
 
+/// A collection of different ways the model can output an apply_patch call
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
+pub enum ApplyPatchModelOutput {
+    Freeform,
+    Function,
+    Shell,
+    ShellViaHeredoc,
+}
+
 pub struct TestCodexBuilder {
     config_mutators: Vec<Box<ConfigMutator>>,
 }
@@ -41,6 +51,14 @@ impl TestCodexBuilder {
         self
     }
 
+    pub fn with_model(self, model: &str) -> Self {
+        let new_model = model.to_string();
+        self.with_config(move |config| {
+            config.model = new_model.clone();
+            config.model_family = find_family_for_model(&new_model).expect("model family");
+        })
+    }
+
     pub async fn build(&mut self, server: &wiremock::MockServer) -> anyhow::Result<TestCodex> {
         let home = Arc::new(TempDir::new()?);
         self.build_with_home(server, home, None).await
@@ -62,6 +80,7 @@ impl TestCodexBuilder {
         resume_from: Option<PathBuf>,
     ) -> anyhow::Result<TestCodex> {
         let (config, cwd) = self.prepare_config(server, &home).await?;
+
         let conversation_manager = ConversationManager::with_auth(CodexAuth::from_api_key("dummy"));
 
         let new_conversation = match resume_from {
@@ -70,15 +89,20 @@ impl TestCodexBuilder {
                     CodexAuth::from_api_key("dummy"),
                 );
                 conversation_manager
-                    .resume_conversation_from_rollout(config, path, auth_manager)
+                    .resume_conversation_from_rollout(config.clone(), path, auth_manager)
+                    .await?
+            }
+            None => {
+                conversation_manager
+                    .new_conversation(config.clone())
                     .await?
             }
-            None => conversation_manager.new_conversation(config).await?,
         };
 
         Ok(TestCodex {
             home,
             cwd,
+            config,
             codex: new_conversation.conversation,
             session_configured: new_conversation.session_configured,
         })
@@ -122,6 +146,7 @@ pub struct TestCodex {
     pub cwd: Arc<TempDir>,
     pub codex: Arc<CodexConversation>,
     pub session_configured: SessionConfiguredEvent,
+    pub config: Config,
 }
 
 impl TestCodex {
@@ -134,14 +159,28 @@ impl TestCodex {
     }
 
     pub async fn submit_turn(&self, prompt: &str) -> Result<()> {
-        self.submit_turn_with_policy(prompt, SandboxPolicy::DangerFullAccess)
-            .await
+        self.submit_turn_with_policies(
+            prompt,
+            AskForApproval::Never,
+            SandboxPolicy::DangerFullAccess,
+        )
+        .await
     }
 
     pub async fn submit_turn_with_policy(
         &self,
         prompt: &str,
         sandbox_policy: SandboxPolicy,
+    ) -> Result<()> {
+        self.submit_turn_with_policies(prompt, AskForApproval::Never, sandbox_policy)
+            .await
+    }
+
+    pub async fn submit_turn_with_policies(
+        &self,
+        prompt: &str,
+        approval_policy: AskForApproval,
+        sandbox_policy: SandboxPolicy,
     ) -> Result<()> {
         let session_model = self.session_configured.model.clone();
         self.codex
@@ -151,7 +190,7 @@ impl TestCodex {
                 }],
                 final_output_json_schema: None,
                 cwd: self.cwd.path().to_path_buf(),
-                approval_policy: AskForApproval::Never,
+                approval_policy,
                 sandbox_policy,
                 model: session_model,
                 effort: None,
@@ -249,6 +288,19 @@ impl TestCodexHarness {
             .expect("output string")
             .to_string()
     }
+
+    pub async fn apply_patch_output(
+        &self,
+        call_id: &str,
+        output_type: ApplyPatchModelOutput,
+    ) -> String {
+        match output_type {
+            ApplyPatchModelOutput::Freeform => self.custom_tool_call_output(call_id).await,
+            ApplyPatchModelOutput::Function
+            | ApplyPatchModelOutput::Shell
+            | ApplyPatchModelOutput::ShellViaHeredoc => self.function_call_stdout(call_id).await,
+        }
+    }
 }
 
 fn custom_tool_call_output<'a>(bodies: &'a [Value], call_id: &str) -> &'a Value {

codex-rs/core/tests/suite/abort_tasks.rs

@@ -13,7 +13,7 @@ use core_test_support::responses::mount_sse_sequence;
 use core_test_support::responses::sse;
 use core_test_support::responses::start_mock_server;
 use core_test_support::test_codex::test_codex;
-use core_test_support::wait_for_event_with_timeout;
+use core_test_support::wait_for_event;
 use regex_lite::Regex;
 use serde_json::json;
 
@@ -42,8 +42,6 @@ async fn interrupt_long_running_tool_emits_turn_aborted() {
 
     let codex = test_codex().build(&server).await.unwrap().codex;
 
-    let wait_timeout = Duration::from_secs(5);
-
     // Kick off a turn that triggers the function call.
     codex
         .submit(Op::UserInput {
@@ -55,22 +53,12 @@ async fn interrupt_long_running_tool_emits_turn_aborted() {
         .unwrap();
 
     // Wait until the exec begins to avoid a race, then interrupt.
-    wait_for_event_with_timeout(
-        &codex,
-        |ev| matches!(ev, EventMsg::ExecCommandBegin(_)),
-        wait_timeout,
-    )
-    .await;
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::ExecCommandBegin(_))).await;
 
     codex.submit(Op::Interrupt).await.unwrap();
 
     // Expect TurnAborted soon after.
-    wait_for_event_with_timeout(
-        &codex,
-        |ev| matches!(ev, EventMsg::TurnAborted(_)),
-        wait_timeout,
-    )
-    .await;
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TurnAborted(_))).await;
 }
 
 /// After an interrupt we expect the next request to the model to include both
@@ -107,8 +95,6 @@ async fn interrupt_tool_records_history_entries() {
     let fixture = test_codex().build(&server).await.unwrap();
     let codex = Arc::clone(&fixture.codex);
 
-    let wait_timeout = Duration::from_millis(100);
-
     codex
         .submit(Op::UserInput {
             items: vec![UserInput::Text {
@@ -118,22 +104,12 @@ async fn interrupt_tool_records_history_entries() {
         .await
         .unwrap();
 
-    wait_for_event_with_timeout(
-        &codex,
-        |ev| matches!(ev, EventMsg::ExecCommandBegin(_)),
-        wait_timeout,
-    )
-    .await;
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::ExecCommandBegin(_))).await;
 
     tokio::time::sleep(Duration::from_secs_f32(0.1)).await;
     codex.submit(Op::Interrupt).await.unwrap();
 
-    wait_for_event_with_timeout(
-        &codex,
-        |ev| matches!(ev, EventMsg::TurnAborted(_)),
-        wait_timeout,
-    )
-    .await;
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TurnAborted(_))).await;
 
     codex
         .submit(Op::UserInput {
@@ -144,12 +120,7 @@ async fn interrupt_tool_records_history_entries() {
         .await
         .unwrap();
 
-    wait_for_event_with_timeout(
-        &codex,
-        |ev| matches!(ev, EventMsg::TaskComplete(_)),
-        wait_timeout,
-    )
-    .await;
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
 
     let requests = response_mock.requests();
     assert!(