serializing noMatch as an object

Co-authored-by: Michael Bolin <mbolin@openai.com>

.github/codex/home/config.toml

@@ -1,3 +1,3 @@
-model = "gpt-5"
+model = "gpt-5.1"
 
 # Consider setting [mcp_servers] here!

.github/workflows/cla.yml

@@ -13,6 +13,9 @@ permissions:
 
 jobs:
   cla:
+    # Only run the CLA assistant for the canonical openai repo so forks are not blocked
+    # and contributors who signed previously do not receive duplicate CLA notifications.
+    if: ${{ github.repository_owner == 'openai' }}
     runs-on: ubuntu-latest
     steps:
       - uses: contributor-assistant/github-action@v2.6.1
@@ -43,4 +46,6 @@ jobs:
           path-to-document: https://github.com/openai/codex/blob/main/docs/CLA.md
           path-to-signatures: signatures/cla.json
           branch: cla-signatures
-          allowlist: dependabot[bot]
+          allowlist: |
+            codex
+            dependabot[bot]

.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@v8
+        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/issue-deduplicator.yml

@@ -46,7 +46,7 @@ jobs:
         with:
           openai-api-key: ${{ secrets.CODEX_OPENAI_API_KEY }}
           allow-users: "*"
-          model: gpt-5
+          model: gpt-5.1
           prompt: |
             You are an assistant that triages new GitHub issues by identifying potential duplicates.
 

.github/workflows/rust-ci.yml

@@ -95,8 +95,8 @@ jobs:
       run:
         working-directory: codex-rs
     env:
-      # Speed up repeated builds across CI runs by caching compiled objects.
-      RUSTC_WRAPPER: sccache
+      # Speed up repeated builds across CI runs by caching compiled objects (non-Windows).
+      USE_SCCACHE: ${{ startsWith(matrix.runner, 'windows') && 'false' || 'true' }}
       CARGO_INCREMENTAL: "0"
       SCCACHE_CACHE_SIZE: 10G
 
@@ -170,12 +170,14 @@ jobs:
 
       # Install and restore sccache cache
       - name: Install sccache
+        if: ${{ env.USE_SCCACHE == 'true' }}
         uses: taiki-e/install-action@44c6d64aa62cd779e873306675c7a58e86d6d532 # v2
         with:
           tool: sccache
           version: 0.7.5
 
       - name: Configure sccache backend
+        if: ${{ env.USE_SCCACHE == 'true' }}
         shell: bash
         run: |
           set -euo pipefail
@@ -188,8 +190,13 @@ jobs:
             echo "Using sccache local disk + actions/cache fallback"
           fi
 
+      - name: Enable sccache wrapper
+        if: ${{ env.USE_SCCACHE == 'true' }}
+        shell: bash
+        run: echo "RUSTC_WRAPPER=sccache" >> "$GITHUB_ENV"
+
       - name: Restore sccache cache (fallback)
-        if: ${{ env.SCCACHE_GHA_ENABLED != 'true' }}
+        if: ${{ env.USE_SCCACHE == 'true' && env.SCCACHE_GHA_ENABLED != 'true' }}
         id: cache_sccache_restore
         uses: actions/cache/restore@v4
         with:
@@ -274,7 +281,7 @@ jobs:
           key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ hashFiles('codex-rs/rust-toolchain.toml') }}
 
       - name: Save sccache cache (fallback)
-        if: always() && !cancelled() && env.SCCACHE_GHA_ENABLED != 'true'
+        if: always() && !cancelled() && env.USE_SCCACHE == 'true' && env.SCCACHE_GHA_ENABLED != 'true'
         continue-on-error: true
         uses: actions/cache/save@v4
         with:
@@ -282,12 +289,12 @@ jobs:
           key: sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ github.run_id }}
 
       - name: sccache stats
-        if: always()
+        if: always() && env.USE_SCCACHE == 'true'
         continue-on-error: true
         run: sccache --show-stats || true
 
       - name: sccache summary
-        if: always()
+        if: always() && env.USE_SCCACHE == 'true'
         shell: bash
         run: |
           {
@@ -326,7 +333,8 @@ jobs:
       run:
         working-directory: codex-rs
     env:
-      RUSTC_WRAPPER: sccache
+      # Speed up repeated builds across CI runs by caching compiled objects (non-Windows).
+      USE_SCCACHE: ${{ startsWith(matrix.runner, 'windows') && 'false' || 'true' }}
       CARGO_INCREMENTAL: "0"
       SCCACHE_CACHE_SIZE: 10G
 
@@ -370,12 +378,14 @@ jobs:
             cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-
 
       - name: Install sccache
+        if: ${{ env.USE_SCCACHE == 'true' }}
         uses: taiki-e/install-action@44c6d64aa62cd779e873306675c7a58e86d6d532 # v2
         with:
           tool: sccache
           version: 0.7.5
 
       - name: Configure sccache backend
+        if: ${{ env.USE_SCCACHE == 'true' }}
         shell: bash
         run: |
           set -euo pipefail
@@ -388,8 +398,13 @@ jobs:
             echo "Using sccache local disk + actions/cache fallback"
           fi
 
+      - name: Enable sccache wrapper
+        if: ${{ env.USE_SCCACHE == 'true' }}
+        shell: bash
+        run: echo "RUSTC_WRAPPER=sccache" >> "$GITHUB_ENV"
+
       - name: Restore sccache cache (fallback)
-        if: ${{ env.SCCACHE_GHA_ENABLED != 'true' }}
+        if: ${{ env.USE_SCCACHE == 'true' && env.SCCACHE_GHA_ENABLED != 'true' }}
         id: cache_sccache_restore
         uses: actions/cache/restore@v4
         with:
@@ -406,10 +421,10 @@ jobs:
 
       - name: tests
         id: test
-        continue-on-error: true
         run: cargo nextest run --all-features --no-fail-fast --target ${{ matrix.target }} --cargo-profile ci-test
         env:
           RUST_BACKTRACE: 1
+          NEXTEST_STATUS_LEVEL: leak
 
       - name: Save cargo home cache
         if: always() && !cancelled() && steps.cache_cargo_home_restore.outputs.cache-hit != 'true'
@@ -424,7 +439,7 @@ jobs:
           key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ hashFiles('codex-rs/rust-toolchain.toml') }}
 
       - name: Save sccache cache (fallback)
-        if: always() && !cancelled() && env.SCCACHE_GHA_ENABLED != 'true'
+        if: always() && !cancelled() && env.USE_SCCACHE == 'true' && env.SCCACHE_GHA_ENABLED != 'true'
         continue-on-error: true
         uses: actions/cache/save@v4
         with:
@@ -432,12 +447,12 @@ jobs:
           key: sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ github.run_id }}
 
       - name: sccache stats
-        if: always()
+        if: always() && env.USE_SCCACHE == 'true'
         continue-on-error: true
         run: sccache --show-stats || true
 
       - name: sccache summary
-        if: always()
+        if: always() && env.USE_SCCACHE == 'true'
         shell: bash
         run: |
           {

.gitignore

@@ -64,6 +64,9 @@ apply_patch/
 # coverage
 coverage/
 
+# personal files
+personal/
+
 # os
 .DS_Store
 Thumbs.db

AGENTS.md

@@ -85,6 +85,7 @@ If you don’t have the tool:
 - `ResponsesRequest` exposes helpers (`body_json`, `input`, `function_call_output`, `custom_tool_call_output`, `call_output`, `header`, `path`, `query_param`) so assertions can target structured payloads instead of manual JSON digging.
 - Build SSE payloads with the provided `ev_*` constructors and the `sse(...)`.
 - Prefer `wait_for_event` over `wait_for_event_with_timeout`.
+- Prefer `mount_sse_once` over `mount_sse_once_match` or `mount_sse_sequence`
 
 - Typical pattern:
 

README.md

@@ -69,6 +69,39 @@ Codex can access MCP servers. To configure them, refer to the [config docs](./do
 
 Codex CLI supports a rich set of configuration options, with preferences stored in `~/.codex/config.toml`. For full configuration options, see [Configuration](./docs/config.md).
 
+### Execpolicy quickstart
+
+Codex can enforce your own rules-based execution policy before it runs shell commands.
+
+1. Create a policy directory: `mkdir -p ~/.codex/policy`.
+2. Create one or more `.codexpolicy` files into that folder. Codex automatically loads every `.codexpolicy` file in there on startup.
+3. Write `prefix_rule` entries to describe the commands you want to allow, prompt, or block:
+
+```starlark
+prefix_rule(
+    pattern = ["git", ["push", "fetch"]],
+    decision = "prompt",  # allow | prompt | forbidden
+    match = [["git", "push", "origin", "main"]],  # examples that must match
+    not_match = [["git", "status"]],              # examples that must not match
+)
+```
+
+- `pattern` is a list of shell tokens, evaluated from left to right; wrap tokens in a nested list to express alternatives (e.g., match both `push` and `fetch`).
+- `decision` sets the severity; Codex picks the strictest decision when multiple rules match.
+- `match` and `not_match` act as (optional) unit tests. Codex validates them when it loads your policy, so you get feedback if an example has unexpected behavior.
+
+In this example rule, if Codex wants to run commands with the prefix `git push` or `git fetch`, it will first ask for user approval.
+
+Note: If Codex wants to run a command that matches with multiple rules, it will use the strictest decision among the matched rules (forbidden > prompt > allow).
+
+Use the `codex execpolicycheck` subcommand to preview decisions before you save a rule (see the [`execpolicy2` README](./codex-rs/execpolicy2/README.md) for syntax details):
+
+```shell
+codex execpolicycheck --policy ~/.codex/policy/default.codexpolicy git push origin main
+```
+
+Pass multiple `--policy` flags to test how several files combine, and use `--pretty` for formatted JSON output. See the [`codex-rs/execpolicy2` README](./codex-rs/execpolicy2/README.md) for a more detailed walkthrough of the available syntax.
+
 ---
 
 ### Docs & FAQ

codex-rs/.config/nextest.toml

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

codex-rs/Cargo.lock

@@ -850,6 +850,7 @@ dependencies = [
  "codex-protocol",
  "codex-utils-json-to-toml",
  "core_test_support",
+ "mcp-types",
  "opentelemetry-appender-tracing",
  "os_info",
  "pretty_assertions",
@@ -873,7 +874,6 @@ dependencies = [
  "clap",
  "codex-protocol",
  "mcp-types",
- "paste",
  "pretty_assertions",
  "schemars 0.8.22",
  "serde",
@@ -883,6 +883,19 @@ dependencies = [
  "uuid",
 ]
 
+[[package]]
+name = "codex-app-server-test-client"
+version = "0.0.0"
+dependencies = [
+ "anyhow",
+ "clap",
+ "codex-app-server-protocol",
+ "codex-protocol",
+ "serde",
+ "serde_json",
+ "uuid",
+]
+
 [[package]]
 name = "codex-apply-patch"
 version = "0.0.0"
@@ -976,6 +989,7 @@ dependencies = [
  "codex-common",
  "codex-core",
  "codex-exec",
+ "codex-execpolicy2",
  "codex-login",
  "codex-mcp-server",
  "codex-process-hardening",
@@ -1047,7 +1061,10 @@ dependencies = [
  "clap",
  "codex-app-server-protocol",
  "codex-core",
+ "codex-lmstudio",
+ "codex-ollama",
  "codex-protocol",
+ "once_cell",
  "serde",
  "toml",
 ]
@@ -1069,6 +1086,7 @@ dependencies = [
  "codex-apply-patch",
  "codex-arg0",
  "codex-async-utils",
+ "codex-execpolicy2",
  "codex-file-search",
  "codex-git",
  "codex-keyring-store",
@@ -1114,6 +1132,7 @@ dependencies = [
  "similar",
  "strum_macros 0.27.2",
  "tempfile",
+ "test-case",
  "test-log",
  "thiserror 2.0.17",
  "time",
@@ -1143,7 +1162,6 @@ dependencies = [
  "codex-arg0",
  "codex-common",
  "codex-core",
- "codex-ollama",
  "codex-protocol",
  "core_test_support",
  "libc",
@@ -1186,6 +1204,21 @@ dependencies = [
  "tempfile",
 ]
 
+[[package]]
+name = "codex-execpolicy2"
+version = "0.0.0"
+dependencies = [
+ "anyhow",
+ "clap",
+ "multimap",
+ "pretty_assertions",
+ "serde",
+ "serde_json",
+ "shlex",
+ "starlark",
+ "thiserror 2.0.17",
+]
+
 [[package]]
 name = "codex-feedback"
 version = "0.0.0"
@@ -1247,6 +1280,19 @@ dependencies = [
  "tokio",
 ]
 
+[[package]]
+name = "codex-lmstudio"
+version = "0.0.0"
+dependencies = [
+ "codex-core",
+ "reqwest",
+ "serde_json",
+ "tokio",
+ "tracing",
+ "which",
+ "wiremock",
+]
+
 [[package]]
 name = "codex-login"
 version = "0.0.0"
@@ -1411,6 +1457,7 @@ dependencies = [
  "tracing",
  "urlencoding",
  "webbrowser",
+ "which",
 ]
 
 [[package]]
@@ -1438,12 +1485,12 @@ dependencies = [
  "codex-ansi-escape",
  "codex-app-server-protocol",
  "codex-arg0",
+ "codex-backend-client",
  "codex-common",
  "codex-core",
  "codex-feedback",
  "codex-file-search",
  "codex-login",
- "codex-ollama",
  "codex-protocol",
  "codex-windows-sandbox",
  "color-eyre",
@@ -1466,6 +1513,7 @@ dependencies = [
  "ratatui",
  "ratatui-macros",
  "regex-lite",
+ "reqwest",
  "serde",
  "serde_json",
  "serial_test",
@@ -1548,9 +1596,11 @@ name = "codex-utils-tokenizer"
 version = "0.0.0"
 dependencies = [
  "anyhow",
+ "codex-utils-cache",
  "pretty_assertions",
  "thiserror 2.0.17",
  "tiktoken-rs",
+ "tokio",
 ]
 
 [[package]]
@@ -6160,6 +6210,39 @@ 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"
@@ -6267,9 +6350,9 @@ dependencies = [
 
 [[package]]
 name = "tiktoken-rs"
-version = "0.7.0"
+version = "0.9.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "25563eeba904d770acf527e8b370fe9a5547bacd20ff84a0b6c3bc41288e5625"
+checksum = "3a19830747d9034cd9da43a60eaa8e552dfda7712424aebf187b7a60126bae0d"
 dependencies = [
  "anyhow",
  "base64",

codex-rs/Cargo.toml

@@ -5,6 +5,7 @@ members = [
     "async-utils",
     "app-server",
     "app-server-protocol",
+    "app-server-test-client",
     "apply-patch",
     "arg0",
     "feedback",
@@ -16,9 +17,11 @@ members = [
     "core",
     "exec",
     "execpolicy",
+    "execpolicy2",
     "keyring-store",
     "file-search",
     "linux-sandbox",
+    "lmstudio",
     "login",
     "mcp-server",
     "mcp-types",
@@ -63,11 +66,13 @@ codex-chatgpt = { path = "chatgpt" }
 codex-common = { path = "common" }
 codex-core = { path = "core" }
 codex-exec = { path = "exec" }
+codex-execpolicy2 = { path = "execpolicy2" }
 codex-feedback = { path = "feedback" }
 codex-file-search = { path = "file-search" }
 codex-git = { path = "utils/git" }
 codex-keyring-store = { path = "keyring-store" }
 codex-linux-sandbox = { path = "linux-sandbox" }
+codex-lmstudio = { path = "lmstudio" }
 codex-login = { path = "login" }
 codex-mcp-server = { path = "mcp-server" }
 codex-ollama = { path = "ollama" }
@@ -122,14 +127,14 @@ 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"
 insta = "1.43.2"
 itertools = "0.14.0"
-keyring = "3.6"
+keyring = { version = "3.6", default-features = false }
 landlock = "0.4.1"
 lazy_static = "1"
 libc = "0.2.175"
@@ -140,6 +145,7 @@ 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"
@@ -148,7 +154,6 @@ opentelemetry-semantic-conventions = "0.30.0"
 opentelemetry_sdk = "0.30.0"
 os_info = "3.12.0"
 owo-colors = "4.2.0"
-paste = "1.0.15"
 path-absolutize = "3.1.1"
 pathdiff = "0.2"
 portable-pty = "0.9.0"
@@ -181,6 +186,7 @@ tempfile = "3.23.0"
 test-log = "0.2.18"
 textwrap = "0.16.2"
 thiserror = "2.0.17"
+tiktoken-rs = "0.9"
 time = "0.3"
 tiny_http = "0.12"
 tokio = "1"
@@ -278,8 +284,8 @@ 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/app-server-protocol/Cargo.toml

@@ -15,7 +15,6 @@ anyhow = { workspace = true }
 clap = { workspace = true, features = ["derive"] }
 codex-protocol = { workspace = true }
 mcp-types = { workspace = true }
-paste = { workspace = true }
 schemars = { workspace = true }
 serde = { workspace = true, features = ["derive"] }
 serde_json = { workspace = true }

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;
@@ -120,10 +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();
@@ -152,13 +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();
@@ -171,6 +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);
         }
@@ -181,6 +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;
                 }
@@ -386,14 +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()
         {
@@ -719,6 +708,7 @@ mod tests {
     use uuid::Uuid;
 
     #[test]
+    #[ignore = "timing out"]
     fn generated_ts_has_no_optional_nullable_fields() -> Result<()> {
         // Assert that there are no types of the form "?: T | null" in the generated TS files.
         let output_dir = std::env::temp_dir().join(format!("codex_ts_types_{}", Uuid::now_v7()));

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

@@ -1,6 +1,4 @@
-use std::collections::HashMap;
 use std::path::Path;
-use std::path::PathBuf;
 
 use crate::JSONRPCNotification;
 use crate::JSONRPCRequest;
@@ -9,12 +7,6 @@ use crate::export::GeneratedSchema;
 use crate::export::write_json_schema;
 use crate::protocol::v1;
 use crate::protocol::v2;
-use codex_protocol::ConversationId;
-use codex_protocol::parse_command::ParsedCommand;
-use codex_protocol::protocol::FileChange;
-use codex_protocol::protocol::ReviewDecision;
-use codex_protocol::protocol::SandboxCommandAssessment;
-use paste::paste;
 use schemars::JsonSchema;
 use serde::Deserialize;
 use serde::Serialize;
@@ -277,34 +269,36 @@ macro_rules! server_request_definitions {
     (
         $(
             $(#[$variant_meta:meta])*
-            $variant:ident
+            $variant:ident $(=> $wire:literal)? {
+                params: $params:ty,
+                response: $response:ty,
+            }
         ),* $(,)?
     ) => {
-        paste! {
-            /// Request initiated from the server and sent to the client.
-            #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
-            #[serde(tag = "method", rename_all = "camelCase")]
-            pub enum ServerRequest {
-                $(
-                    $(#[$variant_meta])*
-                    $variant {
-                        #[serde(rename = "id")]
-                        request_id: RequestId,
-                        params: [<$variant Params>],
-                    },
-                )*
-            }
+        /// Request initiated from the server and sent to the client.
+        #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+        #[serde(tag = "method", rename_all = "camelCase")]
+        pub enum ServerRequest {
+            $(
+                $(#[$variant_meta])*
+                $(#[serde(rename = $wire)] #[ts(rename = $wire)])?
+                $variant {
+                    #[serde(rename = "id")]
+                    request_id: RequestId,
+                    params: $params,
+                },
+            )*
+        }
 
-            #[derive(Debug, Clone, PartialEq, JsonSchema)]
-            pub enum ServerRequestPayload {
-                $( $variant([<$variant Params>]), )*
-            }
+        #[derive(Debug, Clone, PartialEq, JsonSchema)]
+        pub enum ServerRequestPayload {
+            $( $variant($params), )*
+        }
 
-            impl ServerRequestPayload {
-                pub fn request_with_id(self, request_id: RequestId) -> ServerRequest {
-                    match self {
-                        $(Self::$variant(params) => ServerRequest::$variant { request_id, params },)*
-                    }
+        impl ServerRequestPayload {
+            pub fn request_with_id(self, request_id: RequestId) -> ServerRequest {
+                match self {
+                    $(Self::$variant(params) => ServerRequest::$variant { request_id, params },)*
                 }
             }
         }
@@ -312,9 +306,9 @@ macro_rules! server_request_definitions {
         pub fn export_server_responses(
             out_dir: &::std::path::Path,
         ) -> ::std::result::Result<(), ::ts_rs::ExportError> {
-            paste! {
-                $(<[<$variant Response>] as ::ts_rs::TS>::export_all_to(out_dir)?;)*
-            }
+            $(
+                <$response as ::ts_rs::TS>::export_all_to(out_dir)?;
+            )*
             Ok(())
         }
 
@@ -323,9 +317,12 @@ macro_rules! server_request_definitions {
             out_dir: &Path,
         ) -> ::anyhow::Result<Vec<GeneratedSchema>> {
             let mut schemas = Vec::new();
-            paste! {
-                $(schemas.push(crate::export::write_json_schema::<[<$variant Response>]>(out_dir, stringify!([<$variant Response>]))?);)*
-            }
+            $(
+                schemas.push(crate::export::write_json_schema::<$response>(
+                    out_dir,
+                    concat!(stringify!($variant), "Response"),
+                )?);
+            )*
             Ok(schemas)
         }
 
@@ -334,9 +331,12 @@ macro_rules! server_request_definitions {
             out_dir: &Path,
         ) -> ::anyhow::Result<Vec<GeneratedSchema>> {
             let mut schemas = Vec::new();
-            paste! {
-                $(schemas.push(crate::export::write_json_schema::<[<$variant Params>]>(out_dir, stringify!([<$variant Params>]))?);)*
-            }
+            $(
+                schemas.push(crate::export::write_json_schema::<$params>(
+                    out_dir,
+                    concat!(stringify!($variant), "Params"),
+                )?);
+            )*
             Ok(schemas)
         }
     };
@@ -426,49 +426,27 @@ impl TryFrom<JSONRPCRequest> for ServerRequest {
 }
 
 server_request_definitions! {
+    /// NEW APIs
+    /// Sent when approval is requested for a specific command execution.
+    /// This request is used for Turns started via turn/start.
+    CommandExecutionRequestApproval => "item/commandExecution/requestApproval" {
+        params: v2::CommandExecutionRequestApprovalParams,
+        response: v2::CommandExecutionRequestApprovalResponse,
+    },
+
+    /// DEPRECATED APIs below
     /// Request to approve a patch.
-    ApplyPatchApproval,
+    /// This request is used for Turns started via the legacy APIs (i.e. SendUserTurn, SendUserMessage).
+    ApplyPatchApproval {
+        params: v1::ApplyPatchApprovalParams,
+        response: v1::ApplyPatchApprovalResponse,
+    },
     /// Request to exec a command.
-    ExecCommandApproval,
-}
-
-#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
-#[serde(rename_all = "camelCase")]
-pub struct ApplyPatchApprovalParams {
-    pub conversation_id: ConversationId,
-    /// Use to correlate this with [codex_core::protocol::PatchApplyBeginEvent]
-    /// and [codex_core::protocol::PatchApplyEndEvent].
-    pub call_id: String,
-    pub file_changes: HashMap<PathBuf, FileChange>,
-    /// Optional explanatory reason (e.g. request for extra write access).
-    pub reason: Option<String>,
-    /// When set, the agent is asking the user to allow writes under this root
-    /// for the remainder of the session (unclear if this is honored today).
-    pub grant_root: Option<PathBuf>,
-}
-
-#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
-#[serde(rename_all = "camelCase")]
-pub struct ExecCommandApprovalParams {
-    pub conversation_id: ConversationId,
-    /// Use to correlate this with [codex_core::protocol::ExecCommandBeginEvent]
-    /// and [codex_core::protocol::ExecCommandEndEvent].
-    pub call_id: String,
-    pub command: Vec<String>,
-    pub cwd: PathBuf,
-    pub reason: Option<String>,
-    pub risk: Option<SandboxCommandAssessment>,
-    pub parsed_cmd: Vec<ParsedCommand>,
-}
-
-#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
-pub struct ExecCommandApprovalResponse {
-    pub decision: ReviewDecision,
-}
-
-#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
-pub struct ApplyPatchApprovalResponse {
-    pub decision: ReviewDecision,
+    /// This request is used for Turns started via the legacy APIs (i.e. SendUserTurn, SendUserMessage).
+    ExecCommandApproval {
+        params: v1::ExecCommandApprovalParams,
+        response: v1::ExecCommandApprovalResponse,
+    },
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
@@ -508,6 +486,9 @@ 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")]
@@ -530,17 +511,20 @@ client_notification_definitions! {
 mod tests {
     use super::*;
     use anyhow::Result;
+    use codex_protocol::ConversationId;
     use codex_protocol::account::PlanType;
+    use codex_protocol::parse_command::ParsedCommand;
     use codex_protocol::protocol::AskForApproval;
     use pretty_assertions::assert_eq;
     use serde_json::json;
+    use std::path::PathBuf;
 
     #[test]
     fn serialize_new_conversation() -> Result<()> {
         let request = ClientRequest::NewConversation {
             request_id: RequestId::Integer(42),
             params: v1::NewConversationParams {
-                model: Some("gpt-5-codex".to_string()),
+                model: Some("gpt-5.1-codex".to_string()),
                 model_provider: None,
                 profile: None,
                 cwd: None,
@@ -558,7 +542,7 @@ mod tests {
                 "method": "newConversation",
                 "id": 42,
                 "params": {
-                    "model": "gpt-5-codex",
+                    "model": "gpt-5.1-codex",
                     "modelProvider": null,
                     "profile": null,
                     "cwd": null,
@@ -613,7 +597,7 @@ mod tests {
     #[test]
     fn serialize_server_request() -> Result<()> {
         let conversation_id = ConversationId::from_string("67e55044-10b1-426f-9247-bb680e5fe0c8")?;
-        let params = ExecCommandApprovalParams {
+        let params = v1::ExecCommandApprovalParams {
             conversation_id,
             call_id: "call-42".to_string(),
             command: vec!["echo".to_string(), "hello".to_string()],

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

@@ -8,8 +8,12 @@ use codex_protocol::config_types::ReasoningSummary;
 use codex_protocol::config_types::SandboxMode;
 use codex_protocol::config_types::Verbosity;
 use codex_protocol::models::ResponseItem;
+use codex_protocol::parse_command::ParsedCommand;
 use codex_protocol::protocol::AskForApproval;
 use codex_protocol::protocol::EventMsg;
+use codex_protocol::protocol::FileChange;
+use codex_protocol::protocol::ReviewDecision;
+use codex_protocol::protocol::SandboxCommandAssessment;
 use codex_protocol::protocol::SandboxPolicy;
 use codex_protocol::protocol::SessionSource;
 use codex_protocol::protocol::TurnAbortReason;
@@ -191,6 +195,46 @@ pub struct GitDiffToRemoteResponse {
     pub diff: String,
 }
 
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+pub struct ApplyPatchApprovalParams {
+    pub conversation_id: ConversationId,
+    /// Use to correlate this with [codex_core::protocol::PatchApplyBeginEvent]
+    /// and [codex_core::protocol::PatchApplyEndEvent].
+    pub call_id: String,
+    pub file_changes: HashMap<PathBuf, FileChange>,
+    /// Optional explanatory reason (e.g. request for extra write access).
+    pub reason: Option<String>,
+    /// When set, the agent is asking the user to allow writes under this root
+    /// for the remainder of the session (unclear if this is honored today).
+    pub grant_root: Option<PathBuf>,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+pub struct ApplyPatchApprovalResponse {
+    pub decision: ReviewDecision,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+pub struct ExecCommandApprovalParams {
+    pub conversation_id: ConversationId,
+    /// Use to correlate this with [codex_core::protocol::ExecCommandBeginEvent]
+    /// and [codex_core::protocol::ExecCommandEndEvent].
+    pub call_id: String,
+    pub command: Vec<String>,
+    pub cwd: PathBuf,
+    pub reason: Option<String>,
+    pub risk: Option<SandboxCommandAssessment>,
+    pub parsed_cmd: Vec<ParsedCommand>,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+pub struct ExecCommandApprovalResponse {
+    pub decision: ReviewDecision,
+}
+
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
 pub struct CancelLoginChatGptParams {

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

@@ -4,10 +4,13 @@ use std::path::PathBuf;
 use crate::protocol::common::AuthMode;
 use codex_protocol::ConversationId;
 use codex_protocol::account::PlanType;
+use codex_protocol::approvals::SandboxCommandAssessment as CoreSandboxCommandAssessment;
 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::parse_command::ParsedCommand as CoreParsedCommand;
 use codex_protocol::protocol::RateLimitSnapshot as CoreRateLimitSnapshot;
 use codex_protocol::protocol::RateLimitWindow as CoreRateLimitWindow;
 use codex_protocol::user_input::UserInput as CoreUserInput;
@@ -19,7 +22,7 @@ use serde_json::Value as JsonValue;
 use ts_rs::TS;
 
 // Macro to declare a camelCased API v2 enum mirroring a core enum which
-// tends to use kebab-case.
+// tends to use either snake_case or kebab-case.
 macro_rules! v2_enum_from_core {
     (
         pub enum $Name:ident from $Src:path { $( $Variant:ident ),+ $(,)? }
@@ -55,13 +58,32 @@ v2_enum_from_core!(
     }
 );
 
+v2_enum_from_core!(
+    pub enum CommandRiskLevel from codex_protocol::approvals::SandboxRiskLevel {
+        Low,
+        Medium,
+        High
+    }
+);
+
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq, JsonSchema, TS)]
-#[serde(tag = "mode", rename_all = "camelCase")]
-#[ts(tag = "mode")]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub enum ApprovalDecision {
+    Accept,
+    Decline,
+    Cancel,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq, JsonSchema, TS)]
+#[serde(tag = "type", rename_all = "camelCase")]
+#[ts(tag = "type")]
 #[ts(export_to = "v2/")]
 pub enum SandboxPolicy {
     DangerFullAccess,
     ReadOnly,
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
     WorkspaceWrite {
         #[serde(default)]
         writable_roots: Vec<PathBuf>,
@@ -118,6 +140,98 @@ impl From<codex_protocol::protocol::SandboxPolicy> for SandboxPolicy {
     }
 }
 
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct SandboxCommandAssessment {
+    pub description: String,
+    pub risk_level: CommandRiskLevel,
+}
+
+impl SandboxCommandAssessment {
+    pub fn into_core(self) -> CoreSandboxCommandAssessment {
+        CoreSandboxCommandAssessment {
+            description: self.description,
+            risk_level: self.risk_level.to_core(),
+        }
+    }
+}
+
+impl From<CoreSandboxCommandAssessment> for SandboxCommandAssessment {
+    fn from(value: CoreSandboxCommandAssessment) -> Self {
+        Self {
+            description: value.description,
+            risk_level: CommandRiskLevel::from(value.risk_level),
+        }
+    }
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(tag = "type", rename_all = "camelCase")]
+#[ts(tag = "type")]
+#[ts(export_to = "v2/")]
+pub enum CommandAction {
+    Read {
+        command: String,
+        name: String,
+        path: PathBuf,
+    },
+    ListFiles {
+        command: String,
+        path: Option<String>,
+    },
+    Search {
+        command: String,
+        query: Option<String>,
+        path: Option<String>,
+    },
+    Unknown {
+        command: String,
+    },
+}
+
+impl CommandAction {
+    pub fn into_core(self) -> CoreParsedCommand {
+        match self {
+            CommandAction::Read {
+                command: cmd,
+                name,
+                path,
+            } => CoreParsedCommand::Read { cmd, name, path },
+            CommandAction::ListFiles { command: cmd, path } => {
+                CoreParsedCommand::ListFiles { cmd, path }
+            }
+            CommandAction::Search {
+                command: cmd,
+                query,
+                path,
+            } => CoreParsedCommand::Search { cmd, query, path },
+            CommandAction::Unknown { command: cmd } => CoreParsedCommand::Unknown { cmd },
+        }
+    }
+}
+
+impl From<CoreParsedCommand> for CommandAction {
+    fn from(value: CoreParsedCommand) -> Self {
+        match value {
+            CoreParsedCommand::Read { cmd, name, path } => CommandAction::Read {
+                command: cmd,
+                name,
+                path,
+            },
+            CoreParsedCommand::ListFiles { cmd, path } => {
+                CommandAction::ListFiles { command: cmd, path }
+            }
+            CoreParsedCommand::Search { cmd, query, path } => CommandAction::Search {
+                command: cmd,
+                query,
+                path,
+            },
+            CoreParsedCommand::Unknown { cmd } => CommandAction::Unknown { command: cmd },
+        }
+    }
+}
+
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(tag = "type", rename_all = "camelCase")]
 #[ts(tag = "type")]
@@ -278,7 +392,7 @@ pub struct ThreadStartParams {
     pub cwd: Option<String>,
     pub approval_policy: Option<AskForApproval>,
     pub sandbox: Option<SandboxMode>,
-    pub config: Option<HashMap<String, serde_json::Value>>,
+    pub config: Option<HashMap<String, JsonValue>>,
     pub base_instructions: Option<String>,
     pub developer_instructions: Option<String>,
 }
@@ -290,11 +404,39 @@ 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)]
@@ -361,6 +503,8 @@ pub struct Thread {
     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)]
@@ -475,31 +619,50 @@ impl From<CoreUserInput> for UserInput {
 #[ts(tag = "type")]
 #[ts(export_to = "v2/")]
 pub enum ThreadItem {
-    UserMessage {
-        id: String,
-        content: Vec<UserInput>,
-    },
-    AgentMessage {
-        id: String,
-        text: String,
-    },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
+    UserMessage { id: String, content: Vec<UserInput> },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
+    AgentMessage { id: String, text: String },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
     Reasoning {
         id: String,
-        text: String,
+        #[serde(default)]
+        summary: Vec<String>,
+        #[serde(default)]
+        content: Vec<String>,
     },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
     CommandExecution {
         id: String,
+        /// The command to be executed.
         command: String,
-        aggregated_output: String,
-        exit_code: Option<i32>,
+        /// The command's working directory.
+        cwd: PathBuf,
         status: CommandExecutionStatus,
+        /// A best-effort parsing of the command to understand the action(s) it will perform.
+        /// This returns a list of CommandAction objects because a single shell command may
+        /// be composed of many commands piped together.
+        command_actions: Vec<CommandAction>,
+        /// The command's output, aggregated from stdout and stderr.
+        aggregated_output: Option<String>,
+        /// The command's exit code.
+        exit_code: Option<i32>,
+        /// The duration of the command execution in milliseconds.
         duration_ms: Option<i64>,
     },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
     FileChange {
         id: String,
         changes: Vec<FileUpdateChange>,
         status: PatchApplyStatus,
     },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
     McpToolCall {
         id: String,
         server: String,
@@ -509,22 +672,18 @@ pub enum ThreadItem {
         result: Option<McpToolCallResult>,
         error: Option<McpToolCallError>,
     },
-    WebSearch {
-        id: String,
-        query: String,
-    },
-    TodoList {
-        id: String,
-        items: Vec<TodoItem>,
-    },
-    ImageView {
-        id: String,
-        path: String,
-    },
-    CodeReview {
-        id: String,
-        review: String,
-    },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
+    WebSearch { id: String, query: String },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
+    TodoList { id: String, items: Vec<TodoItem> },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
+    ImageView { id: String, path: String },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
+    CodeReview { id: String, review: String },
 }
 
 impl From<CoreTurnItem> for ThreadItem {
@@ -544,17 +703,11 @@ impl From<CoreTurnItem> for ThreadItem {
                     .collect::<String>();
                 ThreadItem::AgentMessage { id: agent.id, text }
             }
-            CoreTurnItem::Reasoning(reasoning) => {
-                let text = if !reasoning.summary_text.is_empty() {
-                    reasoning.summary_text.join("\n")
-                } else {
-                    reasoning.raw_content.join("\n")
-                };
-                ThreadItem::Reasoning {
-                    id: reasoning.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,
@@ -612,7 +765,7 @@ pub enum McpToolCallStatus {
 #[ts(export_to = "v2/")]
 pub struct McpToolCallResult {
     pub content: Vec<McpContentBlock>,
-    pub structured_content: JsonValue,
+    pub structured_content: Option<JsonValue>,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
@@ -688,6 +841,32 @@ 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/")]
@@ -704,6 +883,39 @@ pub struct McpToolCallProgressNotification {
     pub message: String,
 }
 
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct CommandExecutionRequestApprovalParams {
+    pub thread_id: String,
+    pub turn_id: String,
+    pub item_id: String,
+    /// Optional explanatory reason (e.g. request for network access).
+    pub reason: Option<String>,
+    /// Optional model-provided risk assessment describing the blocked command.
+    pub risk: Option<SandboxCommandAssessment>,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct CommandExecutionRequestAcceptSettings {
+    /// If true, automatically approve this command for the duration of the session.
+    #[serde(default)]
+    pub for_session: bool,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct CommandExecutionRequestApprovalResponse {
+    pub decision: ApprovalDecision,
+    /// Optional approval settings for when the decision is `accept`.
+    /// Ignored if the decision is `decline` or `cancel`.
+    #[serde(default)]
+    pub accept_settings: Option<CommandExecutionRequestAcceptSettings>,
+}
+
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
 #[ts(export_to = "v2/")]
@@ -836,7 +1048,8 @@ mod tests {
             ThreadItem::from(reasoning_item),
             ThreadItem::Reasoning {
                 id: "reasoning-1".to_string(),
-                text: "line one\nline two".to_string(),
+                summary: vec!["line one".to_string(), "line two".to_string()],
+                content: vec![],
             }
         );
 

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

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

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

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

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

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

codex-rs/app-server/Cargo.toml

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

codex-rs/app-server/README.md

@@ -2,6 +2,16 @@
 
 `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.
 
+## Table of Contents
+- [Protocol](#protocol)
+- [Message Schema](#message-schema)
+- [Lifecycle Overview](#lifecycle-overview)
+- [Initialization](#initialization)
+- [Core primitives](#core-primitives)
+- [Thread & turn endpoints](#thread--turn-endpoints)
+- [Auth endpoints](#auth-endpoints)
+- [Events (work-in-progress)](#v2-streaming-events-work-in-progress)
+
 ## Protocol
 
 Similar to [MCP](https://modelcontextprotocol.io/), `codex app-server` supports bidirectional communication, streaming JSONL over stdio. The protocol is JSON-RPC 2.0, though the `"jsonrpc":"2.0"` header is omitted.
@@ -15,6 +25,14 @@ codex app-server generate-ts --out DIR
 codex app-server generate-json-schema --out DIR
 ```
 
+## Lifecycle Overview
+
+- Initialize once: Immediately after launching the codex app-server process, send an `initialize` request with your client metadata, then emit an `initialized` notification. Any other request before this handshake gets rejected.
+- Start (or resume) a thread: Call `thread/start` to open a fresh conversation. The response returns the thread object and you’ll also get a `thread/started` notification. If you’re continuing an existing conversation, call `thread/resume` with its ID instead.
+- Begin a turn: To send user input, call `turn/start` with the target `threadId` and the user's input. Optional fields let you override model, cwd, sandbox policy, etc. This immediately returns the new turn object and triggers a `turn/started` notification.
+- Stream events: After `turn/start`, keep reading JSON-RPC notifications on stdout. You’ll see `item/started`, `item/completed`, deltas like `item/agentMessage/delta`, tool progress, etc. These represent streaming model output plus any side effects (commands, tool calls, reasoning notes).
+- Finish the turn: When the model is done (or the turn is interrupted via making the `turn/interrupt` call), the server sends `turn/completed` with the final turn state and token usage.
+
 ## 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.
@@ -56,7 +74,7 @@ Start a fresh thread when you need a new Codex conversation.
 { "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",
+    "model": "gpt-5.1-codex",
     "cwd": "/Users/me/project",
     "approvalPolicy": "never",
     "sandbox": "workspaceWrite",
@@ -137,7 +155,7 @@ You can optionally specify config overrides on the new turn. If specified, these
         "writableRoots": ["/Users/me/project"],
         "networkAccess": true
     },
-    "model": "gpt-5-codex",
+    "model": "gpt-5.1-codex",
     "effort": "medium",
     "summary": "concise"
 } }
@@ -258,3 +276,33 @@ Field 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.
+
+
+## Events (work-in-progress)
+
+Event notifications are the server-initiated event stream for thread lifecycles, turn lifecycles, and the items within them. After you start or resume a thread, keep reading stdout for `thread/started`, `turn/*`, and `item/*` notifications.
+
+### Turn events
+
+The app-server streams JSON-RPC notifications while a turn is running. Each turn starts with `turn/started` (initial `turn`) and ends with `turn/completed` (final `turn` plus token `usage`), and clients subscribe to the events they care about, rendering each item incrementally as updates arrive. The per-item lifecycle is always: `item/started` → zero or more item-specific deltas → `item/completed`.
+
+#### Thread items
+
+`ThreadItem` is the tagged union carried in turn responses and `item/*` notifications. Currently we support events for the following items:
+- `userMessage` — `{id, content}` where `content` is a list of user inputs (`text`, `image`, or `localImage`).
+- `agentMessage` — `{id, text}` containing the accumulated agent reply.
+- `reasoning` — `{id, summary, content}` where `summary` holds streamed reasoning summaries (applicable for most OpenAI models) and `content` holds raw reasoning blocks (applicable for e.g. open source models).
+- `mcpToolCall` — `{id, server, tool, status, arguments, result?, error?}` describing MCP calls; `status` is `inProgress`, `completed`, or `failed`.
+- `webSearch` — `{id, query}` for a web search request issued by the agent.
+
+All items emit two shared lifecycle events:
+- `item/started` — emits the full `item` when a new unit of work begins so the UI can render it immediately; the `item.id` in this payload matches the `itemId` used by deltas.
+- `item/completed` — sends the final `item` once that work finishes (e.g., after a tool call or message completes); treat this as the authoritative state.
+
+There are additional item-specific events:
+#### agentMessage
+- `item/agentMessage/delta` — appends streamed text for the agent message; concatenate `delta` values for the same `itemId` in order to reconstruct the full reply.
+#### reasoning
+- `item/reasoning/summaryTextDelta` — streams readable reasoning summaries; `summaryIndex` increments when a new summary section opens.
+- `item/reasoning/summaryPartAdded` — marks the boundary between reasoning summary sections for an `itemId`; subsequent `summaryTextDelta` entries share the same `summaryIndex`.
+- `item/reasoning/textDelta` — streams raw reasoning text (only applicable for e.g. open source models); use `contentIndex` to group deltas that belong together before showing them in the UI.

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

@@ -0,0 +1,628 @@
+use crate::codex_message_processor::ApiVersion;
+use crate::codex_message_processor::PendingInterrupts;
+use crate::outgoing_message::OutgoingMessageSender;
+use codex_app_server_protocol::AccountRateLimitsUpdatedNotification;
+use codex_app_server_protocol::AgentMessageDeltaNotification;
+use codex_app_server_protocol::ApplyPatchApprovalParams;
+use codex_app_server_protocol::ApplyPatchApprovalResponse;
+use codex_app_server_protocol::ApprovalDecision;
+use codex_app_server_protocol::CommandAction as V2ParsedCommand;
+use codex_app_server_protocol::CommandExecutionOutputDeltaNotification;
+use codex_app_server_protocol::CommandExecutionRequestApprovalParams;
+use codex_app_server_protocol::CommandExecutionRequestApprovalResponse;
+use codex_app_server_protocol::CommandExecutionStatus;
+use codex_app_server_protocol::ExecCommandApprovalParams;
+use codex_app_server_protocol::ExecCommandApprovalResponse;
+use codex_app_server_protocol::InterruptConversationResponse;
+use codex_app_server_protocol::ItemCompletedNotification;
+use codex_app_server_protocol::ItemStartedNotification;
+use codex_app_server_protocol::McpToolCallError;
+use codex_app_server_protocol::McpToolCallResult;
+use codex_app_server_protocol::McpToolCallStatus;
+use codex_app_server_protocol::ReasoningSummaryPartAddedNotification;
+use codex_app_server_protocol::ReasoningSummaryTextDeltaNotification;
+use codex_app_server_protocol::ReasoningTextDeltaNotification;
+use codex_app_server_protocol::SandboxCommandAssessment as V2SandboxCommandAssessment;
+use codex_app_server_protocol::ServerNotification;
+use codex_app_server_protocol::ServerRequestPayload;
+use codex_app_server_protocol::ThreadItem;
+use codex_app_server_protocol::TurnInterruptResponse;
+use codex_core::CodexConversation;
+use codex_core::parse_command::shlex_join;
+use codex_core::protocol::ApplyPatchApprovalRequestEvent;
+use codex_core::protocol::Event;
+use codex_core::protocol::EventMsg;
+use codex_core::protocol::ExecApprovalRequestEvent;
+use codex_core::protocol::ExecCommandEndEvent;
+use codex_core::protocol::McpToolCallBeginEvent;
+use codex_core::protocol::McpToolCallEndEvent;
+use codex_core::protocol::Op;
+use codex_core::protocol::ReviewDecision;
+use codex_protocol::ConversationId;
+use std::convert::TryFrom;
+use std::sync::Arc;
+use tokio::sync::oneshot;
+use tracing::error;
+
+type JsonValue = serde_json::Value;
+
+pub(crate) async fn apply_bespoke_event_handling(
+    event: Event,
+    conversation_id: ConversationId,
+    conversation: Arc<CodexConversation>,
+    outgoing: Arc<OutgoingMessageSender>,
+    pending_interrupts: PendingInterrupts,
+    api_version: ApiVersion,
+) {
+    let Event { id: event_id, msg } = event;
+    match msg {
+        EventMsg::ApplyPatchApprovalRequest(ApplyPatchApprovalRequestEvent {
+            call_id,
+            changes,
+            reason,
+            grant_root,
+        }) => {
+            let params = ApplyPatchApprovalParams {
+                conversation_id,
+                call_id,
+                file_changes: changes,
+                reason,
+                grant_root,
+            };
+            let rx = outgoing
+                .send_request(ServerRequestPayload::ApplyPatchApproval(params))
+                .await;
+            tokio::spawn(async move {
+                on_patch_approval_response(event_id, rx, conversation).await;
+            });
+        }
+        EventMsg::ExecApprovalRequest(ExecApprovalRequestEvent {
+            call_id,
+            turn_id,
+            command,
+            cwd,
+            reason,
+            risk,
+            parsed_cmd,
+        }) => match api_version {
+            ApiVersion::V1 => {
+                let params = ExecCommandApprovalParams {
+                    conversation_id,
+                    call_id,
+                    command,
+                    cwd,
+                    reason,
+                    risk,
+                    parsed_cmd,
+                };
+                let rx = outgoing
+                    .send_request(ServerRequestPayload::ExecCommandApproval(params))
+                    .await;
+                tokio::spawn(async move {
+                    on_exec_approval_response(event_id, rx, conversation).await;
+                });
+            }
+            ApiVersion::V2 => {
+                let params = CommandExecutionRequestApprovalParams {
+                    thread_id: conversation_id.to_string(),
+                    turn_id: turn_id.clone(),
+                    // Until we migrate the core to be aware of a first class CommandExecutionItem
+                    // and emit the corresponding EventMsg, we repurpose the call_id as the item_id.
+                    item_id: call_id.clone(),
+                    reason,
+                    risk: risk.map(V2SandboxCommandAssessment::from),
+                };
+                let rx = outgoing
+                    .send_request(ServerRequestPayload::CommandExecutionRequestApproval(
+                        params,
+                    ))
+                    .await;
+                tokio::spawn(async move {
+                    on_command_execution_request_approval_response(event_id, rx, conversation)
+                        .await;
+                });
+            }
+        },
+        // TODO(celia): properly construct McpToolCall TurnItem in core.
+        EventMsg::McpToolCallBegin(begin_event) => {
+            let notification = construct_mcp_tool_call_notification(begin_event).await;
+            outgoing
+                .send_server_notification(ServerNotification::ItemStarted(notification))
+                .await;
+        }
+        EventMsg::McpToolCallEnd(end_event) => {
+            let notification = construct_mcp_tool_call_end_notification(end_event).await;
+            outgoing
+                .send_server_notification(ServerNotification::ItemCompleted(notification))
+                .await;
+        }
+        EventMsg::AgentMessageContentDelta(event) => {
+            let notification = AgentMessageDeltaNotification {
+                item_id: event.item_id,
+                delta: event.delta,
+            };
+            outgoing
+                .send_server_notification(ServerNotification::AgentMessageDelta(notification))
+                .await;
+        }
+        EventMsg::ReasoningContentDelta(event) => {
+            let notification = ReasoningSummaryTextDeltaNotification {
+                item_id: event.item_id,
+                delta: event.delta,
+                summary_index: event.summary_index,
+            };
+            outgoing
+                .send_server_notification(ServerNotification::ReasoningSummaryTextDelta(
+                    notification,
+                ))
+                .await;
+        }
+        EventMsg::ReasoningRawContentDelta(event) => {
+            let notification = ReasoningTextDeltaNotification {
+                item_id: event.item_id,
+                delta: event.delta,
+                content_index: event.content_index,
+            };
+            outgoing
+                .send_server_notification(ServerNotification::ReasoningTextDelta(notification))
+                .await;
+        }
+        EventMsg::AgentReasoningSectionBreak(event) => {
+            let notification = ReasoningSummaryPartAddedNotification {
+                item_id: event.item_id,
+                summary_index: event.summary_index,
+            };
+            outgoing
+                .send_server_notification(ServerNotification::ReasoningSummaryPartAdded(
+                    notification,
+                ))
+                .await;
+        }
+        EventMsg::TokenCount(token_count_event) => {
+            if let Some(rate_limits) = token_count_event.rate_limits {
+                outgoing
+                    .send_server_notification(ServerNotification::AccountRateLimitsUpdated(
+                        AccountRateLimitsUpdatedNotification {
+                            rate_limits: rate_limits.into(),
+                        },
+                    ))
+                    .await;
+            }
+        }
+        EventMsg::ItemStarted(item_started_event) => {
+            let item: ThreadItem = item_started_event.item.clone().into();
+            let notification = ItemStartedNotification { item };
+            outgoing
+                .send_server_notification(ServerNotification::ItemStarted(notification))
+                .await;
+        }
+        EventMsg::ItemCompleted(item_completed_event) => {
+            let item: ThreadItem = item_completed_event.item.clone().into();
+            let notification = ItemCompletedNotification { item };
+            outgoing
+                .send_server_notification(ServerNotification::ItemCompleted(notification))
+                .await;
+        }
+        EventMsg::ExecCommandBegin(exec_command_begin_event) => {
+            let item = ThreadItem::CommandExecution {
+                id: exec_command_begin_event.call_id.clone(),
+                command: shlex_join(&exec_command_begin_event.command),
+                cwd: exec_command_begin_event.cwd,
+                status: CommandExecutionStatus::InProgress,
+                command_actions: exec_command_begin_event
+                    .parsed_cmd
+                    .into_iter()
+                    .map(V2ParsedCommand::from)
+                    .collect(),
+                aggregated_output: None,
+                exit_code: None,
+                duration_ms: None,
+            };
+            let notification = ItemStartedNotification { item };
+            outgoing
+                .send_server_notification(ServerNotification::ItemStarted(notification))
+                .await;
+        }
+        EventMsg::ExecCommandOutputDelta(exec_command_output_delta_event) => {
+            let notification = CommandExecutionOutputDeltaNotification {
+                item_id: exec_command_output_delta_event.call_id.clone(),
+                delta: String::from_utf8_lossy(&exec_command_output_delta_event.chunk).to_string(),
+            };
+            outgoing
+                .send_server_notification(ServerNotification::CommandExecutionOutputDelta(
+                    notification,
+                ))
+                .await;
+        }
+        EventMsg::ExecCommandEnd(exec_command_end_event) => {
+            let ExecCommandEndEvent {
+                call_id,
+                command,
+                cwd,
+                parsed_cmd,
+                aggregated_output,
+                exit_code,
+                duration,
+                ..
+            } = exec_command_end_event;
+
+            let status = if exit_code == 0 {
+                CommandExecutionStatus::Completed
+            } else {
+                CommandExecutionStatus::Failed
+            };
+
+            let aggregated_output = if aggregated_output.is_empty() {
+                None
+            } else {
+                Some(aggregated_output)
+            };
+
+            let duration_ms = i64::try_from(duration.as_millis()).unwrap_or(i64::MAX);
+
+            let item = ThreadItem::CommandExecution {
+                id: call_id,
+                command: shlex_join(&command),
+                cwd,
+                status,
+                command_actions: parsed_cmd.into_iter().map(V2ParsedCommand::from).collect(),
+                aggregated_output,
+                exit_code: Some(exit_code),
+                duration_ms: Some(duration_ms),
+            };
+
+            let notification = ItemCompletedNotification { item };
+            outgoing
+                .send_server_notification(ServerNotification::ItemCompleted(notification))
+                .await;
+        }
+        // If this is a TurnAborted, reply to any pending interrupt requests.
+        EventMsg::TurnAborted(turn_aborted_event) => {
+            let pending = {
+                let mut map = pending_interrupts.lock().await;
+                map.remove(&conversation_id).unwrap_or_default()
+            };
+            if !pending.is_empty() {
+                for (rid, ver) in pending {
+                    match ver {
+                        ApiVersion::V1 => {
+                            let response = InterruptConversationResponse {
+                                abort_reason: turn_aborted_event.reason.clone(),
+                            };
+                            outgoing.send_response(rid, response).await;
+                        }
+                        ApiVersion::V2 => {
+                            let response = TurnInterruptResponse {};
+                            outgoing.send_response(rid, response).await;
+                        }
+                    }
+                }
+            }
+        }
+
+        _ => {}
+    }
+}
+
+async fn on_patch_approval_response(
+    event_id: String,
+    receiver: oneshot::Receiver<JsonValue>,
+    codex: Arc<CodexConversation>,
+) {
+    let response = receiver.await;
+    let value = match response {
+        Ok(value) => value,
+        Err(err) => {
+            error!("request failed: {err:?}");
+            if let Err(submit_err) = codex
+                .submit(Op::PatchApproval {
+                    id: event_id.clone(),
+                    decision: ReviewDecision::Denied,
+                })
+                .await
+            {
+                error!("failed to submit denied PatchApproval after request failure: {submit_err}");
+            }
+            return;
+        }
+    };
+
+    let response =
+        serde_json::from_value::<ApplyPatchApprovalResponse>(value).unwrap_or_else(|err| {
+            error!("failed to deserialize ApplyPatchApprovalResponse: {err}");
+            ApplyPatchApprovalResponse {
+                decision: ReviewDecision::Denied,
+            }
+        });
+
+    if let Err(err) = codex
+        .submit(Op::PatchApproval {
+            id: event_id,
+            decision: response.decision,
+        })
+        .await
+    {
+        error!("failed to submit PatchApproval: {err}");
+    }
+}
+
+async fn on_exec_approval_response(
+    event_id: String,
+    receiver: oneshot::Receiver<JsonValue>,
+    conversation: Arc<CodexConversation>,
+) {
+    let response = receiver.await;
+    let value = match response {
+        Ok(value) => value,
+        Err(err) => {
+            error!("request failed: {err:?}");
+            return;
+        }
+    };
+
+    // Try to deserialize `value` and then make the appropriate call to `codex`.
+    let response =
+        serde_json::from_value::<ExecCommandApprovalResponse>(value).unwrap_or_else(|err| {
+            error!("failed to deserialize ExecCommandApprovalResponse: {err}");
+            // If we cannot deserialize the response, we deny the request to be
+            // conservative.
+            ExecCommandApprovalResponse {
+                decision: ReviewDecision::Denied,
+            }
+        });
+
+    if let Err(err) = conversation
+        .submit(Op::ExecApproval {
+            id: event_id,
+            decision: response.decision,
+        })
+        .await
+    {
+        error!("failed to submit ExecApproval: {err}");
+    }
+}
+
+async fn on_command_execution_request_approval_response(
+    event_id: String,
+    receiver: oneshot::Receiver<JsonValue>,
+    conversation: Arc<CodexConversation>,
+) {
+    let response = receiver.await;
+    let value = match response {
+        Ok(value) => value,
+        Err(err) => {
+            error!("request failed: {err:?}");
+            return;
+        }
+    };
+
+    let response = serde_json::from_value::<CommandExecutionRequestApprovalResponse>(value)
+        .unwrap_or_else(|err| {
+            error!("failed to deserialize CommandExecutionRequestApprovalResponse: {err}");
+            CommandExecutionRequestApprovalResponse {
+                decision: ApprovalDecision::Decline,
+                accept_settings: None,
+            }
+        });
+
+    let CommandExecutionRequestApprovalResponse {
+        decision,
+        accept_settings,
+    } = response;
+
+    let decision = match (decision, accept_settings) {
+        (ApprovalDecision::Accept, Some(settings)) if settings.for_session => {
+            ReviewDecision::ApprovedForSession
+        }
+        (ApprovalDecision::Accept, _) => ReviewDecision::Approved,
+        (ApprovalDecision::Decline, _) => ReviewDecision::Denied,
+        (ApprovalDecision::Cancel, _) => ReviewDecision::Abort,
+    };
+    if let Err(err) = conversation
+        .submit(Op::ExecApproval {
+            id: event_id,
+            decision,
+        })
+        .await
+    {
+        error!("failed to submit ExecApproval: {err}");
+    }
+}
+
+/// similar to handle_mcp_tool_call_begin in exec
+async fn construct_mcp_tool_call_notification(
+    begin_event: McpToolCallBeginEvent,
+) -> ItemStartedNotification {
+    let item = ThreadItem::McpToolCall {
+        id: begin_event.call_id,
+        server: begin_event.invocation.server,
+        tool: begin_event.invocation.tool,
+        status: McpToolCallStatus::InProgress,
+        arguments: begin_event.invocation.arguments.unwrap_or(JsonValue::Null),
+        result: None,
+        error: None,
+    };
+    ItemStartedNotification { item }
+}
+
+/// simiilar to handle_mcp_tool_call_end in exec
+async fn construct_mcp_tool_call_end_notification(
+    end_event: McpToolCallEndEvent,
+) -> ItemCompletedNotification {
+    let status = if end_event.is_success() {
+        McpToolCallStatus::Completed
+    } else {
+        McpToolCallStatus::Failed
+    };
+
+    let (result, error) = match &end_event.result {
+        Ok(value) => (
+            Some(McpToolCallResult {
+                content: value.content.clone(),
+                structured_content: value.structured_content.clone(),
+            }),
+            None,
+        ),
+        Err(message) => (
+            None,
+            Some(McpToolCallError {
+                message: message.clone(),
+            }),
+        ),
+    };
+
+    let item = ThreadItem::McpToolCall {
+        id: end_event.call_id,
+        server: end_event.invocation.server,
+        tool: end_event.invocation.tool,
+        status,
+        arguments: end_event.invocation.arguments.unwrap_or(JsonValue::Null),
+        result,
+        error,
+    };
+    ItemCompletedNotification { item }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use codex_core::protocol::McpInvocation;
+    use mcp_types::CallToolResult;
+    use mcp_types::ContentBlock;
+    use mcp_types::TextContent;
+    use pretty_assertions::assert_eq;
+    use serde_json::Value as JsonValue;
+    use std::time::Duration;
+
+    #[tokio::test]
+    async fn test_construct_mcp_tool_call_begin_notification_with_args() {
+        let begin_event = McpToolCallBeginEvent {
+            call_id: "call_123".to_string(),
+            invocation: McpInvocation {
+                server: "codex".to_string(),
+                tool: "list_mcp_resources".to_string(),
+                arguments: Some(serde_json::json!({"server": ""})),
+            },
+        };
+
+        let notification = construct_mcp_tool_call_notification(begin_event.clone()).await;
+
+        let expected = ItemStartedNotification {
+            item: ThreadItem::McpToolCall {
+                id: begin_event.call_id,
+                server: begin_event.invocation.server,
+                tool: begin_event.invocation.tool,
+                status: McpToolCallStatus::InProgress,
+                arguments: serde_json::json!({"server": ""}),
+                result: None,
+                error: None,
+            },
+        };
+
+        assert_eq!(notification, expected);
+    }
+
+    #[tokio::test]
+    async fn test_construct_mcp_tool_call_begin_notification_without_args() {
+        let begin_event = McpToolCallBeginEvent {
+            call_id: "call_456".to_string(),
+            invocation: McpInvocation {
+                server: "codex".to_string(),
+                tool: "list_mcp_resources".to_string(),
+                arguments: None,
+            },
+        };
+
+        let notification = construct_mcp_tool_call_notification(begin_event.clone()).await;
+
+        let expected = ItemStartedNotification {
+            item: ThreadItem::McpToolCall {
+                id: begin_event.call_id,
+                server: begin_event.invocation.server,
+                tool: begin_event.invocation.tool,
+                status: McpToolCallStatus::InProgress,
+                arguments: JsonValue::Null,
+                result: None,
+                error: None,
+            },
+        };
+
+        assert_eq!(notification, expected);
+    }
+
+    #[tokio::test]
+    async fn test_construct_mcp_tool_call_end_notification_success() {
+        let content = vec![ContentBlock::TextContent(TextContent {
+            annotations: None,
+            text: "{\"resources\":[]}".to_string(),
+            r#type: "text".to_string(),
+        })];
+        let result = CallToolResult {
+            content: content.clone(),
+            is_error: Some(false),
+            structured_content: None,
+        };
+
+        let end_event = McpToolCallEndEvent {
+            call_id: "call_789".to_string(),
+            invocation: McpInvocation {
+                server: "codex".to_string(),
+                tool: "list_mcp_resources".to_string(),
+                arguments: Some(serde_json::json!({"server": ""})),
+            },
+            duration: Duration::from_nanos(92708),
+            result: Ok(result),
+        };
+
+        let notification = construct_mcp_tool_call_end_notification(end_event.clone()).await;
+
+        let expected = ItemCompletedNotification {
+            item: ThreadItem::McpToolCall {
+                id: end_event.call_id,
+                server: end_event.invocation.server,
+                tool: end_event.invocation.tool,
+                status: McpToolCallStatus::Completed,
+                arguments: serde_json::json!({"server": ""}),
+                result: Some(McpToolCallResult {
+                    content,
+                    structured_content: None,
+                }),
+                error: None,
+            },
+        };
+
+        assert_eq!(notification, expected);
+    }
+
+    #[tokio::test]
+    async fn test_construct_mcp_tool_call_end_notification_error() {
+        let end_event = McpToolCallEndEvent {
+            call_id: "call_err".to_string(),
+            invocation: McpInvocation {
+                server: "codex".to_string(),
+                tool: "list_mcp_resources".to_string(),
+                arguments: None,
+            },
+            duration: Duration::from_millis(1),
+            result: Err("boom".to_string()),
+        };
+
+        let notification = construct_mcp_tool_call_end_notification(end_event.clone()).await;
+
+        let expected = ItemCompletedNotification {
+            item: ThreadItem::McpToolCall {
+                id: end_event.call_id,
+                server: end_event.invocation.server,
+                tool: end_event.invocation.tool,
+                status: McpToolCallStatus::Failed,
+                arguments: JsonValue::Null,
+                result: None,
+                error: Some(McpToolCallError {
+                    message: "boom".to_string(),
+                }),
+            },
+        };
+
+        assert_eq!(notification, expected);
+    }
+}

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

@@ -1,3 +1,4 @@
+use crate::bespoke_event_handling::apply_bespoke_event_handling;
 use crate::error_code::INTERNAL_ERROR_CODE;
 use crate::error_code::INVALID_REQUEST_ERROR_CODE;
 use crate::fuzzy_file_search::run_fuzzy_file_search;
@@ -8,12 +9,9 @@ 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::ApplyPatchApprovalParams;
-use codex_app_server_protocol::ApplyPatchApprovalResponse;
 use codex_app_server_protocol::ArchiveConversationParams;
 use codex_app_server_protocol::ArchiveConversationResponse;
 use codex_app_server_protocol::AskForApproval;
@@ -25,8 +23,6 @@ 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;
 use codex_app_server_protocol::ExecOneOffCommandParams;
 use codex_app_server_protocol::ExecOneOffCommandResponse;
 use codex_app_server_protocol::FeedbackUploadParams;
@@ -45,9 +41,6 @@ use codex_app_server_protocol::GetUserSavedConfigResponse;
 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;
@@ -65,7 +58,6 @@ use codex_app_server_protocol::NewConversationResponse;
 use codex_app_server_protocol::RemoveConversationListenerParams;
 use codex_app_server_protocol::RemoveConversationSubscriptionResponse;
 use codex_app_server_protocol::RequestId;
-use codex_app_server_protocol::Result as JsonRpcResult;
 use codex_app_server_protocol::ResumeConversationParams;
 use codex_app_server_protocol::ResumeConversationResponse;
 use codex_app_server_protocol::SandboxMode;
@@ -74,7 +66,6 @@ use codex_app_server_protocol::SendUserMessageResponse;
 use codex_app_server_protocol::SendUserTurnParams;
 use codex_app_server_protocol::SendUserTurnResponse;
 use codex_app_server_protocol::ServerNotification;
-use codex_app_server_protocol::ServerRequestPayload;
 use codex_app_server_protocol::SessionConfiguredNotification;
 use codex_app_server_protocol::SetDefaultModelParams;
 use codex_app_server_protocol::SetDefaultModelResponse;
@@ -91,7 +82,6 @@ use codex_app_server_protocol::ThreadStartResponse;
 use codex_app_server_protocol::ThreadStartedNotification;
 use codex_app_server_protocol::Turn;
 use codex_app_server_protocol::TurnInterruptParams;
-use codex_app_server_protocol::TurnInterruptResponse;
 use codex_app_server_protocol::TurnStartParams;
 use codex_app_server_protocol::TurnStartResponse;
 use codex_app_server_protocol::TurnStartedNotification;
@@ -123,12 +113,8 @@ use codex_core::find_conversation_path_by_id_str;
 use codex_core::get_platform_sandbox;
 use codex_core::git_info::git_diff_to_remote;
 use codex_core::parse_cursor;
-use codex_core::protocol::ApplyPatchApprovalRequestEvent;
-use codex_core::protocol::Event;
 use codex_core::protocol::EventMsg;
-use codex_core::protocol::ExecApprovalRequestEvent;
 use codex_core::protocol::Op;
-use codex_core::protocol::ReviewDecision;
 use codex_core::read_head_for_summary;
 use codex_feedback::CodexFeedback;
 use codex_login::ServerOptions as LoginServerOptions;
@@ -163,7 +149,7 @@ use tracing::warn;
 use uuid::Uuid;
 
 type PendingInterruptQueue = Vec<(RequestId, ApiVersion)>;
-type PendingInterrupts = Arc<Mutex<HashMap<ConversationId, PendingInterruptQueue>>>;
+pub(crate) type PendingInterrupts = Arc<Mutex<HashMap<ConversationId, PendingInterruptQueue>>>;
 
 // Duration before a ChatGPT login attempt is abandoned.
 const LOGIN_CHATGPT_TIMEOUT: Duration = Duration::from_secs(10 * 60);
@@ -172,8 +158,8 @@ struct ActiveLogin {
     login_id: Uuid,
 }
 
-impl ActiveLogin {
-    fn drop(&self) {
+impl Drop for ActiveLogin {
+    fn drop(&mut self) {
         self.shutdown_handle.shutdown();
     }
 }
@@ -194,7 +180,7 @@ pub(crate) struct CodexMessageProcessor {
 }
 
 #[derive(Clone, Copy, Debug)]
-enum ApiVersion {
+pub(crate) enum ApiVersion {
     V1,
     V2,
 }
@@ -431,7 +417,7 @@ impl CodexMessageProcessor {
         {
             let mut guard = self.active_login.lock().await;
             if let Some(active) = guard.take() {
-                active.drop();
+                drop(active);
             }
         }
 
@@ -539,7 +525,7 @@ impl CodexMessageProcessor {
                     {
                         let mut guard = self.active_login.lock().await;
                         if let Some(existing) = guard.take() {
-                            existing.drop();
+                            drop(existing);
                         }
                         *guard = Some(ActiveLogin {
                             shutdown_handle: shutdown_handle.clone(),
@@ -629,7 +615,7 @@ impl CodexMessageProcessor {
                     {
                         let mut guard = self.active_login.lock().await;
                         if let Some(existing) = guard.take() {
-                            existing.drop();
+                            drop(existing);
                         }
                         *guard = Some(ActiveLogin {
                             shutdown_handle: shutdown_handle.clone(),
@@ -718,7 +704,7 @@ impl CodexMessageProcessor {
         let mut guard = self.active_login.lock().await;
         if guard.as_ref().map(|l| l.login_id) == Some(login_id) {
             if let Some(active) = guard.take() {
-                active.drop();
+                drop(active);
             }
             Ok(())
         } else {
@@ -772,7 +758,7 @@ impl CodexMessageProcessor {
         {
             let mut guard = self.active_login.lock().await;
             if let Some(active) = guard.take() {
-                active.drop();
+                drop(active);
             }
         }
 
@@ -1201,21 +1187,17 @@ 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 {
@@ -1244,16 +1226,15 @@ impl CodexMessageProcessor {
                 {
                     Ok(summary) => summary_to_thread(summary),
                     Err(err) => {
-                        warn!(
-                            "failed to load summary for new thread {}: {}",
-                            conversation_id, err
-                        );
-                        Thread {
-                            id: conversation_id.to_string(),
-                            preview: String::new(),
-                            model_provider: self.config.model_provider_id.clone(),
-                            created_at: chrono::Utc::now().timestamp(),
-                        }
+                        self.send_internal_error(
+                            request_id,
+                            format!(
+                                "failed to load rollout `{}` for conversation {conversation_id}: {err}",
+                                rollout_path.display()
+                            ),
+                        )
+                        .await;
+                        return;
                     }
                 };
 
@@ -1264,7 +1245,7 @@ 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(conversation_id, false)
+                    .attach_conversation_listener(conversation_id, false, ApiVersion::V2)
                     .await
                 {
                     tracing::warn!(
@@ -1292,6 +1273,31 @@ 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,
@@ -1374,91 +1380,150 @@ impl CodexMessageProcessor {
     }
 
     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(_) => {
-                let thread = summary_to_thread(summary);
-
+            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)
+                    .attach_conversation_listener(conversation_id, false, ApiVersion::V2)
                     .await
                 {
                     tracing::warn!(
@@ -1468,6 +1533,25 @@ impl CodexMessageProcessor {
                     );
                 }
 
+                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;
             }
@@ -1878,6 +1962,15 @@ 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,
@@ -2283,7 +2376,7 @@ impl CodexMessageProcessor {
             experimental_raw_events,
         } = params;
         match self
-            .attach_conversation_listener(conversation_id, experimental_raw_events)
+            .attach_conversation_listener(conversation_id, experimental_raw_events, ApiVersion::V1)
             .await
         {
             Ok(subscription_id) => {
@@ -2324,6 +2417,7 @@ impl CodexMessageProcessor {
         &mut self,
         conversation_id: ConversationId,
         experimental_raw_events: bool,
+        api_version: ApiVersion,
     ) -> Result<Uuid, JSONRPCErrorError> {
         let conversation = match self
             .conversation_manager
@@ -2347,6 +2441,7 @@ impl CodexMessageProcessor {
 
         let outgoing_for_task = self.outgoing.clone();
         let pending_interrupts = self.pending_interrupts.clone();
+        let api_version_for_task = api_version;
         tokio::spawn(async move {
             loop {
                 tokio::select! {
@@ -2402,6 +2497,7 @@ impl CodexMessageProcessor {
                             conversation.clone(),
                             outgoing_for_task.clone(),
                             pending_interrupts.clone(),
+                            api_version_for_task,
                         )
                         .await;
                     }
@@ -2544,115 +2640,6 @@ impl CodexMessageProcessor {
     }
 }
 
-async fn apply_bespoke_event_handling(
-    event: Event,
-    conversation_id: ConversationId,
-    conversation: Arc<CodexConversation>,
-    outgoing: Arc<OutgoingMessageSender>,
-    pending_interrupts: PendingInterrupts,
-) {
-    let Event { id: event_id, msg } = event;
-    match msg {
-        EventMsg::ApplyPatchApprovalRequest(ApplyPatchApprovalRequestEvent {
-            call_id,
-            changes,
-            reason,
-            grant_root,
-        }) => {
-            let params = ApplyPatchApprovalParams {
-                conversation_id,
-                call_id,
-                file_changes: changes,
-                reason,
-                grant_root,
-            };
-            let rx = outgoing
-                .send_request(ServerRequestPayload::ApplyPatchApproval(params))
-                .await;
-            // TODO(mbolin): Enforce a timeout so this task does not live indefinitely?
-            tokio::spawn(async move {
-                on_patch_approval_response(event_id, rx, conversation).await;
-            });
-        }
-        EventMsg::ExecApprovalRequest(ExecApprovalRequestEvent {
-            call_id,
-            command,
-            cwd,
-            reason,
-            risk,
-            parsed_cmd,
-        }) => {
-            let params = ExecCommandApprovalParams {
-                conversation_id,
-                call_id,
-                command,
-                cwd,
-                reason,
-                risk,
-                parsed_cmd,
-            };
-            let rx = outgoing
-                .send_request(ServerRequestPayload::ExecCommandApproval(params))
-                .await;
-
-            // TODO(mbolin): Enforce a timeout so this task does not live indefinitely?
-            tokio::spawn(async move {
-                on_exec_approval_response(event_id, rx, conversation).await;
-            });
-        }
-        EventMsg::TokenCount(token_count_event) => {
-            if let Some(rate_limits) = token_count_event.rate_limits {
-                outgoing
-                    .send_server_notification(ServerNotification::AccountRateLimitsUpdated(
-                        AccountRateLimitsUpdatedNotification {
-                            rate_limits: rate_limits.into(),
-                        },
-                    ))
-                    .await;
-            }
-        }
-        EventMsg::ItemStarted(item_started_event) => {
-            let item: ThreadItem = item_started_event.item.clone().into();
-            let notification = ItemStartedNotification { item };
-            outgoing
-                .send_server_notification(ServerNotification::ItemStarted(notification))
-                .await;
-        }
-        EventMsg::ItemCompleted(item_completed_event) => {
-            let item: ThreadItem = item_completed_event.item.clone().into();
-            let notification = ItemCompletedNotification { item };
-            outgoing
-                .send_server_notification(ServerNotification::ItemCompleted(notification))
-                .await;
-        }
-        // If this is a TurnAborted, reply to any pending interrupt requests.
-        EventMsg::TurnAborted(turn_aborted_event) => {
-            let pending = {
-                let mut map = pending_interrupts.lock().await;
-                map.remove(&conversation_id).unwrap_or_default()
-            };
-            if !pending.is_empty() {
-                for (rid, ver) in pending {
-                    match ver {
-                        ApiVersion::V1 => {
-                            let response = InterruptConversationResponse {
-                                abort_reason: turn_aborted_event.reason.clone(),
-                            };
-                            outgoing.send_response(rid, response).await;
-                        }
-                        ApiVersion::V2 => {
-                            let response = TurnInterruptResponse {};
-                            outgoing.send_response(rid, response).await;
-                        }
-                    }
-                }
-            }
-        }
-
-        _ => {}
-    }
-}
-
 async fn derive_config_from_params(
     overrides: ConfigOverrides,
     cli_overrides: Option<std::collections::HashMap<String, serde_json::Value>>,
@@ -2666,84 +2653,6 @@ async fn derive_config_from_params(
     Config::load_with_cli_overrides(cli_overrides, overrides).await
 }
 
-async fn on_patch_approval_response(
-    event_id: String,
-    receiver: oneshot::Receiver<JsonRpcResult>,
-    codex: Arc<CodexConversation>,
-) {
-    let response = receiver.await;
-    let value = match response {
-        Ok(value) => value,
-        Err(err) => {
-            error!("request failed: {err:?}");
-            if let Err(submit_err) = codex
-                .submit(Op::PatchApproval {
-                    id: event_id.clone(),
-                    decision: ReviewDecision::Denied,
-                })
-                .await
-            {
-                error!("failed to submit denied PatchApproval after request failure: {submit_err}");
-            }
-            return;
-        }
-    };
-
-    let response =
-        serde_json::from_value::<ApplyPatchApprovalResponse>(value).unwrap_or_else(|err| {
-            error!("failed to deserialize ApplyPatchApprovalResponse: {err}");
-            ApplyPatchApprovalResponse {
-                decision: ReviewDecision::Denied,
-            }
-        });
-
-    if let Err(err) = codex
-        .submit(Op::PatchApproval {
-            id: event_id,
-            decision: response.decision,
-        })
-        .await
-    {
-        error!("failed to submit PatchApproval: {err}");
-    }
-}
-
-async fn on_exec_approval_response(
-    event_id: String,
-    receiver: oneshot::Receiver<JsonRpcResult>,
-    conversation: Arc<CodexConversation>,
-) {
-    let response = receiver.await;
-    let value = match response {
-        Ok(value) => value,
-        Err(err) => {
-            error!("request failed: {err:?}");
-            return;
-        }
-    };
-
-    // Try to deserialize `value` and then make the appropriate call to `codex`.
-    let response =
-        serde_json::from_value::<ExecCommandApprovalResponse>(value).unwrap_or_else(|err| {
-            error!("failed to deserialize ExecCommandApprovalResponse: {err}");
-            // If we cannot deserialize the response, we deny the request to be
-            // conservative.
-            ExecCommandApprovalResponse {
-                decision: ReviewDecision::Denied,
-            }
-        });
-
-    if let Err(err) = conversation
-        .submit(Op::ExecApproval {
-            id: event_id,
-            decision: response.decision,
-        })
-        .await
-    {
-        error!("failed to submit ExecApproval: {err}");
-    }
-}
-
 async fn read_summary_from_rollout(
     path: &Path,
     fallback_provider: &str,
@@ -2867,6 +2776,7 @@ fn parse_datetime(timestamp: Option<&str>) -> Option<DateTime<Utc>> {
 fn summary_to_thread(summary: ConversationSummary) -> Thread {
     let ConversationSummary {
         conversation_id,
+        path,
         preview,
         timestamp,
         model_provider,
@@ -2880,6 +2790,7 @@ fn summary_to_thread(summary: ConversationSummary) -> Thread {
         preview,
         model_provider,
         created_at: created_at.map(|dt| dt.timestamp()).unwrap_or(0),
+        path,
     }
 }
 

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

@@ -19,6 +19,10 @@ pub(crate) async fn run_fuzzy_file_search(
     roots: Vec<String>,
     cancellation_flag: Arc<AtomicBool>,
 ) -> Vec<FuzzyFileSearchResult> {
+    if roots.is_empty() {
+        return Vec::new();
+    }
+
     #[expect(clippy::expect_used)]
     let limit_per_root =
         NonZero::new(LIMIT_PER_ROOT).expect("LIMIT_PER_ROOT should be a valid non-zero usize");

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

@@ -28,6 +28,7 @@ use tracing_subscriber::filter::Targets;
 use tracing_subscriber::layer::SubscriberExt;
 use tracing_subscriber::util::SubscriberInitExt;
 
+mod bespoke_event_handling;
 mod codex_message_processor;
 mod error_code;
 mod fuzzy_file_search;
@@ -46,7 +47,7 @@ pub async fn run_main(
 ) -> IoResult<()> {
     // Set up channels.
     let (incoming_tx, mut incoming_rx) = mpsc::channel::<JSONRPCMessage>(CHANNEL_CAPACITY);
-    let (outgoing_tx, mut outgoing_rx) = mpsc::unbounded_channel::<OutgoingMessage>();
+    let (outgoing_tx, mut outgoing_rx) = mpsc::channel::<OutgoingMessage>(CHANNEL_CAPACITY);
 
     // Task: read from stdin, push to `incoming_tx`.
     let stdin_reader_handle = tokio::spawn({

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

@@ -19,12 +19,12 @@ use crate::error_code::INTERNAL_ERROR_CODE;
 /// Sends messages to the client and manages request callbacks.
 pub(crate) struct OutgoingMessageSender {
     next_request_id: AtomicI64,
-    sender: mpsc::UnboundedSender<OutgoingMessage>,
+    sender: mpsc::Sender<OutgoingMessage>,
     request_id_to_callback: Mutex<HashMap<RequestId, oneshot::Sender<Result>>>,
 }
 
 impl OutgoingMessageSender {
-    pub(crate) fn new(sender: mpsc::UnboundedSender<OutgoingMessage>) -> Self {
+    pub(crate) fn new(sender: mpsc::Sender<OutgoingMessage>) -> Self {
         Self {
             next_request_id: AtomicI64::new(0),
             sender,
@@ -45,8 +45,12 @@ impl OutgoingMessageSender {
         }
 
         let outgoing_message =
-            OutgoingMessage::Request(request.request_with_id(outgoing_message_id));
-        let _ = self.sender.send(outgoing_message);
+            OutgoingMessage::Request(request.request_with_id(outgoing_message_id.clone()));
+        if let Err(err) = self.sender.send(outgoing_message).await {
+            warn!("failed to send request {outgoing_message_id:?} to client: {err:?}");
+            let mut request_id_to_callback = self.request_id_to_callback.lock().await;
+            request_id_to_callback.remove(&outgoing_message_id);
+        }
         rx_approve
     }
 
@@ -72,7 +76,9 @@ impl OutgoingMessageSender {
         match serde_json::to_value(response) {
             Ok(result) => {
                 let outgoing_message = OutgoingMessage::Response(OutgoingResponse { id, result });
-                let _ = self.sender.send(outgoing_message);
+                if let Err(err) = self.sender.send(outgoing_message).await {
+                    warn!("failed to send response to client: {err:?}");
+                }
             }
             Err(err) => {
                 self.send_error(
@@ -89,21 +95,29 @@ impl OutgoingMessageSender {
     }
 
     pub(crate) async fn send_server_notification(&self, notification: ServerNotification) {
-        let _ = self
+        if let Err(err) = self
             .sender
-            .send(OutgoingMessage::AppServerNotification(notification));
+            .send(OutgoingMessage::AppServerNotification(notification))
+            .await
+        {
+            warn!("failed to send server notification to client: {err:?}");
+        }
     }
 
     /// All notifications should be migrated to [`ServerNotification`] and
     /// [`OutgoingMessage::Notification`] should be removed.
     pub(crate) async fn send_notification(&self, notification: OutgoingNotification) {
         let outgoing_message = OutgoingMessage::Notification(notification);
-        let _ = self.sender.send(outgoing_message);
+        if let Err(err) = self.sender.send(outgoing_message).await {
+            warn!("failed to send notification to client: {err:?}");
+        }
     }
 
     pub(crate) async fn send_error(&self, id: RequestId, error: JSONRPCErrorError) {
         let outgoing_message = OutgoingMessage::Error(OutgoingError { id, error });
-        let _ = self.sender.send(outgoing_message);
+        if let Err(err) = self.sender.send(outgoing_message).await {
+            warn!("failed to send error to client: {err:?}");
+        }
     }
 }
 

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

@@ -27,7 +27,7 @@ fn create_config_toml(codex_home: &Path) -> std::io::Result<()> {
     std::fs::write(
         config_toml,
         r#"
-model = "gpt-5-codex"
+model = "gpt-5.1-codex"
 approval_policy = "on-request"
 sandbox_mode = "workspace-write"
 model_reasoning_summary = "detailed"
@@ -87,7 +87,7 @@ async fn get_config_toml_parses_all_fields() -> Result<()> {
             }),
             forced_chatgpt_workspace_id: Some("12345678-0000-0000-0000-000000000000".into()),
             forced_login_method: Some(ForcedLoginMethod::Chatgpt),
-            model: Some("gpt-5-codex".into()),
+            model: Some("gpt-5.1-codex".into()),
             model_reasoning_effort: Some(ReasoningEffort::High),
             model_reasoning_summary: Some(ReasoningSummary::Detailed),
             model_verbosity: Some(Verbosity::Medium),

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

@@ -57,7 +57,7 @@ fn create_config_toml(codex_home: &Path) -> std::io::Result<()> {
     std::fs::write(
         config_toml,
         r#"
-model = "gpt-5-codex"
+model = "gpt-5.1-codex"
 model_reasoning_effort = "medium"
 "#,
     )

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

@@ -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 \
@@ -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
@@ -154,8 +169,30 @@ 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(())
 }
 

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);
 
@@ -150,46 +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??;

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())?;
@@ -25,7 +27,7 @@ async fn thread_resume_returns_existing_thread() -> Result<()> {
     // Start a thread.
     let start_id = mcp
         .send_thread_start_request(ThreadStartParams {
-            model: Some("gpt-5-codex".to_string()),
+            model: Some("gpt-5.1-codex".to_string()),
             ..Default::default()
         })
         .await?;
@@ -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(
@@ -54,6 +57,105 @@ async fn thread_resume_returns_existing_thread() -> Result<()> {
     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-5.1-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-5.1-codex".to_string()),
+            ..Default::default()
+        })
+        .await?;
+    let start_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
+    )
+    .await??;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(start_resp)?;
+
+    let history_text = "Hello from history";
+    let history = vec![ResponseItem::Message {
+        id: None,
+        role: "user".to_string(),
+        content: vec![ContentItem::InputText {
+            text: history_text.to_string(),
+        }],
+    }];
+
+    // Resume with explicit history and override the model.
+    let resume_id = mcp
+        .send_thread_resume_request(ThreadResumeParams {
+            thread_id: thread.id,
+            history: Some(history),
+            model: Some("mock-model".to_string()),
+            model_provider: Some("mock_provider".to_string()),
+            ..Default::default()
+        })
+        .await?;
+    let resume_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(resume_id)),
+    )
+    .await??;
+    let ThreadResumeResponse { thread: resumed } =
+        to_response::<ThreadResumeResponse>(resume_resp)?;
+    assert!(!resumed.id.is_empty());
+    assert_eq!(resumed.model_provider, "mock_provider");
+    assert_eq!(resumed.preview, history_text);
+
+    Ok(())
+}
+
 // Helper to create a config.toml pointing at the mock model server.
 fn create_config_toml(codex_home: &std::path::Path, server_uri: &str) -> std::io::Result<()> {
     let config_toml = codex_home.join("config.toml");

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

@@ -29,7 +29,7 @@ async fn thread_start_creates_thread_and_emits_started() -> Result<()> {
     // Start a v2 thread with an explicit model override.
     let req_id = mcp
         .send_thread_start_request(ThreadStartParams {
-            model: Some("gpt-5".to_string()),
+            model: Some("gpt-5.1".to_string()),
             ..Default::default()
         })
         .await?;

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

@@ -5,10 +5,13 @@ use app_test_support::create_mock_chat_completions_server;
 use app_test_support::create_mock_chat_completions_server_unchecked;
 use app_test_support::create_shell_sse_response;
 use app_test_support::to_response;
+use codex_app_server_protocol::CommandExecutionStatus;
+use codex_app_server_protocol::ItemStartedNotification;
 use codex_app_server_protocol::JSONRPCNotification;
 use codex_app_server_protocol::JSONRPCResponse;
 use codex_app_server_protocol::RequestId;
 use codex_app_server_protocol::ServerRequest;
+use codex_app_server_protocol::ThreadItem;
 use codex_app_server_protocol::ThreadStartParams;
 use codex_app_server_protocol::ThreadStartResponse;
 use codex_app_server_protocol::TurnStartParams;
@@ -17,9 +20,6 @@ use codex_app_server_protocol::TurnStartedNotification;
 use codex_app_server_protocol::UserInput as V2UserInput;
 use codex_core::protocol_config_types::ReasoningEffort;
 use codex_core::protocol_config_types::ReasoningSummary;
-use codex_protocol::parse_command::ParsedCommand;
-use codex_protocol::protocol::Event;
-use codex_protocol::protocol::EventMsg;
 use core_test_support::skip_if_no_network;
 use pretty_assertions::assert_eq;
 use std::path::Path;
@@ -235,7 +235,7 @@ async fn turn_start_exec_approval_toggle_v2() -> Result<()> {
     .await??;
     let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(start_resp)?;
 
-    // turn/start — expect ExecCommandApproval request from server
+    // turn/start — expect CommandExecutionRequestApproval request from server
     let first_turn_id = mcp
         .send_turn_start_request(TurnStartParams {
             thread_id: thread.id.clone(),
@@ -258,16 +258,10 @@ async fn turn_start_exec_approval_toggle_v2() -> Result<()> {
         mcp.read_stream_until_request_message(),
     )
     .await??;
-    let ServerRequest::ExecCommandApproval { request_id, params } = server_req else {
-        panic!("expected ExecCommandApproval request");
+    let ServerRequest::CommandExecutionRequestApproval { request_id, params } = server_req else {
+        panic!("expected CommandExecutionRequestApproval request");
     };
-    assert_eq!(params.call_id, "call1");
-    assert_eq!(
-        params.parsed_cmd,
-        vec![ParsedCommand::Unknown {
-            cmd: "python3 -c 'print(42)'".to_string()
-        }]
-    );
+    assert_eq!(params.item_id, "call1");
 
     // Approve and wait for task completion
     mcp.send_response(
@@ -302,7 +296,7 @@ async fn turn_start_exec_approval_toggle_v2() -> Result<()> {
     )
     .await??;
 
-    // Ensure we do NOT receive an ExecCommandApproval request before task completes
+    // Ensure we do NOT receive a CommandExecutionRequestApproval request before task completes
     timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_notification_message("codex/event/task_complete"),
@@ -314,8 +308,6 @@ async fn turn_start_exec_approval_toggle_v2() -> Result<()> {
 
 #[tokio::test]
 async fn turn_start_updates_sandbox_and_cwd_between_turns_v2() -> Result<()> {
-    // When returning Result from a test, pass an Ok(()) to the skip macro
-    // so the early return type matches. The no-arg form returns unit.
     skip_if_no_network!(Ok(()));
 
     let tmp = TempDir::new()?;
@@ -424,29 +416,35 @@ async fn turn_start_updates_sandbox_and_cwd_between_turns_v2() -> Result<()> {
     )
     .await??;
 
-    let exec_begin_notification = timeout(
-        DEFAULT_READ_TIMEOUT,
-        mcp.read_stream_until_notification_message("codex/event/exec_command_begin"),
-    )
+    let command_exec_item = timeout(DEFAULT_READ_TIMEOUT, async {
+        loop {
+            let item_started_notification = mcp
+                .read_stream_until_notification_message("item/started")
+                .await?;
+            let params = item_started_notification
+                .params
+                .clone()
+                .expect("item/started params");
+            let item_started: ItemStartedNotification =
+                serde_json::from_value(params).expect("deserialize item/started notification");
+            if matches!(item_started.item, ThreadItem::CommandExecution { .. }) {
+                return Ok::<ThreadItem, anyhow::Error>(item_started.item);
+            }
+        }
+    })
     .await??;
-    let params = exec_begin_notification
-        .params
-        .clone()
-        .expect("exec_command_begin params");
-    let event: Event = serde_json::from_value(params).expect("deserialize exec begin event");
-    let exec_begin = match event.msg {
-        EventMsg::ExecCommandBegin(exec_begin) => exec_begin,
-        other => panic!("expected ExecCommandBegin event, got {other:?}"),
+    let ThreadItem::CommandExecution {
+        cwd,
+        command,
+        status,
+        ..
+    } = command_exec_item
+    else {
+        unreachable!("loop ensures we break on command execution items");
     };
-    assert_eq!(exec_begin.cwd, second_cwd);
-    assert_eq!(
-        exec_begin.command,
-        vec![
-            "bash".to_string(),
-            "-lc".to_string(),
-            "echo second turn".to_string()
-        ]
-    );
+    assert_eq!(cwd, second_cwd);
+    assert_eq!(command, "bash -lc 'echo second turn'");
+    assert_eq!(status, CommandExecutionStatus::InProgress);
 
     timeout(
         DEFAULT_READ_TIMEOUT,

codex-rs/cli/Cargo.toml

@@ -26,6 +26,7 @@ codex-cloud-tasks = { path = "../cloud-tasks" }
 codex-common = { workspace = true, features = ["cli"] }
 codex-core = { workspace = true }
 codex-exec = { workspace = true }
+codex-execpolicy2 = { workspace = true }
 codex-login = { workspace = true }
 codex-mcp-server = { workspace = true }
 codex-process-hardening = { workspace = true }

codex-rs/cli/src/debug_sandbox.rs

@@ -155,11 +155,11 @@ async fn run_command_under_sandbox(
                 run_windows_sandbox_capture(
                     policy_str,
                     &sandbox_cwd,
+                    base_dir.as_path(),
                     command_vec,
                     &cwd_clone,
                     env_map,
                     None,
-                    Some(base_dir.as_path()),
                 )
             })
             .await;

codex-rs/cli/src/main.rs

@@ -18,6 +18,7 @@ use codex_cli::login::run_logout;
 use codex_cloud_tasks::Cli as CloudTasksCli;
 use codex_common::CliConfigOverrides;
 use codex_exec::Cli as ExecCli;
+use codex_execpolicy2::ExecPolicyCheckCommand;
 use codex_responses_api_proxy::Args as ResponsesApiProxyArgs;
 use codex_tui::AppExitInfo;
 use codex_tui::Cli as TuiCli;
@@ -112,6 +113,10 @@ enum Subcommand {
     #[clap(hide = true, name = "stdio-to-uds")]
     StdioToUds(StdioToUdsCommand),
 
+    /// Check execpolicy files against a command.
+    #[clap(name = "execpolicycheck")]
+    ExecPolicyCheck(ExecPolicyCheckCommand),
+
     /// Inspect feature flags.
     Features(FeaturesCli),
 }
@@ -323,6 +328,12 @@ fn run_update_action(action: UpdateAction) -> anyhow::Result<()> {
     Ok(())
 }
 
+fn run_execpolicycheck(cmd: ExecPolicyCheckCommand) -> anyhow::Result<()> {
+    let json = cmd.run()?;
+    println!("{json}");
+    Ok(())
+}
+
 #[derive(Debug, Default, Parser, Clone)]
 struct FeatureToggles {
     /// Enable a feature (repeatable). Equivalent to `-c features.<name>=true`.
@@ -559,6 +570,7 @@ 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::ExecPolicyCheck(cmd)) => run_execpolicycheck(cmd)?,
         Some(Subcommand::Features(FeaturesCli { sub })) => match sub {
             FeaturesSubcommand::List => {
                 // Respect root-level `-c` overrides plus top-level flags like `--profile`.
@@ -761,9 +773,9 @@ mod tests {
 
     #[test]
     fn resume_model_flag_applies_when_no_root_flags() {
-        let interactive = finalize_from_args(["codex", "resume", "-m", "gpt-5-test"].as_ref());
+        let interactive = finalize_from_args(["codex", "resume", "-m", "gpt-5.1-test"].as_ref());
 
-        assert_eq!(interactive.model.as_deref(), Some("gpt-5-test"));
+        assert_eq!(interactive.model.as_deref(), Some("gpt-5.1-test"));
         assert!(interactive.resume_picker);
         assert!(!interactive.resume_last);
         assert_eq!(interactive.resume_session_id, None);
@@ -808,7 +820,7 @@ mod tests {
                 "--ask-for-approval",
                 "on-request",
                 "-m",
-                "gpt-5-test",
+                "gpt-5.1-test",
                 "-p",
                 "my-profile",
                 "-C",
@@ -819,7 +831,7 @@ mod tests {
             .as_ref(),
         );
 
-        assert_eq!(interactive.model.as_deref(), Some("gpt-5-test"));
+        assert_eq!(interactive.model.as_deref(), Some("gpt-5.1-test"));
         assert!(interactive.oss);
         assert_eq!(interactive.config_profile.as_deref(), Some("my-profile"));
         assert_matches!(

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

@@ -1721,6 +1721,7 @@ mod tests {
     use ratatui::layout::Rect;
 
     #[test]
+    #[ignore = "very slow"]
     fn composer_input_renders_typed_characters() {
         let mut composer = ComposerInput::new();
         let key = KeyEvent::new(KeyCode::Char('a'), KeyModifiers::NONE);

codex-rs/common/Cargo.toml

@@ -8,9 +8,12 @@ workspace = true
 
 [dependencies]
 clap = { workspace = true, features = ["derive", "wrap_help"], optional = true }
-codex-core = { workspace = true }
-codex-protocol = { workspace = true }
 codex-app-server-protocol = { workspace = true }
+codex-core = { workspace = true }
+codex-lmstudio = { workspace = true }
+codex-ollama = { workspace = true }
+codex-protocol = { workspace = true }
+once_cell = { workspace = true }
 serde = { workspace = true, optional = true }
 toml = { workspace = true, optional = true }
 

codex-rs/common/src/approval_presets.rs

@@ -24,21 +24,21 @@ pub fn builtin_approval_presets() -> Vec<ApprovalPreset> {
         ApprovalPreset {
             id: "read-only",
             label: "Read Only",
-            description: "Codex can read files and answer questions. Codex requires approval to make edits, run commands, or access network.",
+            description: "Requires approval to edit files and run commands.",
             approval: AskForApproval::OnRequest,
             sandbox: SandboxPolicy::ReadOnly,
         },
         ApprovalPreset {
             id: "auto",
-            label: "Auto",
-            description: "Codex can read files, make edits, and run commands in the workspace. Codex requires approval to work outside the workspace or access network.",
+            label: "Agent",
+            description: "Read and edit files, and run commands.",
             approval: AskForApproval::OnRequest,
             sandbox: SandboxPolicy::new_workspace_write_policy(),
         },
         ApprovalPreset {
             id: "full-access",
-            label: "Full Access",
-            description: "Codex can read files, make edits, and run commands with network access, without approval. Exercise caution.",
+            label: "Agent (full access)",
+            description: "Codex can edit files outside this workspace and run commands with network access. Exercise caution when using.",
             approval: AskForApproval::Never,
             sandbox: SandboxPolicy::DangerFullAccess,
         },

codex-rs/common/src/lib.rs

@@ -37,3 +37,5 @@ pub mod model_presets;
 // Shared approval presets (AskForApproval + Sandbox) used by TUI and MCP server
 // Not to be confused with AskForApproval, which we should probably rename to EscalationPolicy.
 pub mod approval_presets;
+// Shared OSS provider utilities used by TUI and exec
+pub mod oss;

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/common/src/oss.rs

@@ -0,0 +1,60 @@
+//! OSS provider utilities shared between TUI and exec.
+
+use codex_core::LMSTUDIO_OSS_PROVIDER_ID;
+use codex_core::OLLAMA_OSS_PROVIDER_ID;
+use codex_core::config::Config;
+
+/// Returns the default model for a given OSS provider.
+pub fn get_default_model_for_oss_provider(provider_id: &str) -> Option<&'static str> {
+    match provider_id {
+        LMSTUDIO_OSS_PROVIDER_ID => Some(codex_lmstudio::DEFAULT_OSS_MODEL),
+        OLLAMA_OSS_PROVIDER_ID => Some(codex_ollama::DEFAULT_OSS_MODEL),
+        _ => None,
+    }
+}
+
+/// Ensures the specified OSS provider is ready (models downloaded, service reachable).
+pub async fn ensure_oss_provider_ready(
+    provider_id: &str,
+    config: &Config,
+) -> Result<(), std::io::Error> {
+    match provider_id {
+        LMSTUDIO_OSS_PROVIDER_ID => {
+            codex_lmstudio::ensure_oss_ready(config)
+                .await
+                .map_err(|e| std::io::Error::other(format!("OSS setup failed: {e}")))?;
+        }
+        OLLAMA_OSS_PROVIDER_ID => {
+            codex_ollama::ensure_oss_ready(config)
+                .await
+                .map_err(|e| std::io::Error::other(format!("OSS setup failed: {e}")))?;
+        }
+        _ => {
+            // Unknown provider, skip setup
+        }
+    }
+    Ok(())
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_get_default_model_for_provider_lmstudio() {
+        let result = get_default_model_for_oss_provider(LMSTUDIO_OSS_PROVIDER_ID);
+        assert_eq!(result, Some(codex_lmstudio::DEFAULT_OSS_MODEL));
+    }
+
+    #[test]
+    fn test_get_default_model_for_provider_ollama() {
+        let result = get_default_model_for_oss_provider(OLLAMA_OSS_PROVIDER_ID);
+        assert_eq!(result, Some(codex_ollama::DEFAULT_OSS_MODEL));
+    }
+
+    #[test]
+    fn test_get_default_model_for_provider_unknown() {
+        let result = get_default_model_for_oss_provider("unknown-provider");
+        assert_eq!(result, None);
+    }
+}

codex-rs/core/Cargo.toml

@@ -22,6 +22,7 @@ chrono = { workspace = true, features = ["serde"] }
 codex-app-server-protocol = { workspace = true }
 codex-apply-patch = { workspace = true }
 codex-async-utils = { workspace = true }
+codex-execpolicy2 = { workspace = true }
 codex-file-search = { workspace = true }
 codex-git = { workspace = true }
 codex-keyring-store = { workspace = true }
@@ -40,12 +41,7 @@ eventsource-stream = { workspace = true }
 futures = { workspace = true }
 http = { workspace = true }
 indexmap = { workspace = true }
-keyring = { workspace = true, features = [
-    "apple-native",
-    "crypto-rust",
-    "linux-native-async-persistent",
-    "windows-native",
-] }
+keyring = { workspace = true, features = ["crypto-rust"] }
 libc = { workspace = true }
 mcp-types = { workspace = true }
 os_info = { workspace = true }
@@ -60,6 +56,7 @@ 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 = [
@@ -89,9 +86,11 @@ wildmatch = { workspace = true }
 [target.'cfg(target_os = "linux")'.dependencies]
 landlock = { workspace = true }
 seccompiler = { workspace = true }
+keyring = { workspace = true, features = ["linux-native-async-persistent"] }
 
 [target.'cfg(target_os = "macos")'.dependencies]
 core-foundation = "0.9"
+keyring = { workspace = true, features = ["apple-native"] }
 
 # Build OpenSSL from source for musl builds.
 [target.x86_64-unknown-linux-musl.dependencies]
@@ -101,6 +100,12 @@ openssl-sys = { workspace = true, features = ["vendored"] }
 [target.aarch64-unknown-linux-musl.dependencies]
 openssl-sys = { workspace = true, features = ["vendored"] }
 
+[target.'cfg(target_os = "windows")'.dependencies]
+keyring = { workspace = true, features = ["windows-native"] }
+
+[target.'cfg(any(target_os = "freebsd", target_os = "openbsd"))'.dependencies]
+keyring = { workspace = true, features = ["sync-secret-service"] }
+
 [dev-dependencies]
 assert_cmd = { workspace = true }
 assert_matches = { workspace = true }

codex-rs/core/gpt_5_1_prompt.md

@@ -0,0 +1,368 @@
+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:
+
+- 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

@@ -2,8 +2,6 @@ You are Codex, based on GPT-5. You are running as a coding agent in the Codex CL
 
 ## General
 
-- The arguments to `shell` will be passed to execvp(). Most terminal commands should be prefixed with ["bash", "-lc"].
-- 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.)
 
 ## Editing constraints

codex-rs/core/src/auth.rs

@@ -227,6 +227,14 @@ impl CodexAuth {
             })
     }
 
+    /// 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
@@ -335,7 +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,

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> {
@@ -88,23 +93,16 @@ pub fn try_parse_word_only_commands_sequence(tree: &Tree, src: &str) -> Option<V
     Some(commands)
 }
 
-pub fn is_well_known_sh_shell(shell: &str) -> bool {
-    if shell == "bash" || shell == "zsh" {
-        return true;
-    }
-
-    let shell_name = std::path::Path::new(shell)
-        .file_name()
-        .and_then(|s| s.to_str())
-        .unwrap_or(shell);
-    matches!(shell_name, "bash" | "zsh")
-}
-
 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") || !is_well_known_sh_shell(shell) {
+    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))

codex-rs/core/src/chat_completions.rs

@@ -81,6 +81,7 @@ pub(crate) async fn stream_chat_completions(
             ResponseItem::CustomToolCallOutput { .. } => {}
             ResponseItem::WebSearchCall { .. } => {}
             ResponseItem::GhostSnapshot { .. } => {}
+            ResponseItem::CompactionSummary { .. } => {}
         }
     }
 
@@ -320,7 +321,8 @@ pub(crate) async fn stream_chat_completions(
             }
             ResponseItem::Reasoning { .. }
             | ResponseItem::WebSearchCall { .. }
-            | ResponseItem::Other => {
+            | ResponseItem::Other
+            | ResponseItem::CompactionSummary { .. } => {
                 // Omit these items from the conversation history.
                 continue;
             }
@@ -338,7 +340,7 @@ 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 +479,14 @@ 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;
     }
 }
@@ -669,7 +675,9 @@ async fn process_chat_sse<S>(
             }
 
             // Emit end-of-turn when finish_reason signals completion.
-            if let Some(finish_reason) = choice.get("finish_reason").and_then(|v| v.as_str()) {
+            if let Some(finish_reason) = choice.get("finish_reason").and_then(|v| v.as_str())
+                && !finish_reason.is_empty()
+            {
                 match finish_reason {
                     "tool_calls" if fn_call_state.active => {
                         // First, flush the terminal raw reasoning so UIs can finalize
@@ -898,20 +906,26 @@ 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

@@ -26,6 +26,7 @@ use tokio::sync::mpsc;
 use tokio::time::timeout;
 use tokio_util::io::ReaderStream;
 use tracing::debug;
+use tracing::enabled;
 use tracing::trace;
 use tracing::warn;
 
@@ -35,10 +36,10 @@ 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;
@@ -78,6 +79,18 @@ struct Error {
     resets_at: Option<i64>,
 }
 
+#[derive(Debug, Serialize)]
+struct CompactHistoryRequest<'a> {
+    model: &'a str,
+    input: &'a [ResponseItem],
+    instructions: &'a str,
+}
+
+#[derive(Debug, Deserialize)]
+struct CompactHistoryResponse {
+    output: Vec<ResponseItem>,
+}
+
 #[derive(Debug, Clone)]
 pub struct ModelClient {
     config: Arc<Config>,
@@ -199,12 +212,18 @@ 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 +234,9 @@ 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 +315,9 @@ 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
@@ -500,6 +520,70 @@ impl ModelClient {
     pub fn get_auth_manager(&self) -> Option<Arc<AuthManager>> {
         self.auth_manager.clone()
     }
+
+    pub async fn compact_conversation_history(&self, prompt: &Prompt) -> Result<Vec<ResponseItem>> {
+        if prompt.input.is_empty() {
+            return Ok(Vec::new());
+        }
+        let auth_manager = self.auth_manager.clone();
+        let auth = auth_manager.as_ref().and_then(|m| m.auth());
+        let mut req_builder = self
+            .provider
+            .create_compact_request_builder(&self.client, &auth)
+            .await?;
+        if let SessionSource::SubAgent(sub) = &self.session_source {
+            let subagent = if let crate::protocol::SubAgentSource::Other(label) = sub {
+                label.clone()
+            } else {
+                serde_json::to_value(sub)
+                    .ok()
+                    .and_then(|v| v.as_str().map(std::string::ToString::to_string))
+                    .unwrap_or_else(|| "other".to_string())
+            };
+            req_builder = req_builder.header("x-openai-subagent", subagent);
+        }
+        if let Some(auth) = auth.as_ref()
+            && auth.mode == AuthMode::ChatGPT
+            && let Some(account_id) = auth.get_account_id()
+        {
+            req_builder = req_builder.header("chatgpt-account-id", account_id);
+        }
+        let payload = CompactHistoryRequest {
+            model: &self.config.model,
+            input: &prompt.input,
+            instructions: &prompt.get_full_instructions(&self.config.model_family),
+        };
+
+        if enabled!(tracing::Level::TRACE) {
+            trace!(
+                "POST to {}: {}",
+                self.provider
+                    .get_compact_url(&auth)
+                    .unwrap_or("<none>".to_string()),
+                serde_json::to_value(&payload).unwrap_or_default()
+            );
+        }
+
+        let response = req_builder
+            .json(&payload)
+            .send()
+            .await
+            .map_err(|source| CodexErr::ConnectionFailed(ConnectionFailedError { source }))?;
+        let status = response.status();
+        let body = response
+            .text()
+            .await
+            .map_err(|source| CodexErr::ConnectionFailed(ConnectionFailedError { source }))?;
+        if !status.is_success() {
+            return Err(CodexErr::UnexpectedStatus(UnexpectedResponseError {
+                status,
+                body,
+                request_id: None,
+            }));
+        }
+        let CompactHistoryResponse { output } = serde_json::from_str(&body)?;
+        Ok(output)
+    }
 }
 
 enum StreamAttemptError {
@@ -553,6 +637,8 @@ struct SseEvent {
     response: Option<Value>,
     item: Option<Value>,
     delta: Option<String>,
+    summary_index: Option<i64>,
+    content_index: Option<i64>,
 }
 
 #[derive(Debug, Deserialize)]
@@ -812,16 +898,22 @@ 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;
                     }
@@ -898,10 +990,12 @@ 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" => {}
@@ -1208,7 +1302,7 @@ mod tests {
 
     #[tokio::test]
     async fn error_when_error_event() {
-        let raw_error = r#"{"type":"response.failed","sequence_number":3,"response":{"id":"resp_689bcf18d7f08194bf3440ba62fe05d803fee0cdac429894","object":"response","created_at":1755041560,"status":"failed","background":false,"error":{"code":"rate_limit_exceeded","message":"Rate limit reached for gpt-5 in organization org-AAA on tokens per min (TPM): Limit 30000, Used 22999, Requested 12528. Please try again in 11.054s. Visit https://platform.openai.com/account/rate-limits to learn more."}, "usage":null,"user":null,"metadata":{}}}"#;
+        let raw_error = r#"{"type":"response.failed","sequence_number":3,"response":{"id":"resp_689bcf18d7f08194bf3440ba62fe05d803fee0cdac429894","object":"response","created_at":1755041560,"status":"failed","background":false,"error":{"code":"rate_limit_exceeded","message":"Rate limit reached for gpt-5.1 in organization org-AAA on tokens per min (TPM): Limit 30000, Used 22999, Requested 12528. Please try again in 11.054s. Visit https://platform.openai.com/account/rate-limits to learn more."}, "usage":null,"user":null,"metadata":{}}}"#;
 
         let sse1 = format!("event: response.failed\ndata: {raw_error}\n\n");
         let provider = ModelProviderInfo {
@@ -1237,7 +1331,7 @@ mod tests {
             Err(CodexErr::Stream(msg, delay)) => {
                 assert_eq!(
                     msg,
-                    "Rate limit reached for gpt-5 in organization org-AAA on tokens per min (TPM): Limit 30000, Used 22999, Requested 12528. Please try again in 11.054s. Visit https://platform.openai.com/account/rate-limits to learn more."
+                    "Rate limit reached for gpt-5.1 in organization org-AAA on tokens per min (TPM): Limit 30000, Used 22999, Requested 12528. Please try again in 11.054s. Visit https://platform.openai.com/account/rate-limits to learn more."
                 );
                 assert_eq!(*delay, Some(Duration::from_secs_f64(11.054)));
             }
@@ -1456,7 +1550,7 @@ mod tests {
     fn test_try_parse_retry_after() {
         let err = Error {
             r#type: None,
-            message: Some("Rate limit reached for gpt-5 in organization org- on tokens per min (TPM): Limit 1, Used 1, Requested 19304. Please try again in 28ms. Visit https://platform.openai.com/account/rate-limits to learn more.".to_string()),
+            message: Some("Rate limit reached for gpt-5.1 in organization org- on tokens per min (TPM): Limit 1, Used 1, Requested 19304. Please try again in 28ms. Visit https://platform.openai.com/account/rate-limits to learn more.".to_string()),
             code: Some("rate_limit_exceeded".to_string()),
             plan_type: None,
             resets_at: None
@@ -1470,7 +1564,7 @@ mod tests {
     fn test_try_parse_retry_after_no_delay() {
         let err = Error {
             r#type: None,
-            message: Some("Rate limit reached for gpt-5 in organization <ORG> on tokens per min (TPM): Limit 30000, Used 6899, Requested 24050. Please try again in 1.898s. Visit https://platform.openai.com/account/rate-limits to learn more.".to_string()),
+            message: Some("Rate limit reached for gpt-5.1 in organization <ORG> on tokens per min (TPM): Limit 30000, Used 6899, Requested 24050. Please try again in 1.898s. Visit https://platform.openai.com/account/rate-limits to learn more.".to_string()),
             code: Some("rate_limit_exceeded".to_string()),
             plan_type: None,
             resets_at: None

codex-rs/core/src/client_common.rs

@@ -136,7 +136,7 @@ fn reserialize_shell_outputs(items: &mut [ResponseItem]) {
 }
 
 fn is_shell_tool_name(name: &str) -> bool {
-    matches!(name, "shell" | "container.exec")
+    matches!(name, "shell" | "container.exec" | "shell_command")
 }
 
 #[derive(Deserialize)]
@@ -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,
@@ -430,7 +427,11 @@ mod tests {
                 expects_apply_patch_instructions: false,
             },
             InstructionsTestCase {
-                slug: "gpt-5-codex",
+                slug: "gpt-5.1-codex",
+                expects_apply_patch_instructions: false,
+            },
+            InstructionsTestCase {
+                slug: "gpt-5.1-codex",
                 expects_apply_patch_instructions: false,
             },
         ];
@@ -456,7 +457,7 @@ mod tests {
         let input: Vec<ResponseItem> = vec![];
         let tools: Vec<serde_json::Value> = vec![];
         let req = ResponsesApiRequest {
-            model: "gpt-5",
+            model: "gpt-5.1",
             instructions: "i",
             input: &input,
             tools: &tools,
@@ -497,7 +498,7 @@ mod tests {
             create_text_param_for_request(None, &Some(schema.clone())).expect("text controls");
 
         let req = ResponsesApiRequest {
-            model: "gpt-5",
+            model: "gpt-5.1",
             instructions: "i",
             input: &input,
             tools: &tools,
@@ -533,7 +534,7 @@ mod tests {
         let input: Vec<ResponseItem> = vec![];
         let tools: Vec<serde_json::Value> = vec![];
         let req = ResponsesApiRequest {
-            model: "gpt-5",
+            model: "gpt-5.1",
             instructions: "i",
             input: &input,
             tools: &tools,

codex-rs/core/src/codex.rs

@@ -9,12 +9,11 @@ use crate::client_common::REVIEW_PROMPT;
 use crate::compact;
 use crate::features::Feature;
 use crate::function_tool::FunctionCallError;
-use crate::mcp::auth::McpAuthStatusEntry;
-use crate::mcp_connection_manager::DEFAULT_STARTUP_TIMEOUT;
 use crate::parse_command::parse_command;
 use crate::parse_turn_item;
 use crate::response_processing::process_items;
 use crate::terminal;
+use crate::truncate::TruncationPolicy;
 use crate::user_notification::UserNotifier;
 use crate::util::error_or_panic;
 use async_channel::Receiver;
@@ -45,6 +44,7 @@ use mcp_types::ReadResourceResult;
 use serde_json;
 use serde_json::Value;
 use tokio::sync::Mutex;
+use tokio::sync::RwLock;
 use tokio::sync::oneshot;
 use tokio_util::sync::CancellationToken;
 use tracing::debug;
@@ -56,8 +56,8 @@ use crate::ModelProviderInfo;
 use crate::client::ModelClient;
 use crate::client_common::Prompt;
 use crate::client_common::ResponseEvent;
+use crate::compact::collect_user_messages;
 use crate::config::Config;
-use crate::config::types::McpServerTransportConfig;
 use crate::config::types::ShellEnvironmentPolicy;
 use crate::context_manager::ContextManager;
 use crate::environment_context::EnvironmentContext;
@@ -65,10 +65,6 @@ use crate::error::CodexErr;
 use crate::error::Result as CodexResult;
 #[cfg(test)]
 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;
@@ -96,6 +92,7 @@ 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;
@@ -121,6 +118,7 @@ use crate::user_instructions::UserInstructions;
 use crate::user_notification::UserNotification;
 use crate::util::backoff;
 use codex_async_utils::OrCancelExt;
+use codex_execpolicy2::Policy as ExecPolicy;
 use codex_otel::otel_event_manager::OtelEventManager;
 use codex_protocol::config_types::ReasoningEffort as ReasoningEffortConfig;
 use codex_protocol::config_types::ReasoningSummary as ReasoningSummaryConfig;
@@ -132,6 +130,7 @@ use codex_protocol::protocol::InitialHistory;
 use codex_protocol::user_input::UserInput;
 use codex_utils_readiness::Readiness;
 use codex_utils_readiness::ReadinessFlag;
+use codex_utils_tokenizer::warm_model_cache;
 
 /// The high-level interface to the Codex system.
 /// It operates as a queue pair where you send submissions and receive events.
@@ -165,6 +164,10 @@ impl Codex {
 
         let user_instructions = get_user_instructions(&config).await;
 
+        let exec_policy = crate::exec_policy::exec_policy_for(&config.features, &config.codex_home)
+            .await
+            .map_err(|err| CodexErr::Fatal(format!("failed to load execpolicy: {err}")))?;
+
         let config = Arc::new(config);
 
         let session_configuration = SessionConfiguration {
@@ -181,6 +184,7 @@ impl Codex {
             cwd: config.cwd.clone(),
             original_config_do_not_use: Arc::clone(&config),
             features: config.features.clone(),
+            exec_policy,
             session_source,
         };
 
@@ -278,6 +282,8 @@ pub(crate) struct TurnContext {
     pub(crate) final_output_json_schema: Option<Value>,
     pub(crate) codex_linux_sandbox_exe: Option<PathBuf>,
     pub(crate) tool_call_gate: Arc<ReadinessFlag>,
+    pub(crate) exec_policy: Arc<ExecPolicy>,
+    pub(crate) truncation_policy: TruncationPolicy,
 }
 
 impl TurnContext {
@@ -334,6 +340,8 @@ pub(crate) struct SessionConfiguration {
 
     /// Set of feature flags for this session
     features: Features,
+    /// Execpolicy policy, applied only when enabled by feature flag.
+    exec_policy: Arc<ExecPolicy>,
 
     // TODO(pakrym): Remove config from here
     original_config_do_not_use: Arc<Config>,
@@ -404,7 +412,7 @@ impl Session {
         );
 
         let client = ModelClient::new(
-            Arc::new(per_turn_config),
+            Arc::new(per_turn_config.clone()),
             auth_manager,
             otel_event_manager,
             provider,
@@ -434,6 +442,8 @@ impl Session {
             final_output_json_schema: None,
             codex_linux_sandbox_exe: config.codex_linux_sandbox_exe.clone(),
             tool_call_gate: Arc::new(ReadinessFlag::new()),
+            exec_policy: session_configuration.exec_policy.clone(),
+            truncation_policy: TruncationPolicy::new(&per_turn_config),
         }
     }
 
@@ -474,21 +484,13 @@ impl Session {
             ),
         };
 
-        // Error messages to dispatch after SessionConfigured is sent.
-        let mut post_session_configured_events = Vec::<Event>::new();
-
         // Kick off independent async setup tasks in parallel to reduce startup latency.
         //
         // - initialize RolloutRecorder with new or resumed session info
-        // - spin up MCP connection manager
         // - perform default shell discovery
         // - load history metadata
         let rollout_fut = RolloutRecorder::new(&config, rollout_params);
 
-        let mcp_fut = McpConnectionManager::new(
-            config.mcp_servers.clone(),
-            config.mcp_oauth_credentials_store_mode,
-        );
         let default_shell_fut = shell::default_user_shell();
         let history_meta_fut = crate::message_history::history_metadata(&config);
         let auth_statuses_fut = compute_auth_statuses(
@@ -497,15 +499,8 @@ impl Session {
         );
 
         // Join all independent futures.
-        let (
-            rollout_recorder,
-            mcp_res,
-            default_shell,
-            (history_log_id, history_entry_count),
-            auth_statuses,
-        ) = tokio::join!(
+        let (rollout_recorder, default_shell, (history_log_id, history_entry_count), auth_statuses) = tokio::join!(
             rollout_fut,
-            mcp_fut,
             default_shell_fut,
             history_meta_fut,
             auth_statuses_fut
@@ -517,34 +512,7 @@ impl Session {
         })?;
         let rollout_path = rollout_recorder.rollout_path.clone();
 
-        // Handle MCP manager result and record any startup failures.
-        let (mcp_connection_manager, failed_clients) = match mcp_res {
-            Ok((mgr, failures)) => (mgr, failures),
-            Err(e) => {
-                let message = format!("Failed to create MCP connection manager: {e:#}");
-                error!("{message}");
-                post_session_configured_events.push(Event {
-                    id: INITIAL_SUBMIT_ID.to_owned(),
-                    msg: EventMsg::Error(ErrorEvent { message }),
-                });
-                (McpConnectionManager::default(), Default::default())
-            }
-        };
-
-        // Surface individual client start-up failures to the user.
-        if !failed_clients.is_empty() {
-            for (server_name, err) in failed_clients {
-                let auth_entry = auth_statuses.get(&server_name);
-                let display_message = mcp_init_error_display(&server_name, auth_entry, &err);
-                warn!("MCP client for `{server_name}` failed to start: {err:#}");
-                post_session_configured_events.push(Event {
-                    id: INITIAL_SUBMIT_ID.to_owned(),
-                    msg: EventMsg::Error(ErrorEvent {
-                        message: display_message,
-                    }),
-                });
-            }
-        }
+        let mut post_session_configured_events = Vec::<Event>::new();
 
         for (alias, feature) in session_configuration.features.legacy_feature_usages() {
             let canonical = feature.key();
@@ -589,8 +557,12 @@ impl Session {
         // Create the mutable state for the Session.
         let state = SessionState::new(session_configuration.clone());
 
+        // Warm the tokenizer cache for the session model without blocking startup.
+        warm_model_cache(&session_configuration.model);
+
         let services = SessionServices {
-            mcp_connection_manager,
+            mcp_connection_manager: Arc::new(RwLock::new(McpConnectionManager::default())),
+            mcp_startup_cancellation_token: CancellationToken::new(),
             unified_exec_manager: UnifiedExecSessionManager::default(),
             notifier: UserNotifier::new(config.notify.clone()),
             rollout: Mutex::new(Some(rollout_recorder)),
@@ -630,6 +602,18 @@ impl Session {
         for event in events {
             sess.send_event_raw(event).await;
         }
+        sess.services
+            .mcp_connection_manager
+            .write()
+            .await
+            .initialize(
+                config.mcp_servers.clone(),
+                config.mcp_oauth_credentials_store_mode,
+                auth_statuses.clone(),
+                tx_event.clone(),
+                sess.services.mcp_startup_cancellation_token.clone(),
+            )
+            .await;
 
         // record_initial_history can emit events. We record only after the SessionConfiguredEvent is emitted.
         sess.record_initial_history(initial_history).await;
@@ -707,7 +691,8 @@ impl Session {
                 let reconstructed_history =
                     self.reconstruct_history_from_rollout(&turn_context, &rollout_items);
                 if !reconstructed_history.is_empty() {
-                    self.record_into_history(&reconstructed_history).await;
+                    self.record_into_history(&reconstructed_history, &turn_context)
+                        .await;
                 }
 
                 // If persisting, persist all rollout items as-is (recorder filters)
@@ -889,6 +874,7 @@ impl Session {
         let parsed_cmd = parse_command(&command);
         let event = EventMsg::ExecApprovalRequest(ExecApprovalRequestEvent {
             call_id,
+            turn_id: turn_context.sub_id.clone(),
             command,
             cwd,
             reason,
@@ -963,7 +949,7 @@ impl Session {
         turn_context: &TurnContext,
         items: &[ResponseItem],
     ) {
-        self.record_into_history(items).await;
+        self.record_into_history(items, turn_context).await;
         self.persist_rollout_response_items(items).await;
         self.send_raw_response_items(turn_context, items).await;
     }
@@ -977,17 +963,25 @@ impl Session {
         for item in rollout_items {
             match item {
                 RolloutItem::ResponseItem(response_item) => {
-                    history.record_items(std::iter::once(response_item));
+                    history.record_items(
+                        std::iter::once(response_item),
+                        turn_context.truncation_policy,
+                    );
                 }
                 RolloutItem::Compacted(compacted) => {
                     let snapshot = history.get_history();
-                    let user_messages = collect_user_messages(&snapshot);
-                    let rebuilt = build_compacted_history(
-                        self.build_initial_context(turn_context),
-                        &user_messages,
-                        &compacted.message,
-                    );
-                    history.replace(rebuilt);
+                    // TODO(jif) clean
+                    if let Some(replacement) = &compacted.replacement_history {
+                        history.replace(replacement.clone());
+                    } else {
+                        let user_messages = collect_user_messages(&snapshot);
+                        let rebuilt = compact::build_compacted_history(
+                            self.build_initial_context(turn_context),
+                            &user_messages,
+                            &compacted.message,
+                        );
+                        history.replace(rebuilt);
+                    }
                 }
                 _ => {}
             }
@@ -996,9 +990,13 @@ impl Session {
     }
 
     /// Append ResponseItems to the in-memory conversation history only.
-    pub(crate) async fn record_into_history(&self, items: &[ResponseItem]) {
+    pub(crate) async fn record_into_history(
+        &self,
+        items: &[ResponseItem],
+        turn_context: &TurnContext,
+    ) {
         let mut state = self.state.lock().await;
-        state.record_items(items.iter());
+        state.record_items(items.iter(), turn_context.truncation_policy);
     }
 
     pub(crate) async fn replace_history(&self, items: Vec<ResponseItem>) {
@@ -1015,6 +1013,15 @@ impl Session {
         self.persist_rollout_items(&rollout_items).await;
     }
 
+    pub async fn enabled(&self, feature: Feature) -> bool {
+        self.state
+            .lock()
+            .await
+            .session_configuration
+            .features
+            .enabled(feature)
+    }
+
     async fn send_raw_response_items(&self, turn_context: &TurnContext, items: &[ResponseItem]) {
         for item in items {
             self.send_event(
@@ -1082,6 +1089,36 @@ impl Session {
         self.send_token_count_event(turn_context).await;
     }
 
+    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,
@@ -1162,14 +1199,7 @@ impl Session {
         turn_context: Arc<TurnContext>,
         cancellation_token: CancellationToken,
     ) {
-        if !self
-            .state
-            .lock()
-            .await
-            .session_configuration
-            .features
-            .enabled(Feature::GhostCommit)
-        {
+        if !self.enabled(Feature::GhostCommit).await {
             return;
         }
         let token = match turn_context.tool_call_gate.subscribe().await {
@@ -1223,6 +1253,8 @@ impl Session {
     ) -> anyhow::Result<ListResourcesResult> {
         self.services
             .mcp_connection_manager
+            .read()
+            .await
             .list_resources(server, params)
             .await
     }
@@ -1234,6 +1266,8 @@ impl Session {
     ) -> anyhow::Result<ListResourceTemplatesResult> {
         self.services
             .mcp_connection_manager
+            .read()
+            .await
             .list_resource_templates(server, params)
             .await
     }
@@ -1245,6 +1279,8 @@ impl Session {
     ) -> anyhow::Result<ReadResourceResult> {
         self.services
             .mcp_connection_manager
+            .read()
+            .await
             .read_resource(server, params)
             .await
     }
@@ -1257,19 +1293,29 @@ impl Session {
     ) -> anyhow::Result<CallToolResult> {
         self.services
             .mcp_connection_manager
+            .read()
+            .await
             .call_tool(server, tool, arguments)
             .await
     }
 
-    pub(crate) fn parse_mcp_tool_name(&self, tool_name: &str) -> Option<(String, String)> {
+    pub(crate) async fn parse_mcp_tool_name(&self, tool_name: &str) -> Option<(String, String)> {
         self.services
             .mcp_connection_manager
+            .read()
+            .await
             .parse_tool_name(tool_name)
+            .await
     }
 
     pub async fn interrupt_task(self: &Arc<Self>) {
         info!("interrupt received: abort current task, if any");
-        self.abort_all_tasks(TurnAbortReason::Interrupted).await;
+        let has_active_turn = { self.active_turn.lock().await.is_some() };
+        if has_active_turn {
+            self.abort_all_tasks(TurnAbortReason::Interrupted).await;
+        } else {
+            self.cancel_mcp_startup().await;
+        }
     }
 
     pub(crate) fn notifier(&self) -> &UserNotifier {
@@ -1283,6 +1329,10 @@ impl Session {
     fn show_raw_agent_reasoning(&self) -> bool {
         self.services.show_raw_agent_reasoning
     }
+
+    async fn cancel_mcp_startup(&self) {
+        self.services.mcp_startup_cancellation_token.cancel();
+    }
 }
 
 async fn submission_loop(sess: Arc<Session>, config: Arc<Config>, rx_sub: Receiver<Submission>) {
@@ -1390,7 +1440,7 @@ mod handlers {
     use codex_protocol::protocol::ReviewDecision;
     use codex_protocol::protocol::ReviewRequest;
     use codex_protocol::protocol::TurnAbortReason;
-    use codex_protocol::user_input::UserInput;
+
     use std::sync::Arc;
     use tracing::info;
     use tracing::warn;
@@ -1540,17 +1590,15 @@ mod handlers {
     }
 
     pub async fn list_mcp_tools(sess: &Session, config: &Arc<Config>, sub_id: String) {
-        // This is a cheap lookup from the connection manager's cache.
-        let tools = sess.services.mcp_connection_manager.list_all_tools();
-        let (auth_status_entries, resources, resource_templates) = tokio::join!(
+        let mcp_connection_manager = sess.services.mcp_connection_manager.read().await;
+        let (tools, auth_status_entries, resources, resource_templates) = tokio::join!(
+            mcp_connection_manager.list_all_tools(),
             compute_auth_statuses(
                 config.mcp_servers.iter(),
                 config.mcp_oauth_credentials_store_mode,
             ),
-            sess.services.mcp_connection_manager.list_all_resources(),
-            sess.services
-                .mcp_connection_manager
-                .list_all_resource_templates()
+            mcp_connection_manager.list_all_resources(),
+            mcp_connection_manager.list_all_resource_templates(),
         );
         let auth_statuses = auth_status_entries
             .iter()
@@ -1559,7 +1607,10 @@ mod handlers {
         let event = Event {
             id: sub_id,
             msg: EventMsg::McpListToolsResponse(crate::protocol::McpListToolsResponseEvent {
-                tools,
+                tools: tools
+                    .into_iter()
+                    .map(|(name, tool)| (name, tool.tool))
+                    .collect(),
                 resources,
                 resource_templates,
                 auth_statuses,
@@ -1597,16 +1648,9 @@ mod handlers {
         let turn_context = sess
             .new_turn_with_sub_id(sub_id, SessionSettingsUpdate::default())
             .await;
-        // Attempt to inject input into current task
-        if let Err(items) = sess
-            .inject_input(vec![UserInput::Text {
-                text: turn_context.compact_prompt().to_string(),
-            }])
-            .await
-        {
-            sess.spawn_task(Arc::clone(&turn_context), items, CompactTask)
-                .await;
-        }
+
+        sess.spawn_task(Arc::clone(&turn_context), vec![], CompactTask)
+            .await;
     }
 
     pub async fn shutdown(sess: &Arc<Session>, sub_id: String) -> bool {
@@ -1732,6 +1776,8 @@ async fn spawn_review_thread(
         final_output_json_schema: None,
         codex_linux_sandbox_exe: parent_turn_context.codex_linux_sandbox_exe.clone(),
         tool_call_gate: Arc::new(ReadinessFlag::new()),
+        exec_policy: parent_turn_context.exec_policy.clone(),
+        truncation_policy: TruncationPolicy::new(&per_turn_config),
     };
 
     // Seed the child task with the review prompt as the initial user message.
@@ -1784,7 +1830,6 @@ pub(crate) async fn run_task(
     // Although from the perspective of codex.rs, TurnDiffTracker has the lifecycle of a Task which contains
     // many turns, from the perspective of the user, it is a single turn.
     let turn_diff_tracker = Arc::new(tokio::sync::Mutex::new(TurnDiffTracker::new()));
-    let mut auto_compact_recently_attempted = false;
 
     loop {
         // Note that pending_input would be something like a message the user
@@ -1839,27 +1884,12 @@ pub(crate) async fn run_task(
                 let (responses, items_to_record_in_conversation_history) =
                     process_items(processed_items, &sess, &turn_context).await;
 
+                // as long as compaction works well in getting us way below the token limit, we shouldn't worry about being in an infinite loop.
                 if token_limit_reached {
-                    if auto_compact_recently_attempted {
-                        let limit_str = limit.to_string();
-                        let current_tokens = total_usage_tokens
-                            .map(|tokens| tokens.to_string())
-                            .unwrap_or_else(|| "unknown".to_string());
-                        let event = EventMsg::Error(ErrorEvent {
-                            message: format!(
-                                "Conversation is still above the token limit after automatic summarization (limit {limit_str}, current {current_tokens}). Please start a new session or trim your input."
-                            ),
-                        });
-                        sess.send_event(&turn_context, event).await;
-                        break;
-                    }
-                    auto_compact_recently_attempted = true;
                     compact::run_inline_auto_compact_task(sess.clone(), turn_context.clone()).await;
                     continue;
                 }
 
-                auto_compact_recently_attempted = false;
-
                 if responses.is_empty() {
                     last_agent_message = get_last_assistant_message_from_turn(
                         &items_to_record_in_conversation_history,
@@ -1905,22 +1935,52 @@ async fn run_turn(
     input: Vec<ResponseItem>,
     cancellation_token: CancellationToken,
 ) -> CodexResult<TurnRunResult> {
-    let mcp_tools = sess.services.mcp_connection_manager.list_all_tools();
+    let mcp_tools = sess
+        .services
+        .mcp_connection_manager
+        .read()
+        .await
+        .list_all_tools()
+        .or_cancel(&cancellation_token)
+        .await?;
     let router = Arc::new(ToolRouter::from_config(
         &turn_context.tools_config,
-        Some(mcp_tools),
+        Some(
+            mcp_tools
+                .into_iter()
+                .map(|(name, tool)| (name, tool.tool))
+                .collect(),
+        ),
     ));
 
     let model_supports_parallel = turn_context
         .client
         .get_model_family()
         .supports_parallel_tool_calls;
-    let parallel_tool_calls = model_supports_parallel;
+
+    // TODO(jif) revert once testing phase is done.
+    let parallel_tool_calls = model_supports_parallel
+        && sess
+            .state
+            .lock()
+            .await
+            .session_configuration
+            .features
+            .enabled(Feature::ParallelToolCalls);
+    let mut base_instructions = turn_context.base_instructions.clone();
+    if parallel_tool_calls {
+        static INSTRUCTIONS: &str = include_str!("../templates/parallel/instructions.md");
+        static INSERTION_SPOT: &str = "## Editing constraints";
+        base_instructions
+            .as_mut()
+            .map(|base| base.replace(INSERTION_SPOT, INSTRUCTIONS));
+    }
+
     let prompt = Prompt {
         input,
         tools: router.specs(),
         parallel_tool_calls,
-        base_instructions_override: turn_context.base_instructions.clone(),
+        base_instructions_override: base_instructions,
         output_schema: turn_context.final_output_json_schema.clone(),
     };
 
@@ -2077,7 +2137,7 @@ async fn try_run_turn(
             ResponseEvent::Created => {}
             ResponseEvent::OutputItemDone(item) => {
                 let previously_active_item = active_item.take();
-                match ToolRouter::build_tool_call(sess.as_ref(), item.clone()) {
+                match ToolRouter::build_tool_call(sess.as_ref(), item.clone()).await {
                     Ok(Some(call)) => {
                         let payload_preview = call.payload.log_payload().into_owned();
                         tracing::info!("ToolCall: {} {}", call.tool_name, payload_preview);
@@ -2201,13 +2261,17 @@ 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;
@@ -2215,18 +2279,29 @@ 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;
@@ -2273,59 +2348,6 @@ pub(super) fn get_last_assistant_message_from_turn(responses: &[ResponseItem]) -
     })
 }
 
-fn mcp_init_error_display(
-    server_name: &str,
-    entry: Option<&McpAuthStatusEntry>,
-    err: &anyhow::Error,
-) -> String {
-    if let Some(McpServerTransportConfig::StreamableHttp {
-        url,
-        bearer_token_env_var,
-        http_headers,
-        ..
-    }) = &entry.map(|entry| &entry.config.transport)
-        && url == "https://api.githubcopilot.com/mcp/"
-        && bearer_token_env_var.is_none()
-        && http_headers.as_ref().map(HashMap::is_empty).unwrap_or(true)
-    {
-        // GitHub only supports OAUth for first party MCP clients.
-        // That means that the user has to specify a personal access token either via bearer_token_env_var or http_headers.
-        // https://github.com/github/github-mcp-server/issues/921#issuecomment-3221026448
-        format!(
-            "GitHub MCP does not support OAuth. Log in by adding a personal access token (https://github.com/settings/personal-access-tokens) to your environment and config.toml:\n[mcp_servers.{server_name}]\nbearer_token_env_var = CODEX_GITHUB_PERSONAL_ACCESS_TOKEN"
-        )
-    } else if is_mcp_client_auth_required_error(err) {
-        format!(
-            "The {server_name} MCP server is not logged in. Run `codex mcp login {server_name}`."
-        )
-    } else if is_mcp_client_startup_timeout_error(err) {
-        let startup_timeout_secs = match entry {
-            Some(entry) => match entry.config.startup_timeout_sec {
-                Some(timeout) => timeout,
-                None => DEFAULT_STARTUP_TIMEOUT,
-            },
-            None => DEFAULT_STARTUP_TIMEOUT,
-        }
-        .as_secs();
-        format!(
-            "MCP client for `{server_name}` timed out after {startup_timeout_secs} seconds. Add or adjust `startup_timeout_sec` in your config.toml:\n[mcp_servers.{server_name}]\nstartup_timeout_sec = XX"
-        )
-    } else {
-        format!("MCP client for `{server_name}` failed to start: {err:#}")
-    }
-}
-
-fn is_mcp_client_auth_required_error(error: &anyhow::Error) -> bool {
-    // StreamableHttpError::AuthRequired from the MCP SDK.
-    error.to_string().contains("Auth required")
-}
-
-fn is_mcp_client_startup_timeout_error(error: &anyhow::Error) -> bool {
-    let error_message = error.to_string();
-    error_message.contains("request timed out")
-        || error_message.contains("timed out handshaking with MCP server")
-}
-
 use crate::features::Features;
 #[cfg(test)]
 pub(crate) use tests::make_session_and_context;
@@ -2335,10 +2357,7 @@ mod tests {
     use super::*;
     use crate::config::ConfigOverrides;
     use crate::config::ConfigToml;
-    use crate::config::types::McpServerConfig;
-    use crate::config::types::McpServerTransportConfig;
     use crate::exec::ExecToolCallOutput;
-    use crate::mcp::auth::McpAuthStatusEntry;
     use crate::tools::format_exec_output_str;
 
     use crate::protocol::CompactedItem;
@@ -2358,7 +2377,6 @@ mod tests {
     use codex_app_server_protocol::AuthMode;
     use codex_protocol::models::ContentItem;
     use codex_protocol::models::ResponseItem;
-    use codex_protocol::protocol::McpAuthStatus;
     use std::time::Duration;
     use tokio::time::sleep;
 
@@ -2566,13 +2584,15 @@ mod tests {
             cwd: config.cwd.clone(),
             original_config_do_not_use: Arc::clone(&config),
             features: Features::default(),
+            exec_policy: Arc::new(codex_execpolicy2::PolicyParser::new().build()),
             session_source: SessionSource::Exec,
         };
 
         let state = SessionState::new(session_configuration.clone());
 
         let services = SessionServices {
-            mcp_connection_manager: McpConnectionManager::default(),
+            mcp_connection_manager: Arc::new(RwLock::new(McpConnectionManager::default())),
+            mcp_startup_cancellation_token: CancellationToken::new(),
             unified_exec_manager: UnifiedExecSessionManager::default(),
             notifier: UserNotifier::new(None),
             rollout: Mutex::new(None),
@@ -2642,13 +2662,15 @@ mod tests {
             cwd: config.cwd.clone(),
             original_config_do_not_use: Arc::clone(&config),
             features: Features::default(),
+            exec_policy: Arc::new(codex_execpolicy2::PolicyParser::new().build()),
             session_source: SessionSource::Exec,
         };
 
         let state = SessionState::new(session_configuration.clone());
 
         let services = SessionServices {
-            mcp_connection_manager: McpConnectionManager::default(),
+            mcp_connection_manager: Arc::new(RwLock::new(McpConnectionManager::default())),
+            mcp_startup_cancellation_token: CancellationToken::new(),
             unified_exec_manager: UnifiedExecSessionManager::default(),
             notifier: UserNotifier::new(None),
             rollout: Mutex::new(None),
@@ -2829,9 +2851,23 @@ mod tests {
     #[tokio::test]
     async fn fatal_tool_error_stops_turn_and_reports_error() {
         let (session, turn_context, _rx) = make_session_and_context_with_rx();
+        let tools = {
+            session
+                .services
+                .mcp_connection_manager
+                .read()
+                .await
+                .list_all_tools()
+                .await
+        };
         let router = ToolRouter::from_config(
             &turn_context.tools_config,
-            Some(session.services.mcp_connection_manager.list_all_tools()),
+            Some(
+                tools
+                    .into_iter()
+                    .map(|(name, tool)| (name, tool.tool))
+                    .collect(),
+            ),
         );
         let item = ResponseItem::CustomToolCall {
             id: None,
@@ -2842,6 +2878,7 @@ mod tests {
         };
 
         let call = ToolRouter::build_tool_call(session.as_ref(), item.clone())
+            .await
             .expect("build tool call")
             .expect("tool call present");
         let tracker = Arc::new(tokio::sync::Mutex::new(TurnDiffTracker::new()));
@@ -2874,7 +2911,7 @@ mod tests {
         for item in &initial_context {
             rollout_items.push(RolloutItem::ResponseItem(item.clone()));
         }
-        live_history.record_items(initial_context.iter());
+        live_history.record_items(initial_context.iter(), turn_context.truncation_policy);
 
         let user1 = ResponseItem::Message {
             id: None,
@@ -2883,7 +2920,7 @@ mod tests {
                 text: "first user".to_string(),
             }],
         };
-        live_history.record_items(std::iter::once(&user1));
+        live_history.record_items(std::iter::once(&user1), turn_context.truncation_policy);
         rollout_items.push(RolloutItem::ResponseItem(user1.clone()));
 
         let assistant1 = ResponseItem::Message {
@@ -2893,13 +2930,13 @@ mod tests {
                 text: "assistant reply one".to_string(),
             }],
         };
-        live_history.record_items(std::iter::once(&assistant1));
+        live_history.record_items(std::iter::once(&assistant1), turn_context.truncation_policy);
         rollout_items.push(RolloutItem::ResponseItem(assistant1.clone()));
 
         let summary1 = "summary one";
         let snapshot1 = live_history.get_history();
         let user_messages1 = collect_user_messages(&snapshot1);
-        let rebuilt1 = build_compacted_history(
+        let rebuilt1 = compact::build_compacted_history(
             session.build_initial_context(turn_context),
             &user_messages1,
             summary1,
@@ -2907,6 +2944,7 @@ mod tests {
         live_history.replace(rebuilt1);
         rollout_items.push(RolloutItem::Compacted(CompactedItem {
             message: summary1.to_string(),
+            replacement_history: None,
         }));
 
         let user2 = ResponseItem::Message {
@@ -2916,7 +2954,7 @@ mod tests {
                 text: "second user".to_string(),
             }],
         };
-        live_history.record_items(std::iter::once(&user2));
+        live_history.record_items(std::iter::once(&user2), turn_context.truncation_policy);
         rollout_items.push(RolloutItem::ResponseItem(user2.clone()));
 
         let assistant2 = ResponseItem::Message {
@@ -2926,13 +2964,13 @@ mod tests {
                 text: "assistant reply two".to_string(),
             }],
         };
-        live_history.record_items(std::iter::once(&assistant2));
+        live_history.record_items(std::iter::once(&assistant2), turn_context.truncation_policy);
         rollout_items.push(RolloutItem::ResponseItem(assistant2.clone()));
 
         let summary2 = "summary two";
         let snapshot2 = live_history.get_history();
         let user_messages2 = collect_user_messages(&snapshot2);
-        let rebuilt2 = build_compacted_history(
+        let rebuilt2 = compact::build_compacted_history(
             session.build_initial_context(turn_context),
             &user_messages2,
             summary2,
@@ -2940,6 +2978,7 @@ mod tests {
         live_history.replace(rebuilt2);
         rollout_items.push(RolloutItem::Compacted(CompactedItem {
             message: summary2.to_string(),
+            replacement_history: None,
         }));
 
         let user3 = ResponseItem::Message {
@@ -2949,7 +2988,7 @@ mod tests {
                 text: "third user".to_string(),
             }],
         };
-        live_history.record_items(std::iter::once(&user3));
+        live_history.record_items(std::iter::once(&user3), turn_context.truncation_policy);
         rollout_items.push(RolloutItem::ResponseItem(user3.clone()));
 
         let assistant3 = ResponseItem::Message {
@@ -2959,7 +2998,7 @@ mod tests {
                 text: "assistant reply three".to_string(),
             }],
         };
-        live_history.record_items(std::iter::once(&assistant3));
+        live_history.record_items(std::iter::once(&assistant3), turn_context.truncation_policy);
         rollout_items.push(RolloutItem::ResponseItem(assistant3.clone()));
 
         (rollout_items, live_history.get_history())
@@ -3091,7 +3130,6 @@ mod tests {
         pretty_assertions::assert_eq!(exec_output.metadata, ResponseExecMetadata { exit_code: 0 });
         assert!(exec_output.output.contains("hi"));
     }
-
     #[tokio::test]
     async fn unified_exec_rejects_escalated_permissions_when_policy_not_on_request() {
         use crate::protocol::AskForApproval;
@@ -3133,89 +3171,4 @@ mod tests {
 
         pretty_assertions::assert_eq!(output, expected);
     }
-
-    #[test]
-    fn mcp_init_error_display_prompts_for_github_pat() {
-        let server_name = "github";
-        let entry = McpAuthStatusEntry {
-            config: McpServerConfig {
-                transport: McpServerTransportConfig::StreamableHttp {
-                    url: "https://api.githubcopilot.com/mcp/".to_string(),
-                    bearer_token_env_var: None,
-                    http_headers: None,
-                    env_http_headers: None,
-                },
-                enabled: true,
-                startup_timeout_sec: None,
-                tool_timeout_sec: None,
-                enabled_tools: None,
-                disabled_tools: None,
-            },
-            auth_status: McpAuthStatus::Unsupported,
-        };
-        let err = anyhow::anyhow!("OAuth is unsupported");
-
-        let display = mcp_init_error_display(server_name, Some(&entry), &err);
-
-        let expected = format!(
-            "GitHub MCP does not support OAuth. Log in by adding a personal access token (https://github.com/settings/personal-access-tokens) to your environment and config.toml:\n[mcp_servers.{server_name}]\nbearer_token_env_var = CODEX_GITHUB_PERSONAL_ACCESS_TOKEN"
-        );
-
-        assert_eq!(expected, display);
-    }
-
-    #[test]
-    fn mcp_init_error_display_prompts_for_login_when_auth_required() {
-        let server_name = "example";
-        let err = anyhow::anyhow!("Auth required for server");
-
-        let display = mcp_init_error_display(server_name, None, &err);
-
-        let expected = format!(
-            "The {server_name} MCP server is not logged in. Run `codex mcp login {server_name}`."
-        );
-
-        assert_eq!(expected, display);
-    }
-
-    #[test]
-    fn mcp_init_error_display_reports_generic_errors() {
-        let server_name = "custom";
-        let entry = McpAuthStatusEntry {
-            config: McpServerConfig {
-                transport: McpServerTransportConfig::StreamableHttp {
-                    url: "https://example.com".to_string(),
-                    bearer_token_env_var: Some("TOKEN".to_string()),
-                    http_headers: None,
-                    env_http_headers: None,
-                },
-                enabled: true,
-                startup_timeout_sec: None,
-                tool_timeout_sec: None,
-                enabled_tools: None,
-                disabled_tools: None,
-            },
-            auth_status: McpAuthStatus::Unsupported,
-        };
-        let err = anyhow::anyhow!("boom");
-
-        let display = mcp_init_error_display(server_name, Some(&entry), &err);
-
-        let expected = format!("MCP client for `{server_name}` failed to start: {err:#}");
-
-        assert_eq!(expected, display);
-    }
-
-    #[test]
-    fn mcp_init_error_display_includes_startup_timeout_hint() {
-        let server_name = "slow";
-        let err = anyhow::anyhow!("request timed out");
-
-        let display = mcp_init_error_display(server_name, None, &err);
-
-        assert_eq!(
-            "MCP client for `slow` timed out after 10 seconds. Add or adjust `startup_timeout_sec` in your config.toml:\n[mcp_servers.slow]\nstartup_timeout_sec = XX",
-            display
-        );
-    }
 }

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

@@ -1,6 +1,8 @@
 use codex_protocol::protocol::AskForApproval;
 use codex_protocol::protocol::SandboxPolicy;
 
+use crate::sandboxing::SandboxPermissions;
+
 use crate::bash::parse_shell_lc_plain_commands;
 use crate::is_safe_command::is_known_safe_command;
 
@@ -8,7 +10,7 @@ pub fn requires_initial_appoval(
     policy: AskForApproval,
     sandbox_policy: &SandboxPolicy,
     command: &[String],
-    with_escalated_permissions: bool,
+    sandbox_permissions: SandboxPermissions,
 ) -> bool {
     if is_known_safe_command(command) {
         return false;
@@ -24,8 +26,7 @@ pub fn requires_initial_appoval(
             // 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 {
+            if sandbox_permissions.requires_escalated_permissions() {
                 return true;
             }
             command_might_be_dangerous(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

@@ -14,7 +14,9 @@ use crate::protocol::EventMsg;
 use crate::protocol::TaskStartedEvent;
 use crate::protocol::TurnContextItem;
 use crate::protocol::WarningEvent;
-use crate::truncate::truncate_middle;
+use crate::truncate::TruncationPolicy;
+use crate::truncate::approx_token_count;
+use crate::truncate::truncate_text;
 use crate::util::backoff;
 use codex_protocol::items::TurnItem;
 use codex_protocol::models::ContentItem;
@@ -26,6 +28,7 @@ use futures::prelude::*;
 use tracing::error;
 
 pub const SUMMARIZATION_PROMPT: &str = include_str!("../templates/compact/prompt.md");
+pub const SUMMARY_PREFIX: &str = include_str!("../templates/compact/summary_prefix.md");
 const COMPACT_USER_MESSAGE_MAX_TOKENS: usize = 20_000;
 
 pub(crate) async fn run_inline_auto_compact_task(
@@ -58,7 +61,10 @@ async fn run_compact_task_inner(
     let initial_input_for_turn: ResponseInputItem = ResponseInputItem::from(input);
 
     let mut history = sess.clone_history().await;
-    history.record_items(&[initial_input_for_turn.into()]);
+    history.record_items(
+        &[initial_input_for_turn.into()],
+        turn_context.truncation_policy,
+    );
 
     let mut truncated_count = 0usize;
 
@@ -140,7 +146,9 @@ async fn run_compact_task_inner(
     }
 
     let history_snapshot = sess.clone_history().await.get_history();
-    let summary_text = get_last_assistant_message_from_turn(&history_snapshot).unwrap_or_default();
+    let summary_suffix =
+        get_last_assistant_message_from_turn(&history_snapshot).unwrap_or_default();
+    let summary_text = format!("{SUMMARY_PREFIX}\n{summary_suffix}");
     let user_messages = collect_user_messages(&history_snapshot);
 
     let initial_context = sess.build_initial_context(turn_context.as_ref());
@@ -153,8 +161,18 @@ 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(),
+        replacement_history: None,
     });
     sess.persist_rollout_items(&[rollout_item]).await;
 
@@ -192,12 +210,22 @@ pub(crate) fn collect_user_messages(items: &[ResponseItem]) -> Vec<String> {
     items
         .iter()
         .filter_map(|item| match crate::event_mapping::parse_turn_item(item) {
-            Some(TurnItem::UserMessage(user)) => Some(user.message()),
+            Some(TurnItem::UserMessage(user)) => {
+                if is_summary_message(&user.message()) {
+                    None
+                } else {
+                    Some(user.message())
+                }
+            }
             _ => None,
         })
         .collect()
 }
 
+pub(crate) fn is_summary_message(message: &str) -> bool {
+    message.starts_with(format!("{SUMMARY_PREFIX}\n").as_str())
+}
+
 pub(crate) fn build_compacted_history(
     initial_context: Vec<ResponseItem>,
     user_messages: &[String],
@@ -207,7 +235,7 @@ pub(crate) fn build_compacted_history(
         initial_context,
         user_messages,
         summary_text,
-        COMPACT_USER_MESSAGE_MAX_TOKENS * 4,
+        COMPACT_USER_MESSAGE_MAX_TOKENS,
     )
 }
 
@@ -215,20 +243,21 @@ fn build_compacted_history_with_limit(
     mut history: Vec<ResponseItem>,
     user_messages: &[String],
     summary_text: &str,
-    max_bytes: usize,
+    max_tokens: usize,
 ) -> Vec<ResponseItem> {
     let mut selected_messages: Vec<String> = Vec::new();
-    if max_bytes > 0 {
-        let mut remaining = max_bytes;
+    if max_tokens > 0 {
+        let mut remaining = max_tokens;
         for message in user_messages.iter().rev() {
             if remaining == 0 {
                 break;
             }
-            if message.len() <= remaining {
+            let tokens = approx_token_count(message);
+            if tokens <= remaining {
                 selected_messages.push(message.clone());
-                remaining = remaining.saturating_sub(message.len());
+                remaining = remaining.saturating_sub(tokens);
             } else {
-                let (truncated, _) = truncate_middle(message, remaining);
+                let truncated = truncate_text(message, TruncationPolicy::Tokens(remaining));
                 selected_messages.push(truncated);
                 break;
             }
@@ -277,7 +306,8 @@ async fn drain_to_completed(
         };
         match event {
             Ok(ResponseEvent::OutputItemDone(item)) => {
-                sess.record_into_history(std::slice::from_ref(&item)).await;
+                sess.record_into_history(std::slice::from_ref(&item), turn_context)
+                    .await;
             }
             Ok(ResponseEvent::RateLimits(snapshot)) => {
                 sess.update_rate_limits(turn_context, snapshot).await;
@@ -295,6 +325,7 @@ async fn drain_to_completed(
 
 #[cfg(test)]
 mod tests {
+
     use super::*;
     use pretty_assertions::assert_eq;
 
@@ -386,16 +417,16 @@ mod tests {
     }
 
     #[test]
-    fn build_compacted_history_truncates_overlong_user_messages() {
+    fn build_token_limited_compacted_history_truncates_overlong_user_messages() {
         // Use a small truncation limit so the test remains fast while still validating
         // that oversized user content is truncated.
-        let max_bytes = 128;
-        let big = "X".repeat(max_bytes + 50);
+        let max_tokens = 16;
+        let big = "word ".repeat(200);
         let history = super::build_compacted_history_with_limit(
             Vec::new(),
             std::slice::from_ref(&big),
             "SUMMARY",
-            max_bytes,
+            max_tokens,
         );
         assert_eq!(history.len(), 2);
 
@@ -428,7 +459,7 @@ mod tests {
     }
 
     #[test]
-    fn build_compacted_history_appends_summary_message() {
+    fn build_token_limited_compacted_history_appends_summary_message() {
         let initial_context: Vec<ResponseItem> = Vec::new();
         let user_messages = vec!["first user message".to_string()];
         let summary_text = "summary text";

codex-rs/core/src/compact_remote.rs

@@ -0,0 +1,100 @@
+use std::sync::Arc;
+
+use crate::Prompt;
+use crate::codex::Session;
+use crate::codex::TurnContext;
+use crate::error::Result as CodexResult;
+use crate::protocol::AgentMessageEvent;
+use crate::protocol::CompactedItem;
+use crate::protocol::ErrorEvent;
+use crate::protocol::EventMsg;
+use crate::protocol::RolloutItem;
+use crate::protocol::TaskStartedEvent;
+use codex_protocol::models::ResponseInputItem;
+use codex_protocol::models::ResponseItem;
+use codex_protocol::user_input::UserInput;
+
+pub(crate) async fn run_remote_compact_task(
+    sess: Arc<Session>,
+    turn_context: Arc<TurnContext>,
+    input: Vec<UserInput>,
+) -> Option<String> {
+    let start_event = EventMsg::TaskStarted(TaskStartedEvent {
+        model_context_window: turn_context.client.get_model_context_window(),
+    });
+    sess.send_event(&turn_context, start_event).await;
+
+    match run_remote_compact_task_inner(&sess, &turn_context, input).await {
+        Ok(()) => {
+            let event = EventMsg::AgentMessage(AgentMessageEvent {
+                message: "Compact task completed".to_string(),
+            });
+            sess.send_event(&turn_context, event).await;
+        }
+        Err(err) => {
+            let event = EventMsg::Error(ErrorEvent {
+                message: err.to_string(),
+            });
+            sess.send_event(&turn_context, event).await;
+        }
+    }
+
+    None
+}
+
+async fn run_remote_compact_task_inner(
+    sess: &Arc<Session>,
+    turn_context: &Arc<TurnContext>,
+    input: Vec<UserInput>,
+) -> CodexResult<()> {
+    let mut history = sess.clone_history().await;
+    if !input.is_empty() {
+        let initial_input_for_turn: ResponseInputItem = ResponseInputItem::from(input);
+        history.record_items(
+            &[initial_input_for_turn.into()],
+            turn_context.truncation_policy,
+        );
+    }
+
+    let prompt = Prompt {
+        input: history.get_history_for_prompt(),
+        tools: vec![],
+        parallel_tool_calls: false,
+        base_instructions_override: turn_context.base_instructions.clone(),
+        output_schema: None,
+    };
+
+    let mut new_history = turn_context
+        .client
+        .compact_conversation_history(&prompt)
+        .await?;
+    // Required to keep `/undo` available after compaction
+    let ghost_snapshots: Vec<ResponseItem> = history
+        .get_history()
+        .iter()
+        .filter(|item| matches!(item, ResponseItem::GhostSnapshot { .. }))
+        .cloned()
+        .collect();
+
+    if !ghost_snapshots.is_empty() {
+        new_history.extend(ghost_snapshots);
+    }
+    sess.replace_history(new_history.clone()).await;
+
+    if let Some(estimated_tokens) = sess
+        .clone_history()
+        .await
+        .estimate_token_count(turn_context.as_ref())
+    {
+        sess.override_last_token_usage_estimate(turn_context.as_ref(), estimated_tokens)
+            .await;
+    }
+
+    let compacted_item = CompactedItem {
+        message: String::new(),
+        replacement_history: Some(new_history),
+    };
+    sess.persist_rollout_items(&[RolloutItem::Compacted(compacted_item)])
+        .await;
+    Ok(())
+}

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

@@ -3,6 +3,8 @@ use crate::config::types::McpServerConfig;
 use crate::config::types::Notice;
 use anyhow::Context;
 use codex_protocol::config_types::ReasoningEffort;
+use codex_protocol::config_types::TrustLevel;
+use codex_utils_tokenizer::warm_model_cache;
 use std::collections::BTreeMap;
 use std::path::Path;
 use std::path::PathBuf;
@@ -29,11 +31,13 @@ pub enum ConfigEdit {
     SetNoticeHideRateLimitModelNudge(bool),
     /// Toggle the Windows onboarding acknowledgement flag.
     SetWindowsWslSetupAcknowledged(bool),
+    /// Toggle the model migration prompt acknowledgement flag.
+    SetNoticeHideModelMigrationPrompt(String, bool),
     /// Replace the entire `[mcp_servers]` table.
     ReplaceMcpServers(BTreeMap<String, McpServerConfig>),
-    /// Set trust_level = "trusted" under `[projects."<path>"]`,
+    /// Set trust_level under `[projects."<path>"]`,
     /// migrating inline tables to explicit tables.
-    SetProjectTrusted(PathBuf),
+    SetProjectTrustLevel { path: PathBuf, level: TrustLevel },
     /// Set the value stored at the exact dotted path.
     SetPath {
         segments: Vec<String>,
@@ -227,6 +231,9 @@ impl ConfigDocument {
     fn apply(&mut self, edit: &ConfigEdit) -> anyhow::Result<bool> {
         match edit {
             ConfigEdit::SetModel { model, effort } => Ok({
+                if let Some(model) = &model {
+                    warm_model_cache(model)
+                }
                 let mut mutated = false;
                 mutated |= self.write_profile_value(
                     &["model"],
@@ -253,6 +260,13 @@ impl ConfigDocument {
                 &[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"],
@@ -261,10 +275,14 @@ impl ConfigDocument {
             ConfigEdit::ReplaceMcpServers(servers) => Ok(self.replace_mcp_servers(servers)),
             ConfigEdit::SetPath { segments, value } => Ok(self.insert(segments, value.clone())),
             ConfigEdit::ClearPath { segments } => Ok(self.clear_owned(segments)),
-            ConfigEdit::SetProjectTrusted(project_path) => {
+            ConfigEdit::SetProjectTrustLevel { path, level } => {
                 // Delegate to the existing, tested logic in config.rs to
                 // ensure tables are explicit and migration is preserved.
-                crate::config::set_project_trusted_inner(&mut self.doc, project_path.as_path())?;
+                crate::config::set_project_trust_level_inner(
+                    &mut self.doc,
+                    path.as_path(),
+                    *level,
+                )?;
                 Ok(true)
             }
         }
@@ -499,6 +517,15 @@ impl ConfigEditsBuilder {
         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));
@@ -511,9 +538,24 @@ impl ConfigEditsBuilder {
         self
     }
 
-    pub fn set_project_trusted<P: Into<PathBuf>>(mut self, project_path: P) -> Self {
-        self.edits
-            .push(ConfigEdit::SetProjectTrusted(project_path.into()));
+    pub fn set_project_trust_level<P: Into<PathBuf>>(
+        mut self,
+        project_path: P,
+        trust_level: TrustLevel,
+    ) -> Self {
+        self.edits.push(ConfigEdit::SetProjectTrustLevel {
+            path: project_path.into(),
+            level: trust_level,
+        });
+        self
+    }
+
+    /// Enable or disable a feature flag by key under the `[features]` table.
+    pub fn set_feature_enabled(mut self, key: &str, enabled: bool) -> Self {
+        self.edits.push(ConfigEdit::SetPath {
+            segments: vec!["features".to_string(), key.to_string()],
+            value: value(enabled),
+        });
         self
     }
 
@@ -551,7 +593,7 @@ mod tests {
             codex_home,
             None,
             &[ConfigEdit::SetModel {
-                model: Some("gpt-5-codex".to_string()),
+                model: Some("gpt-5.1-codex".to_string()),
                 effort: Some(ReasoningEffort::High),
             }],
         )
@@ -559,7 +601,7 @@ mod tests {
 
         let contents =
             std::fs::read_to_string(codex_home.join(CONFIG_TOML_FILE)).expect("read config");
-        let expected = r#"model = "gpt-5-codex"
+        let expected = r#"model = "gpt-5.1-codex"
 model_reasoning_effort = "high"
 "#;
         assert_eq!(contents, expected);
@@ -689,7 +731,7 @@ model = "o5-preview"
         std::fs::write(
             codex_home.join(CONFIG_TOML_FILE),
             r#"[profiles."team a"]
-model = "gpt-5-codex"
+model = "gpt-5.1-codex"
 "#,
         )
         .expect("seed");
@@ -770,6 +812,35 @@ existing = "value"
         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);
     }
@@ -910,14 +981,14 @@ B = \"2\"
         let codex_home = tmp.path().to_path_buf();
 
         ConfigEditsBuilder::new(&codex_home)
-            .set_model(Some("gpt-5-codex"), Some(ReasoningEffort::High))
+            .set_model(Some("gpt-5.1-codex"), Some(ReasoningEffort::High))
             .apply()
             .await
             .expect("persist");
 
         let contents =
             std::fs::read_to_string(codex_home.join(CONFIG_TOML_FILE)).expect("read config");
-        let expected = r#"model = "gpt-5-codex"
+        let expected = r#"model = "gpt-5.1-codex"
 model_reasoning_effort = "high"
 "#;
         assert_eq!(contents, expected);
@@ -939,11 +1010,11 @@ model_reasoning_effort = "low"
             std::fs::read_to_string(codex_home.join(CONFIG_TOML_FILE)).expect("read config");
         assert_eq!(contents, initial_expected);
 
-        let updated_expected = r#"model = "gpt-5-codex"
+        let updated_expected = r#"model = "gpt-5.1-codex"
 model_reasoning_effort = "high"
 "#;
         ConfigEditsBuilder::new(codex_home)
-            .set_model(Some("gpt-5-codex"), Some(ReasoningEffort::High))
+            .set_model(Some("gpt-5.1-codex"), Some(ReasoningEffort::High))
             .apply_blocking()
             .expect("persist update");
         contents = std::fs::read_to_string(codex_home.join(CONFIG_TOML_FILE)).expect("read config");

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

@@ -25,7 +25,9 @@ use crate::git_info::resolve_root_git_project_for_trust;
 use crate::model_family::ModelFamily;
 use crate::model_family::derive_default_model_family;
 use crate::model_family::find_family_for_model;
+use crate::model_provider_info::LMSTUDIO_OSS_PROVIDER_ID;
 use crate::model_provider_info::ModelProviderInfo;
+use crate::model_provider_info::OLLAMA_OSS_PROVIDER_ID;
 use crate::model_provider_info::built_in_model_providers;
 use crate::openai_model_info::get_model_info;
 use crate::project_doc::DEFAULT_PROJECT_DOC_FILENAME;
@@ -38,6 +40,7 @@ use codex_protocol::config_types::ForcedLoginMethod;
 use codex_protocol::config_types::ReasoningEffort;
 use codex_protocol::config_types::ReasoningSummary;
 use codex_protocol::config_types::SandboxMode;
+use codex_protocol::config_types::TrustLevel;
 use codex_protocol::config_types::Verbosity;
 use codex_rmcp_client::OAuthCredentialsStoreMode;
 use dirs::home_dir;
@@ -59,11 +62,11 @@ pub mod profile;
 pub mod types;
 
 #[cfg(target_os = "windows")]
-pub const OPENAI_DEFAULT_MODEL: &str = "gpt-5";
+pub const OPENAI_DEFAULT_MODEL: &str = "gpt-5.1";
 #[cfg(not(target_os = "windows"))]
-pub const OPENAI_DEFAULT_MODEL: &str = "gpt-5-codex";
-const OPENAI_DEFAULT_REVIEW_MODEL: &str = "gpt-5-codex";
-pub const GPT_5_CODEX_MEDIUM_MODEL: &str = "gpt-5-codex";
+pub const OPENAI_DEFAULT_MODEL: &str = "gpt-5.1-codex";
+const OPENAI_DEFAULT_REVIEW_MODEL: &str = "gpt-5.1-codex";
+pub const GPT_5_CODEX_MEDIUM_MODEL: &str = "gpt-5.1-codex";
 
 /// Maximum number of bytes of the documentation that will be embedded. Larger
 /// files are *silently truncated* to this size so we do not take up too much of
@@ -78,7 +81,7 @@ pub struct Config {
     /// Optional override of model selection.
     pub model: String,
 
-    /// Model used specifically for review sessions. Defaults to "gpt-5-codex".
+    /// Model used specifically for review sessions. Defaults to "gpt-5.1-codex".
     pub review_model: String,
 
     pub model_family: ModelFamily,
@@ -192,6 +195,9 @@ pub struct Config {
     /// Additional filenames to try when looking for project-level docs.
     pub project_doc_fallback_filenames: Vec<String>,
 
+    /// Token budget applied when storing tool/function outputs in the context manager.
+    pub tool_output_token_limit: Option<usize>,
+
     /// Directory containing all Codex state (defaults to `~/.codex` but can be
     /// overridden by the `CODEX_HOME` environment variable).
     pub codex_home: PathBuf,
@@ -382,15 +388,16 @@ fn ensure_no_inline_bearer_tokens(value: &TomlValue) -> std::io::Result<()> {
     Ok(())
 }
 
-pub(crate) fn set_project_trusted_inner(
+pub(crate) fn set_project_trust_level_inner(
     doc: &mut DocumentMut,
     project_path: &Path,
+    trust_level: TrustLevel,
 ) -> anyhow::Result<()> {
     // Ensure we render a human-friendly structure:
     //
     // [projects]
     // [projects."/path/to/project"]
-    // trust_level = "trusted"
+    // trust_level = "trusted" or "untrusted"
     //
     // rather than inline tables like:
     //
@@ -446,20 +453,66 @@ pub(crate) fn set_project_trusted_inner(
         return Err(anyhow::anyhow!("project table missing for {project_key}"));
     };
     proj_tbl.set_implicit(false);
-    proj_tbl["trust_level"] = toml_edit::value("trusted");
+    proj_tbl["trust_level"] = toml_edit::value(trust_level.to_string());
     Ok(())
 }
 
-/// Patch `CODEX_HOME/config.toml` project state.
+/// Patch `CODEX_HOME/config.toml` project state to set trust level.
 /// Use with caution.
-pub fn set_project_trusted(codex_home: &Path, project_path: &Path) -> anyhow::Result<()> {
+pub fn set_project_trust_level(
+    codex_home: &Path,
+    project_path: &Path,
+    trust_level: TrustLevel,
+) -> anyhow::Result<()> {
     use crate::config::edit::ConfigEditsBuilder;
 
     ConfigEditsBuilder::new(codex_home)
-        .set_project_trusted(project_path)
+        .set_project_trust_level(project_path, trust_level)
         .apply_blocking()
 }
 
+/// Save the default OSS provider preference to config.toml
+pub fn set_default_oss_provider(codex_home: &Path, provider: &str) -> std::io::Result<()> {
+    // Validate that the provider is one of the known OSS providers
+    match provider {
+        LMSTUDIO_OSS_PROVIDER_ID | OLLAMA_OSS_PROVIDER_ID => {
+            // Valid provider, continue
+        }
+        _ => {
+            return Err(std::io::Error::new(
+                std::io::ErrorKind::InvalidInput,
+                format!(
+                    "Invalid OSS provider '{provider}'. Must be one of: {LMSTUDIO_OSS_PROVIDER_ID}, {OLLAMA_OSS_PROVIDER_ID}"
+                ),
+            ));
+        }
+    }
+    let config_path = codex_home.join(CONFIG_TOML_FILE);
+
+    // Read existing config or create empty string if file doesn't exist
+    let content = match std::fs::read_to_string(&config_path) {
+        Ok(content) => content,
+        Err(e) if e.kind() == std::io::ErrorKind::NotFound => String::new(),
+        Err(e) => return Err(e),
+    };
+
+    // Parse as DocumentMut for editing while preserving structure
+    let mut doc = content.parse::<DocumentMut>().map_err(|e| {
+        std::io::Error::new(
+            std::io::ErrorKind::InvalidData,
+            format!("failed to parse config.toml: {e}"),
+        )
+    })?;
+
+    // Set the default_oss_provider at root level
+    use toml_edit::value;
+    doc["oss_provider"] = value(provider);
+
+    // Write the modified document back
+    std::fs::write(&config_path, doc.to_string())?;
+    Ok(())
+}
+
 /// Apply a single dotted-path override onto a TOML value.
 fn apply_toml_override(root: &mut TomlValue, path: &str, value: TomlValue) {
     use toml::value::Table;
@@ -586,6 +639,9 @@ pub struct ConfigToml {
     /// Ordered list of fallback filenames to look for when AGENTS.md is missing.
     pub project_doc_fallback_filenames: Option<Vec<String>>,
 
+    /// Token budget applied when storing tool/function outputs in the context manager.
+    pub tool_output_token_limit: Option<usize>,
+
     /// Profile to use from the `profiles` map.
     pub profile: Option<String>,
 
@@ -657,6 +713,8 @@ pub struct ConfigToml {
     pub experimental_use_rmcp_client: Option<bool>,
     pub experimental_use_freeform_apply_patch: Option<bool>,
     pub experimental_sandbox_command_assessment: Option<bool>,
+    /// Preferred OSS provider for local models, e.g. "lmstudio" or "ollama".
+    pub oss_provider: Option<String>,
 }
 
 impl From<ConfigToml> for UserSavedConfig {
@@ -686,15 +744,16 @@ impl From<ConfigToml> for UserSavedConfig {
 
 #[derive(Deserialize, Debug, Clone, PartialEq, Eq)]
 pub struct ProjectConfig {
-    pub trust_level: Option<String>,
+    pub trust_level: Option<TrustLevel>,
 }
 
 impl ProjectConfig {
     pub fn is_trusted(&self) -> bool {
-        match &self.trust_level {
-            Some(trust_level) => trust_level == "trusted",
-            None => false,
-        }
+        matches!(self.trust_level, Some(TrustLevel::Trusted))
+    }
+
+    pub fn is_untrusted(&self) -> bool {
+        matches!(self.trust_level, Some(TrustLevel::Untrusted))
     }
 }
 
@@ -735,9 +794,9 @@ impl ConfigToml {
             .or(profile_sandbox_mode)
             .or(self.sandbox_mode)
             .or_else(|| {
-                // if no sandbox_mode is set, but user has marked directory as trusted, use WorkspaceWrite
+                // if no sandbox_mode is set, but user has marked directory as trusted or untrusted, use WorkspaceWrite
                 self.get_active_project(resolved_cwd).and_then(|p| {
-                    if p.is_trusted() {
+                    if p.is_trusted() || p.is_untrusted() {
                         Some(SandboxMode::WorkspaceWrite)
                     } else {
                         None
@@ -844,6 +903,34 @@ pub struct ConfigOverrides {
     pub additional_writable_roots: Vec<PathBuf>,
 }
 
+/// Resolves the OSS provider from CLI override, profile config, or global config.
+/// Returns `None` if no provider is configured at any level.
+pub fn resolve_oss_provider(
+    explicit_provider: Option<&str>,
+    config_toml: &ConfigToml,
+    config_profile: Option<String>,
+) -> Option<String> {
+    if let Some(provider) = explicit_provider {
+        // Explicit provider specified (e.g., via --local-provider)
+        Some(provider.to_string())
+    } else {
+        // Check profile config first, then global config
+        let profile = config_toml.get_config_profile(config_profile).ok();
+        if let Some(profile) = &profile {
+            // Check if profile has an oss provider
+            if let Some(profile_oss_provider) = &profile.oss_provider {
+                Some(profile_oss_provider.clone())
+            }
+            // If not then check if the toml has an oss provider
+            else {
+                config_toml.oss_provider.clone()
+            }
+        } else {
+            config_toml.oss_provider.clone()
+        }
+    }
+}
+
 impl Config {
     /// Meant to be used exclusively for tests: `load_with_overrides()` should
     /// be used in all other cases.
@@ -958,6 +1045,9 @@ impl Config {
                 if active_project.is_trusted() {
                     // If no explicit approval policy is set, but we trust cwd, default to OnRequest
                     AskForApproval::OnRequest
+                } else if active_project.is_untrusted() {
+                    // If project is explicitly marked untrusted, require approval for non-safe commands
+                    AskForApproval::UnlessTrusted
                 } else {
                     AskForApproval::default()
                 }
@@ -1125,6 +1215,7 @@ impl Config {
                     }
                 })
                 .collect(),
+            tool_output_token_limit: cfg.tool_output_token_limit,
             codex_home,
             history,
             file_opener: cfg.file_opener.unwrap_or(UriBasedFileOpener::VsCode),
@@ -1229,6 +1320,16 @@ impl Config {
             Ok(Some(s))
         }
     }
+
+    pub fn set_windows_sandbox_globally(&mut self, value: bool) {
+        crate::safety::set_windows_sandbox_enabled(value);
+        if value {
+            self.features.enable(Feature::WindowsSandbox);
+        } else {
+            self.features.disable(Feature::WindowsSandbox);
+        }
+        self.forced_auto_mode_downgraded_on_windows = !value;
+    }
 }
 
 fn default_model() -> String {
@@ -1323,7 +1424,7 @@ persistence = "none"
     }
 
     #[test]
-    fn tui_config_missing_notifications_field_defaults_to_disabled() {
+    fn tui_config_missing_notifications_field_defaults_to_enabled() {
         let cfg = r#"
 [tui]
 "#;
@@ -1332,7 +1433,7 @@ persistence = "none"
             .expect("TUI config without notifications should succeed");
         let tui = parsed.tui.expect("config should include tui section");
 
-        assert_eq!(tui.notifications, Notifications::Enabled(false));
+        assert_eq!(tui.notifications, Notifications::Enabled(true));
     }
 
     #[test]
@@ -2532,7 +2633,7 @@ url = "https://example.com/mcp"
         let codex_home = TempDir::new()?;
 
         ConfigEditsBuilder::new(codex_home.path())
-            .set_model(Some("gpt-5-codex"), Some(ReasoningEffort::High))
+            .set_model(Some("gpt-5.1-codex"), Some(ReasoningEffort::High))
             .apply()
             .await?;
 
@@ -2540,7 +2641,7 @@ url = "https://example.com/mcp"
             tokio::fs::read_to_string(codex_home.path().join(CONFIG_TOML_FILE)).await?;
         let parsed: ConfigToml = toml::from_str(&serialized)?;
 
-        assert_eq!(parsed.model.as_deref(), Some("gpt-5-codex"));
+        assert_eq!(parsed.model.as_deref(), Some("gpt-5.1-codex"));
         assert_eq!(parsed.model_reasoning_effort, Some(ReasoningEffort::High));
 
         Ok(())
@@ -2554,7 +2655,7 @@ url = "https://example.com/mcp"
         tokio::fs::write(
             &config_path,
             r#"
-model = "gpt-5-codex"
+model = "gpt-5.1-codex"
 model_reasoning_effort = "medium"
 
 [profiles.dev]
@@ -2590,7 +2691,7 @@ model = "gpt-4.1"
 
         ConfigEditsBuilder::new(codex_home.path())
             .with_profile(Some("dev"))
-            .set_model(Some("gpt-5-codex"), Some(ReasoningEffort::Medium))
+            .set_model(Some("gpt-5.1-codex"), Some(ReasoningEffort::Medium))
             .apply()
             .await?;
 
@@ -2602,7 +2703,7 @@ model = "gpt-4.1"
             .get("dev")
             .expect("profile should be created");
 
-        assert_eq!(profile.model.as_deref(), Some("gpt-5-codex"));
+        assert_eq!(profile.model.as_deref(), Some("gpt-5.1-codex"));
         assert_eq!(
             profile.model_reasoning_effort,
             Some(ReasoningEffort::Medium)
@@ -2624,7 +2725,7 @@ model = "gpt-4"
 model_reasoning_effort = "medium"
 
 [profiles.prod]
-model = "gpt-5-codex"
+model = "gpt-5.1-codex"
 "#,
         )
         .await?;
@@ -2653,7 +2754,7 @@ model = "gpt-5-codex"
                 .profiles
                 .get("prod")
                 .and_then(|profile| profile.model.as_deref()),
-            Some("gpt-5-codex"),
+            Some("gpt-5.1-codex"),
         );
 
         Ok(())
@@ -2768,7 +2869,7 @@ model_provider = "openai"
 approval_policy = "on-failure"
 
 [profiles.gpt5]
-model = "gpt-5"
+model = "gpt-5.1"
 model_provider = "openai"
 approval_policy = "on-failure"
 model_reasoning_effort = "high"
@@ -2877,6 +2978,7 @@ model_verbosity = "high"
                 model_providers: fixture.model_provider_map.clone(),
                 project_doc_max_bytes: PROJECT_DOC_MAX_BYTES,
                 project_doc_fallback_filenames: Vec::new(),
+                tool_output_token_limit: None,
                 codex_home: fixture.codex_home(),
                 history: History::default(),
                 file_opener: UriBasedFileOpener::VsCode,
@@ -2948,6 +3050,7 @@ model_verbosity = "high"
             model_providers: fixture.model_provider_map.clone(),
             project_doc_max_bytes: PROJECT_DOC_MAX_BYTES,
             project_doc_fallback_filenames: Vec::new(),
+            tool_output_token_limit: None,
             codex_home: fixture.codex_home(),
             history: History::default(),
             file_opener: UriBasedFileOpener::VsCode,
@@ -3034,6 +3137,7 @@ model_verbosity = "high"
             model_providers: fixture.model_provider_map.clone(),
             project_doc_max_bytes: PROJECT_DOC_MAX_BYTES,
             project_doc_fallback_filenames: Vec::new(),
+            tool_output_token_limit: None,
             codex_home: fixture.codex_home(),
             history: History::default(),
             file_opener: UriBasedFileOpener::VsCode,
@@ -3084,9 +3188,9 @@ model_verbosity = "high"
             fixture.codex_home(),
         )?;
         let expected_gpt5_profile_config = Config {
-            model: "gpt-5".to_string(),
+            model: "gpt-5.1".to_string(),
             review_model: OPENAI_DEFAULT_REVIEW_MODEL.to_string(),
-            model_family: find_family_for_model("gpt-5").expect("known model slug"),
+            model_family: find_family_for_model("gpt-5.1").expect("known model slug"),
             model_context_window: Some(272_000),
             model_max_output_tokens: Some(128_000),
             model_auto_compact_token_limit: Some(244_800),
@@ -3106,6 +3210,7 @@ model_verbosity = "high"
             model_providers: fixture.model_provider_map.clone(),
             project_doc_max_bytes: PROJECT_DOC_MAX_BYTES,
             project_doc_fallback_filenames: Vec::new(),
+            tool_output_token_limit: None,
             codex_home: fixture.codex_home(),
             history: History::default(),
             file_opener: UriBasedFileOpener::VsCode,
@@ -3164,7 +3269,7 @@ model_verbosity = "high"
         let project_dir = Path::new("/some/path");
         let mut doc = DocumentMut::new();
 
-        set_project_trusted_inner(&mut doc, project_dir)?;
+        set_project_trust_level_inner(&mut doc, project_dir, TrustLevel::Trusted)?;
 
         let contents = doc.to_string();
 
@@ -3204,7 +3309,7 @@ trust_level = "trusted"
         let mut doc = initial.parse::<DocumentMut>()?;
 
         // Run the function; it should convert to explicit tables and set trusted
-        set_project_trusted_inner(&mut doc, project_dir)?;
+        set_project_trust_level_inner(&mut doc, project_dir, TrustLevel::Trusted)?;
 
         let contents = doc.to_string();
 
@@ -3231,7 +3336,7 @@ model = "foo""#;
 
         // Approve a new directory
         let new_project = Path::new("/Users/mbolin/code/codex2");
-        set_project_trusted_inner(&mut doc, new_project)?;
+        set_project_trust_level_inner(&mut doc, new_project, TrustLevel::Trusted)?;
 
         let contents = doc.to_string();
 
@@ -3254,6 +3359,201 @@ trust_level = "trusted"
 
         Ok(())
     }
+
+    #[test]
+    fn test_set_default_oss_provider() -> std::io::Result<()> {
+        let temp_dir = TempDir::new()?;
+        let codex_home = temp_dir.path();
+        let config_path = codex_home.join(CONFIG_TOML_FILE);
+
+        // Test setting valid provider on empty config
+        set_default_oss_provider(codex_home, OLLAMA_OSS_PROVIDER_ID)?;
+        let content = std::fs::read_to_string(&config_path)?;
+        assert!(content.contains("oss_provider = \"ollama\""));
+
+        // Test updating existing config
+        std::fs::write(&config_path, "model = \"gpt-4\"\n")?;
+        set_default_oss_provider(codex_home, LMSTUDIO_OSS_PROVIDER_ID)?;
+        let content = std::fs::read_to_string(&config_path)?;
+        assert!(content.contains("oss_provider = \"lmstudio\""));
+        assert!(content.contains("model = \"gpt-4\""));
+
+        // Test overwriting existing oss_provider
+        set_default_oss_provider(codex_home, OLLAMA_OSS_PROVIDER_ID)?;
+        let content = std::fs::read_to_string(&config_path)?;
+        assert!(content.contains("oss_provider = \"ollama\""));
+        assert!(!content.contains("oss_provider = \"lmstudio\""));
+
+        // Test invalid provider
+        let result = set_default_oss_provider(codex_home, "invalid_provider");
+        assert!(result.is_err());
+        let error = result.unwrap_err();
+        assert_eq!(error.kind(), std::io::ErrorKind::InvalidInput);
+        assert!(error.to_string().contains("Invalid OSS provider"));
+        assert!(error.to_string().contains("invalid_provider"));
+
+        Ok(())
+    }
+
+    #[test]
+    fn test_untrusted_project_gets_workspace_write_sandbox() -> anyhow::Result<()> {
+        let config_with_untrusted = r#"
+[projects."/tmp/test"]
+trust_level = "untrusted"
+"#;
+
+        let cfg = toml::from_str::<ConfigToml>(config_with_untrusted)
+            .expect("TOML deserialization should succeed");
+
+        let resolution = cfg.derive_sandbox_policy(None, None, &PathBuf::from("/tmp/test"));
+
+        // Verify that untrusted projects get WorkspaceWrite (or ReadOnly on Windows due to downgrade)
+        if cfg!(target_os = "windows") {
+            assert!(
+                matches!(resolution.policy, SandboxPolicy::ReadOnly),
+                "Expected ReadOnly on Windows, got {:?}",
+                resolution.policy
+            );
+        } else {
+            assert!(
+                matches!(resolution.policy, SandboxPolicy::WorkspaceWrite { .. }),
+                "Expected WorkspaceWrite for untrusted project, got {:?}",
+                resolution.policy
+            );
+        }
+
+        Ok(())
+    }
+
+    #[test]
+    fn test_resolve_oss_provider_explicit_override() {
+        let config_toml = ConfigToml::default();
+        let result = resolve_oss_provider(Some("custom-provider"), &config_toml, None);
+        assert_eq!(result, Some("custom-provider".to_string()));
+    }
+
+    #[test]
+    fn test_resolve_oss_provider_from_profile() {
+        let mut profiles = std::collections::HashMap::new();
+        let profile = ConfigProfile {
+            oss_provider: Some("profile-provider".to_string()),
+            ..Default::default()
+        };
+        profiles.insert("test-profile".to_string(), profile);
+        let config_toml = ConfigToml {
+            profiles,
+            ..Default::default()
+        };
+
+        let result = resolve_oss_provider(None, &config_toml, Some("test-profile".to_string()));
+        assert_eq!(result, Some("profile-provider".to_string()));
+    }
+
+    #[test]
+    fn test_resolve_oss_provider_from_global_config() {
+        let config_toml = ConfigToml {
+            oss_provider: Some("global-provider".to_string()),
+            ..Default::default()
+        };
+
+        let result = resolve_oss_provider(None, &config_toml, None);
+        assert_eq!(result, Some("global-provider".to_string()));
+    }
+
+    #[test]
+    fn test_resolve_oss_provider_profile_fallback_to_global() {
+        let mut profiles = std::collections::HashMap::new();
+        let profile = ConfigProfile::default(); // No oss_provider set
+        profiles.insert("test-profile".to_string(), profile);
+        let config_toml = ConfigToml {
+            oss_provider: Some("global-provider".to_string()),
+            profiles,
+            ..Default::default()
+        };
+
+        let result = resolve_oss_provider(None, &config_toml, Some("test-profile".to_string()));
+        assert_eq!(result, Some("global-provider".to_string()));
+    }
+
+    #[test]
+    fn test_resolve_oss_provider_none_when_not_configured() {
+        let config_toml = ConfigToml::default();
+        let result = resolve_oss_provider(None, &config_toml, None);
+        assert_eq!(result, None);
+    }
+
+    #[test]
+    fn test_resolve_oss_provider_explicit_overrides_all() {
+        let mut profiles = std::collections::HashMap::new();
+        let profile = ConfigProfile {
+            oss_provider: Some("profile-provider".to_string()),
+            ..Default::default()
+        };
+        profiles.insert("test-profile".to_string(), profile);
+        let config_toml = ConfigToml {
+            oss_provider: Some("global-provider".to_string()),
+            profiles,
+            ..Default::default()
+        };
+
+        let result = resolve_oss_provider(
+            Some("explicit-provider"),
+            &config_toml,
+            Some("test-profile".to_string()),
+        );
+        assert_eq!(result, Some("explicit-provider".to_string()));
+    }
+
+    #[test]
+    fn test_untrusted_project_gets_unless_trusted_approval_policy() -> std::io::Result<()> {
+        let codex_home = TempDir::new()?;
+        let test_project_dir = TempDir::new()?;
+        let test_path = test_project_dir.path();
+
+        let mut projects = std::collections::HashMap::new();
+        projects.insert(
+            test_path.to_string_lossy().to_string(),
+            ProjectConfig {
+                trust_level: Some(TrustLevel::Untrusted),
+            },
+        );
+
+        let cfg = ConfigToml {
+            projects: Some(projects),
+            ..Default::default()
+        };
+
+        let config = Config::load_from_base_config_with_overrides(
+            cfg,
+            ConfigOverrides {
+                cwd: Some(test_path.to_path_buf()),
+                ..Default::default()
+            },
+            codex_home.path().to_path_buf(),
+        )?;
+
+        // Verify that untrusted projects get UnlessTrusted approval policy
+        assert_eq!(
+            config.approval_policy,
+            AskForApproval::UnlessTrusted,
+            "Expected UnlessTrusted approval policy for untrusted project"
+        );
+
+        // Verify that untrusted projects still get WorkspaceWrite sandbox (or ReadOnly on Windows)
+        if cfg!(target_os = "windows") {
+            assert!(
+                matches!(config.sandbox_policy, SandboxPolicy::ReadOnly),
+                "Expected ReadOnly on Windows"
+            );
+        } else {
+            assert!(
+                matches!(config.sandbox_policy, SandboxPolicy::WorkspaceWrite { .. }),
+                "Expected WorkspaceWrite sandbox for untrusted project"
+            );
+        }
+
+        Ok(())
+    }
 }
 
 #[cfg(test)]

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

@@ -33,6 +33,7 @@ pub struct ConfigProfile {
     /// Optional feature toggles scoped to this profile.
     #[serde(default)]
     pub features: Option<crate::features::FeaturesToml>,
+    pub oss_provider: Option<String>,
 }
 
 impl From<ConfigProfile> for codex_app_server_protocol::Profile {

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

@@ -338,7 +338,7 @@ pub enum Notifications {
 
 impl Default for Notifications {
     fn default() -> Self {
-        Self::Enabled(false)
+        Self::Enabled(true)
     }
 }
 
@@ -346,7 +346,7 @@ impl Default for Notifications {
 #[derive(Deserialize, Debug, Clone, PartialEq, Default)]
 pub struct Tui {
     /// Enable desktop notifications from the TUI when the terminal is unfocused.
-    /// Defaults to `false`.
+    /// Defaults to `true`.
     #[serde(default)]
     pub notifications: Notifications,
 }
@@ -362,6 +362,8 @@ pub struct Notice {
     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

@@ -1,13 +1,15 @@
+use crate::codex::TurnContext;
+use crate::context_manager::normalize;
+use crate::truncate::TruncationPolicy;
+use crate::truncate::truncate_function_output_items_with_policy;
+use crate::truncate::truncate_text;
 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;
-
 /// Transcript of conversation history
 #[derive(Debug, Clone, Default)]
 pub(crate) struct ContextManager {
@@ -28,6 +30,10 @@ 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),
@@ -38,7 +44,7 @@ impl ContextManager {
     }
 
     /// `items` is ordered from oldest to newest.
-    pub(crate) fn record_items<I>(&mut self, items: I)
+    pub(crate) fn record_items<I>(&mut self, items: I, policy: TruncationPolicy)
     where
         I: IntoIterator,
         I::Item: std::ops::Deref<Target = ResponseItem>,
@@ -50,7 +56,7 @@ impl ContextManager {
                 continue;
             }
 
-            let processed = Self::process_item(&item);
+            let processed = self.process_item(item_ref, policy);
             self.items.push(processed);
         }
     }
@@ -68,6 +74,28 @@ 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
@@ -116,14 +144,14 @@ impl ContextManager {
         items.retain(|item| !matches!(item, ResponseItem::GhostSnapshot { .. }));
     }
 
-    fn process_item(item: &ResponseItem) -> ResponseItem {
+    fn process_item(&self, item: &ResponseItem, policy: TruncationPolicy) -> ResponseItem {
         match item {
             ResponseItem::FunctionCallOutput { call_id, output } => {
-                let truncated = format_output_for_model_body(output.content.as_str());
+                let truncated = truncate_text(output.content.as_str(), policy);
                 let truncated_items = output
                     .content_items
                     .as_ref()
-                    .map(|items| globally_truncate_function_output_items(items));
+                    .map(|items| truncate_function_output_items_with_policy(items, policy));
                 ResponseItem::FunctionCallOutput {
                     call_id: call_id.clone(),
                     output: FunctionCallOutputPayload {
@@ -134,7 +162,7 @@ impl ContextManager {
                 }
             }
             ResponseItem::CustomToolCallOutput { call_id, output } => {
-                let truncated = format_output_for_model_body(output);
+                let truncated = truncate_text(output, policy);
                 ResponseItem::CustomToolCallOutput {
                     call_id: call_id.clone(),
                     output: truncated,
@@ -146,6 +174,7 @@ impl ContextManager {
             | ResponseItem::FunctionCall { .. }
             | ResponseItem::WebSearchCall { .. }
             | ResponseItem::CustomToolCall { .. }
+            | ResponseItem::CompactionSummary { .. }
             | ResponseItem::GhostSnapshot { .. }
             | ResponseItem::Other => item.clone(),
         }
@@ -163,7 +192,8 @@ fn is_api_message(message: &ResponseItem) -> bool {
         | ResponseItem::CustomToolCallOutput { .. }
         | ResponseItem::LocalShellCall { .. }
         | ResponseItem::Reasoning { .. }
-        | ResponseItem::WebSearchCall { .. } => true,
+        | ResponseItem::WebSearchCall { .. }
+        | ResponseItem::CompactionSummary { .. } => true,
         ResponseItem::GhostSnapshot { .. } => false,
         ResponseItem::Other => false,
     }

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

@@ -1,8 +1,8 @@
 use super::*;
-use crate::context_manager::truncate;
+use crate::truncate;
+use crate::truncate::TruncationPolicy;
 use codex_git::GhostCommit;
 use codex_protocol::models::ContentItem;
-use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::LocalShellAction;
 use codex_protocol::models::LocalShellExecAction;
@@ -12,6 +12,9 @@ use codex_protocol::models::ReasoningItemReasoningSummary;
 use pretty_assertions::assert_eq;
 use regex_lite::Regex;
 
+const EXEC_FORMAT_MAX_LINES: usize = 256;
+const EXEC_FORMAT_MAX_BYTES: usize = 10_000;
+
 fn assistant_msg(text: &str) -> ResponseItem {
     ResponseItem::Message {
         id: None,
@@ -24,7 +27,9 @@ fn assistant_msg(text: &str) -> ResponseItem {
 
 fn create_history_with_items(items: Vec<ResponseItem>) -> ContextManager {
     let mut h = ContextManager::new();
-    h.record_items(items.iter());
+    // Use a generous but fixed token budget; tests only rely on truncation
+    // behavior, not on a specific model's token limit.
+    h.record_items(items.iter(), TruncationPolicy::Tokens(10_000));
     h
 }
 
@@ -54,6 +59,7 @@ fn reasoning_msg(text: &str) -> ResponseItem {
 #[test]
 fn filters_non_api_messages() {
     let mut h = ContextManager::default();
+    let policy = TruncationPolicy::Tokens(10_000);
     // System message is not API messages; Other is ignored.
     let system = ResponseItem::Message {
         id: None,
@@ -63,12 +69,12 @@ fn filters_non_api_messages() {
         }],
     };
     let reasoning = reasoning_msg("thinking...");
-    h.record_items([&system, &reasoning, &ResponseItem::Other]);
+    h.record_items([&system, &reasoning, &ResponseItem::Other], policy);
 
     // User and assistant should be retained.
     let u = user_msg("hi");
     let a = assistant_msg("hello");
-    h.record_items([&u, &a]);
+    h.record_items([&u, &a], policy);
 
     let items = h.contents();
     assert_eq!(
@@ -236,6 +242,9 @@ fn normalization_retains_local_shell_outputs() {
 #[test]
 fn record_items_truncates_function_call_output_content() {
     let mut history = ContextManager::new();
+    // Any reasonably small token budget works; the test only cares that
+    // truncation happens and the marker is present.
+    let policy = TruncationPolicy::Tokens(1_000);
     let long_line = "a very long line to trigger truncation\n";
     let long_output = long_line.repeat(2_500);
     let item = ResponseItem::FunctionCallOutput {
@@ -247,15 +256,20 @@ fn record_items_truncates_function_call_output_content() {
         },
     };
 
-    history.record_items([&item]);
+    history.record_items([&item], policy);
 
     assert_eq!(history.items.len(), 1);
     match &history.items[0] {
         ResponseItem::FunctionCallOutput { output, .. } => {
             assert_ne!(output.content, long_output);
             assert!(
-                output.content.starts_with("Total output lines:"),
-                "expected truncated summary, got {}",
+                output.content.contains("tokens truncated"),
+                "expected token-based truncation marker, got {}",
+                output.content
+            );
+            assert!(
+                output.content.contains("tokens truncated"),
+                "expected truncation marker, got {}",
                 output.content
             );
         }
@@ -266,6 +280,7 @@ fn record_items_truncates_function_call_output_content() {
 #[test]
 fn record_items_truncates_custom_tool_call_output_content() {
     let mut history = ContextManager::new();
+    let policy = TruncationPolicy::Tokens(1_000);
     let line = "custom output that is very long\n";
     let long_output = line.repeat(2_500);
     let item = ResponseItem::CustomToolCallOutput {
@@ -273,21 +288,48 @@ fn record_items_truncates_custom_tool_call_output_content() {
         output: long_output.clone(),
     };
 
-    history.record_items([&item]);
+    history.record_items([&item], policy);
 
     assert_eq!(history.items.len(), 1);
     match &history.items[0] {
         ResponseItem::CustomToolCallOutput { output, .. } => {
             assert_ne!(output, &long_output);
             assert!(
-                output.starts_with("Total output lines:"),
-                "expected truncated summary, got {output}"
+                output.contains("tokens truncated"),
+                "expected token-based truncation marker, got {output}"
+            );
+            assert!(
+                output.contains("tokens truncated") || output.contains("bytes truncated"),
+                "expected truncation marker, got {output}"
             );
         }
         other => panic!("unexpected history item: {other:?}"),
     }
 }
 
+#[test]
+fn record_items_respects_custom_token_limit() {
+    let mut history = ContextManager::new();
+    let policy = TruncationPolicy::Tokens(10);
+    let long_output = "tokenized content repeated many times ".repeat(200);
+    let item = ResponseItem::FunctionCallOutput {
+        call_id: "call-custom-limit".to_string(),
+        output: FunctionCallOutputPayload {
+            content: long_output,
+            success: Some(true),
+            ..Default::default()
+        },
+    };
+
+    history.record_items([&item], policy);
+
+    let stored = match &history.items[0] {
+        ResponseItem::FunctionCallOutput { output, .. } => output,
+        other => panic!("unexpected history item: {other:?}"),
+    };
+    assert!(stored.content.contains("tokens truncated"));
+}
+
 fn assert_truncated_message_matches(message: &str, line: &str, total_lines: usize) {
     let pattern = truncated_message_pattern(line, total_lines);
     let regex = Regex::new(&pattern).unwrap_or_else(|err| {
@@ -301,21 +343,22 @@ fn assert_truncated_message_matches(message: &str, line: &str, total_lines: usiz
         .expect("missing body capture")
         .as_str();
     assert!(
-        body.len() <= truncate::MODEL_FORMAT_MAX_BYTES,
+        body.len() <= EXEC_FORMAT_MAX_BYTES,
         "body exceeds byte limit: {} bytes",
         body.len()
     );
 }
 
 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 = EXEC_FORMAT_MAX_LINES / 2;
+    let tail_lines = EXEC_FORMAT_MAX_LINES - head_lines;
+    let head_take = head_lines.min(total_lines);
+    let tail_take = tail_lines.min(total_lines.saturating_sub(head_take));
     let omitted = total_lines.saturating_sub(head_take + tail_take);
     let escaped_line = regex_lite::escape(line);
     if omitted == 0 {
         return format!(
-            r"(?s)^Total output lines: {total_lines}\n\n(?P<body>{escaped_line}.*\n\[\.{{3}} output truncated to fit {max_bytes} bytes \.{{3}}]\n\n.*)$",
-            max_bytes = truncate::MODEL_FORMAT_MAX_BYTES,
+            r"(?s)^Total output lines: {total_lines}\n\n(?P<body>{escaped_line}.*\n\[\.{{3}} removed \d+ bytes to fit {EXEC_FORMAT_MAX_BYTES} byte limit \.{{3}}]\n\n.*)$",
         );
     }
     format!(
@@ -328,7 +371,7 @@ 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::truncate_with_line_bytes_budget(&large_error, EXEC_FORMAT_MAX_BYTES);
 
     let total_lines = large_error.lines().count();
     assert_truncated_message_matches(&truncated, line, total_lines);
@@ -337,13 +380,13 @@ 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 long_line = "a".repeat(EXEC_FORMAT_MAX_BYTES + 50);
+    let truncated = truncate::truncate_with_line_bytes_budget(&long_line, EXEC_FORMAT_MAX_BYTES);
 
     assert_ne!(truncated, long_line);
+    let removed_bytes = long_line.len().saturating_sub(EXEC_FORMAT_MAX_BYTES);
     let marker_line = format!(
-        "[... output truncated to fit {} bytes ...]",
-        truncate::MODEL_FORMAT_MAX_BYTES
+        "[... removed {removed_bytes} bytes to fit {EXEC_FORMAT_MAX_BYTES} byte limit ...]"
     );
     assert!(
         truncated.contains(&marker_line),
@@ -359,18 +402,21 @@ 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::truncate_with_line_bytes_budget(&content, EXEC_FORMAT_MAX_BYTES),
+        content
+    );
 }
 
 #[test]
 fn format_exec_output_reports_omitted_lines_and_keeps_head_and_tail() {
-    let total_lines = truncate::MODEL_FORMAT_MAX_LINES + 100;
+    let total_lines = EXEC_FORMAT_MAX_LINES + 100;
     let content: String = (0..total_lines)
         .map(|idx| format!("line-{idx}\n"))
         .collect();
 
-    let truncated = truncate::format_output_for_model_body(&content);
-    let omitted = total_lines - truncate::MODEL_FORMAT_MAX_LINES;
+    let truncated = truncate::truncate_with_line_bytes_budget(&content, EXEC_FORMAT_MAX_BYTES);
+    let omitted = total_lines - EXEC_FORMAT_MAX_LINES;
     let expected_marker = format!("[... omitted {omitted} of {total_lines} lines ...]");
 
     assert!(
@@ -391,99 +437,24 @@ fn format_exec_output_reports_omitted_lines_and_keeps_head_and_tail() {
 
 #[test]
 fn format_exec_output_prefers_line_marker_when_both_limits_exceeded() {
-    let total_lines = truncate::MODEL_FORMAT_MAX_LINES + 42;
+    let total_lines = EXEC_FORMAT_MAX_LINES + 42;
     let long_line = "x".repeat(256);
     let content: String = (0..total_lines)
         .map(|idx| format!("line-{idx}-{long_line}\n"))
         .collect();
 
-    let truncated = truncate::format_output_for_model_body(&content);
+    let truncated = truncate::truncate_with_line_bytes_budget(&content, EXEC_FORMAT_MAX_BYTES);
 
     assert!(
         truncated.contains("[... omitted 42 of 298 lines ...]"),
         "expected omitted marker when line count exceeds limit: {truncated}"
     );
     assert!(
-        !truncated.contains("output truncated to fit"),
+        !truncated.contains("byte limit"),
         "line omission marker should take precedence over byte marker: {truncated}"
     );
 }
 
-#[test]
-fn truncates_across_multiple_under_limit_texts_and_reports_omitted() {
-    // Arrange: several text items, none exceeding per-item limit, but total exceeds budget.
-    let budget = truncate::MODEL_FORMAT_MAX_BYTES;
-    let t1_len = (budget / 2).saturating_sub(10);
-    let t2_len = (budget / 2).saturating_sub(10);
-    let remaining_after_t1_t2 = budget.saturating_sub(t1_len + t2_len);
-    let t3_len = 50; // gets truncated to remaining_after_t1_t2
-    let t4_len = 5; // omitted
-    let t5_len = 7; // omitted
-
-    let t1 = "a".repeat(t1_len);
-    let t2 = "b".repeat(t2_len);
-    let t3 = "c".repeat(t3_len);
-    let t4 = "d".repeat(t4_len);
-    let t5 = "e".repeat(t5_len);
-
-    let item = ResponseItem::FunctionCallOutput {
-        call_id: "call-omit".to_string(),
-        output: FunctionCallOutputPayload {
-            content: "irrelevant".to_string(),
-            content_items: Some(vec![
-                FunctionCallOutputContentItem::InputText { text: t1 },
-                FunctionCallOutputContentItem::InputText { text: t2 },
-                FunctionCallOutputContentItem::InputImage {
-                    image_url: "img:mid".to_string(),
-                },
-                FunctionCallOutputContentItem::InputText { text: t3 },
-                FunctionCallOutputContentItem::InputText { text: t4 },
-                FunctionCallOutputContentItem::InputText { text: t5 },
-            ]),
-            success: Some(true),
-        },
-    };
-
-    let mut history = ContextManager::new();
-    history.record_items([&item]);
-    assert_eq!(history.items.len(), 1);
-    let json = serde_json::to_value(&history.items[0]).expect("serialize to json");
-
-    let output = json
-        .get("output")
-        .expect("output field")
-        .as_array()
-        .expect("array output");
-
-    // Expect: t1 (full), t2 (full), image, t3 (truncated), summary mentioning 2 omitted.
-    assert_eq!(output.len(), 5);
-
-    let first = output[0].as_object().expect("first obj");
-    assert_eq!(first.get("type").unwrap(), "input_text");
-    let first_text = first.get("text").unwrap().as_str().unwrap();
-    assert_eq!(first_text.len(), t1_len);
-
-    let second = output[1].as_object().expect("second obj");
-    assert_eq!(second.get("type").unwrap(), "input_text");
-    let second_text = second.get("text").unwrap().as_str().unwrap();
-    assert_eq!(second_text.len(), t2_len);
-
-    assert_eq!(
-        output[2],
-        serde_json::json!({"type": "input_image", "image_url": "img:mid"})
-    );
-
-    let fourth = output[3].as_object().expect("fourth obj");
-    assert_eq!(fourth.get("type").unwrap(), "input_text");
-    let fourth_text = fourth.get("text").unwrap().as_str().unwrap();
-    assert_eq!(fourth_text.len(), remaining_after_t1_t2);
-
-    let summary = output[4].as_object().expect("summary obj");
-    assert_eq!(summary.get("type").unwrap(), "input_text");
-    let summary_text = summary.get("text").unwrap().as_str().unwrap();
-    assert!(summary_text.contains("omitted 2 text items"));
-}
-
 //TODO(aibrahim): run CI in release mode.
 #[cfg(not(debug_assertions))]
 #[test]

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

@@ -1,6 +1,5 @@
 mod history;
 mod normalize;
-mod truncate;
 
+pub(crate) use crate::truncate::truncate_with_line_bytes_budget;
 pub(crate) use history::ContextManager;
-pub(crate) use truncate::format_output_for_model_body;

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

@@ -1,128 +0,0 @@
-use codex_protocol::models::FunctionCallOutputContentItem;
-use codex_utils_string::take_bytes_at_char_boundary;
-use codex_utils_string::take_last_bytes_at_char_boundary;
-
-// 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(crate) fn globally_truncate_function_output_items(
-    items: &[FunctionCallOutputContentItem],
-) -> Vec<FunctionCallOutputContentItem> {
-    let mut out: Vec<FunctionCallOutputContentItem> = Vec::with_capacity(items.len());
-    let mut remaining = MODEL_FORMAT_MAX_BYTES;
-    let mut omitted_text_items = 0usize;
-
-    for it in items {
-        match it {
-            FunctionCallOutputContentItem::InputText { text } => {
-                if remaining == 0 {
-                    omitted_text_items += 1;
-                    continue;
-                }
-
-                let len = text.len();
-                if len <= remaining {
-                    out.push(FunctionCallOutputContentItem::InputText { text: text.clone() });
-                    remaining -= len;
-                } else {
-                    let slice = take_bytes_at_char_boundary(text, remaining);
-                    if !slice.is_empty() {
-                        out.push(FunctionCallOutputContentItem::InputText {
-                            text: slice.to_string(),
-                        });
-                    }
-                    remaining = 0;
-                }
-            }
-            // todo(aibrahim): handle input images; resize
-            FunctionCallOutputContentItem::InputImage { image_url } => {
-                out.push(FunctionCallOutputContentItem::InputImage {
-                    image_url: image_url.clone(),
-                });
-            }
-        }
-    }
-
-    if omitted_text_items > 0 {
-        out.push(FunctionCallOutputContentItem::InputText {
-            text: format!("[omitted {omitted_text_items} text items ...]"),
-        });
-    }
-
-    out
-}
-
-pub(crate) fn format_output_for_model_body(content: &str) -> 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 {
-        return content.to_string();
-    }
-    let output = truncate_formatted_exec_output(content, total_lines);
-    format!("Total output lines: {total_lines}\n\n{output}")
-}
-
-fn truncate_formatted_exec_output(content: &str, total_lines: usize) -> String {
-    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 omitted = segments.len().saturating_sub(head_take + tail_take);
-
-    let head_slice_end: usize = segments
-        .iter()
-        .take(head_take)
-        .map(|segment| segment.len())
-        .sum();
-    let tail_slice_start: usize = if tail_take == 0 {
-        content.len()
-    } else {
-        content.len()
-            - segments
-                .iter()
-                .rev()
-                .take(tail_take)
-                .map(|segment| segment.len())
-                .sum::<usize>()
-    };
-    let head_slice = &content[..head_slice_end];
-    let tail_slice = &content[tail_slice_start..];
-    let truncated_by_bytes = content.len() > MODEL_FORMAT_MAX_BYTES;
-    // this is a bit wrong. We are counting metadata lines and not just shell output lines.
-    let marker = if omitted > 0 {
-        Some(format!(
-            "\n[... omitted {omitted} of {total_lines} lines ...]\n\n"
-        ))
-    } else if truncated_by_bytes {
-        Some(format!(
-            "\n[... output truncated to fit {MODEL_FORMAT_MAX_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 head_part = take_bytes_at_char_boundary(head_slice, head_budget);
-    let mut result = String::with_capacity(MODEL_FORMAT_MAX_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());
-    if remaining == 0 {
-        return result;
-    }
-
-    let tail_part = take_last_bytes_at_char_boundary(tail_slice, remaining);
-    result.push_str(tail_part);
-
-    result
-}

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

@@ -2,7 +2,8 @@ use crate::codex::ProcessedResponseItem;
 use crate::exec::ExecToolCallOutput;
 use crate::token_data::KnownPlan;
 use crate::token_data::PlanType;
-use crate::truncate::truncate_middle;
+use crate::truncate::TruncationPolicy;
+use crate::truncate::truncate_text;
 use chrono::DateTime;
 use chrono::Datelike;
 use chrono::Local;
@@ -461,7 +462,10 @@ pub fn get_error_message_ui(e: &CodexErr) -> String {
         _ => e.to_string(),
     };
 
-    truncate_middle(&message, ERROR_MESSAGE_UI_MAX_BYTES).0
+    truncate_text(
+        &message,
+        TruncationPolicy::Bytes(ERROR_MESSAGE_UI_MAX_BYTES),
+    )
 }
 
 #[cfg(test)]

codex-rs/core/src/exec.rs

@@ -189,16 +189,20 @@ async fn exec_windows_sandbox(
     };
 
     let sandbox_cwd = cwd.clone();
-    let logs_base_dir = find_codex_home().ok();
+    let codex_home = find_codex_home().map_err(|err| {
+        CodexErr::Io(io::Error::other(format!(
+            "windows sandbox: failed to resolve codex_home: {err}"
+        )))
+    })?;
     let spawn_res = tokio::task::spawn_blocking(move || {
         run_windows_sandbox_capture(
             policy_str,
             &sandbox_cwd,
+            codex_home.as_ref(),
             command,
             &cwd,
             env,
             timeout_ms,
-            logs_base_dir.as_deref(),
         )
     })
     .await;
@@ -532,8 +536,52 @@ async fn consume_truncated_output(
         }
     };
 
-    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);
 

codex-rs/core/src/exec_policy.rs

@@ -0,0 +1,365 @@
+use std::io::ErrorKind;
+use std::path::Path;
+use std::path::PathBuf;
+use std::sync::Arc;
+
+use crate::command_safety::is_dangerous_command::requires_initial_appoval;
+use codex_execpolicy2::Decision;
+use codex_execpolicy2::Evaluation;
+use codex_execpolicy2::Policy;
+use codex_execpolicy2::PolicyParser;
+use codex_protocol::protocol::AskForApproval;
+use codex_protocol::protocol::SandboxPolicy;
+use thiserror::Error;
+use tokio::fs;
+
+use crate::bash::parse_shell_lc_plain_commands;
+use crate::features::Feature;
+use crate::features::Features;
+use crate::sandboxing::SandboxPermissions;
+use crate::tools::sandboxing::ApprovalRequirement;
+
+const FORBIDDEN_REASON: &str = "execpolicy forbids this command";
+const PROMPT_REASON: &str = "execpolicy requires approval for this command";
+const POLICY_DIR_NAME: &str = "policy";
+const POLICY_EXTENSION: &str = "codexpolicy";
+
+#[derive(Debug, Error)]
+pub enum ExecPolicyError {
+    #[error("failed to read execpolicy files from {dir}: {source}")]
+    ReadDir {
+        dir: PathBuf,
+        source: std::io::Error,
+    },
+
+    #[error("failed to read execpolicy file {path}: {source}")]
+    ReadFile {
+        path: PathBuf,
+        source: std::io::Error,
+    },
+
+    #[error("failed to parse execpolicy file {path}: {source}")]
+    ParsePolicy {
+        path: String,
+        source: codex_execpolicy2::Error,
+    },
+}
+
+pub(crate) async fn exec_policy_for(
+    features: &Features,
+    codex_home: &Path,
+) -> Result<Arc<Policy>, ExecPolicyError> {
+    if !features.enabled(Feature::ExecPolicy) {
+        return Ok(Arc::new(Policy::empty()));
+    }
+
+    let policy_dir = codex_home.join(POLICY_DIR_NAME);
+    let policy_paths = collect_policy_files(&policy_dir).await?;
+
+    let mut parser = PolicyParser::new();
+    for policy_path in &policy_paths {
+        let contents =
+            fs::read_to_string(policy_path)
+                .await
+                .map_err(|source| ExecPolicyError::ReadFile {
+                    path: policy_path.clone(),
+                    source,
+                })?;
+        let identifier = policy_path.to_string_lossy().to_string();
+        parser
+            .parse(&identifier, &contents)
+            .map_err(|source| ExecPolicyError::ParsePolicy {
+                path: identifier,
+                source,
+            })?;
+    }
+
+    let policy = Arc::new(parser.build());
+    tracing::debug!(
+        "loaded execpolicy from {} files in {}",
+        policy_paths.len(),
+        policy_dir.display()
+    );
+
+    Ok(policy)
+}
+
+fn evaluate_with_policy(
+    policy: &Policy,
+    command: &[String],
+    approval_policy: AskForApproval,
+) -> Option<ApprovalRequirement> {
+    let commands = parse_shell_lc_plain_commands(command).unwrap_or_else(|| vec![command.to_vec()]);
+    let evaluation = policy.check_multiple(commands.iter());
+
+    match evaluation {
+        Evaluation::Match { decision, .. } => match decision {
+            Decision::Forbidden => Some(ApprovalRequirement::Forbidden {
+                reason: FORBIDDEN_REASON.to_string(),
+            }),
+            Decision::Prompt => {
+                let reason = PROMPT_REASON.to_string();
+                if matches!(approval_policy, AskForApproval::Never) {
+                    Some(ApprovalRequirement::Forbidden { reason })
+                } else {
+                    Some(ApprovalRequirement::NeedsApproval {
+                        reason: Some(reason),
+                    })
+                }
+            }
+            Decision::Allow => Some(ApprovalRequirement::Skip),
+        },
+        Evaluation::NoMatch { .. } => None,
+    }
+}
+
+pub(crate) fn create_approval_requirement_for_command(
+    policy: &Policy,
+    command: &[String],
+    approval_policy: AskForApproval,
+    sandbox_policy: &SandboxPolicy,
+    sandbox_permissions: SandboxPermissions,
+) -> ApprovalRequirement {
+    if let Some(requirement) = evaluate_with_policy(policy, command, approval_policy) {
+        return requirement;
+    }
+
+    if requires_initial_appoval(
+        approval_policy,
+        sandbox_policy,
+        command,
+        sandbox_permissions,
+    ) {
+        ApprovalRequirement::NeedsApproval { reason: None }
+    } else {
+        ApprovalRequirement::Skip
+    }
+}
+
+async fn collect_policy_files(dir: &Path) -> Result<Vec<PathBuf>, ExecPolicyError> {
+    let mut read_dir = match fs::read_dir(dir).await {
+        Ok(read_dir) => read_dir,
+        Err(err) if err.kind() == ErrorKind::NotFound => return Ok(Vec::new()),
+        Err(source) => {
+            return Err(ExecPolicyError::ReadDir {
+                dir: dir.to_path_buf(),
+                source,
+            });
+        }
+    };
+
+    let mut policy_paths = Vec::new();
+    while let Some(entry) =
+        read_dir
+            .next_entry()
+            .await
+            .map_err(|source| ExecPolicyError::ReadDir {
+                dir: dir.to_path_buf(),
+                source,
+            })?
+    {
+        let path = entry.path();
+        let file_type = entry
+            .file_type()
+            .await
+            .map_err(|source| ExecPolicyError::ReadDir {
+                dir: dir.to_path_buf(),
+                source,
+            })?;
+
+        if path
+            .extension()
+            .and_then(|ext| ext.to_str())
+            .is_some_and(|ext| ext == POLICY_EXTENSION)
+            && file_type.is_file()
+        {
+            policy_paths.push(path);
+        }
+    }
+
+    policy_paths.sort();
+
+    Ok(policy_paths)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::features::Feature;
+    use crate::features::Features;
+    use codex_protocol::protocol::AskForApproval;
+    use codex_protocol::protocol::SandboxPolicy;
+    use pretty_assertions::assert_eq;
+    use std::fs;
+    use tempfile::tempdir;
+
+    #[tokio::test]
+    async fn returns_empty_policy_when_feature_disabled() {
+        let mut features = Features::with_defaults();
+        features.disable(Feature::ExecPolicy);
+        let temp_dir = tempdir().expect("create temp dir");
+
+        let policy = exec_policy_for(&features, temp_dir.path())
+            .await
+            .expect("policy result");
+
+        let commands = [vec!["rm".to_string()]];
+        assert!(matches!(
+            policy.check_multiple(commands.iter()),
+            Evaluation::NoMatch { .. }
+        ));
+        assert!(!temp_dir.path().join(POLICY_DIR_NAME).exists());
+    }
+
+    #[tokio::test]
+    async fn collect_policy_files_returns_empty_when_dir_missing() {
+        let temp_dir = tempdir().expect("create temp dir");
+
+        let policy_dir = temp_dir.path().join(POLICY_DIR_NAME);
+        let files = collect_policy_files(&policy_dir)
+            .await
+            .expect("collect policy files");
+
+        assert!(files.is_empty());
+    }
+
+    #[tokio::test]
+    async fn loads_policies_from_policy_subdirectory() {
+        let temp_dir = tempdir().expect("create temp dir");
+        let policy_dir = temp_dir.path().join(POLICY_DIR_NAME);
+        fs::create_dir_all(&policy_dir).expect("create policy dir");
+        fs::write(
+            policy_dir.join("deny.codexpolicy"),
+            r#"prefix_rule(pattern=["rm"], decision="forbidden")"#,
+        )
+        .expect("write policy file");
+
+        let policy = exec_policy_for(&Features::with_defaults(), temp_dir.path())
+            .await
+            .expect("policy result");
+        let command = [vec!["rm".to_string()]];
+        assert!(matches!(
+            policy.check_multiple(command.iter()),
+            Evaluation::Match { .. }
+        ));
+    }
+
+    #[tokio::test]
+    async fn ignores_policies_outside_policy_dir() {
+        let temp_dir = tempdir().expect("create temp dir");
+        fs::write(
+            temp_dir.path().join("root.codexpolicy"),
+            r#"prefix_rule(pattern=["ls"], decision="prompt")"#,
+        )
+        .expect("write policy file");
+
+        let policy = exec_policy_for(&Features::with_defaults(), temp_dir.path())
+            .await
+            .expect("policy result");
+        let command = [vec!["ls".to_string()]];
+        assert!(matches!(
+            policy.check_multiple(command.iter()),
+            Evaluation::NoMatch { .. }
+        ));
+    }
+
+    #[test]
+    fn evaluates_bash_lc_inner_commands() {
+        let policy_src = r#"
+prefix_rule(pattern=["rm"], decision="forbidden")
+"#;
+        let mut parser = PolicyParser::new();
+        parser
+            .parse("test.codexpolicy", policy_src)
+            .expect("parse policy");
+        let policy = parser.build();
+
+        let forbidden_script = vec![
+            "bash".to_string(),
+            "-lc".to_string(),
+            "rm -rf /tmp".to_string(),
+        ];
+
+        let requirement =
+            evaluate_with_policy(&policy, &forbidden_script, AskForApproval::OnRequest)
+                .expect("expected match for forbidden command");
+
+        assert_eq!(
+            requirement,
+            ApprovalRequirement::Forbidden {
+                reason: FORBIDDEN_REASON.to_string()
+            }
+        );
+    }
+
+    #[test]
+    fn approval_requirement_prefers_execpolicy_match() {
+        let policy_src = r#"prefix_rule(pattern=["rm"], decision="prompt")"#;
+        let mut parser = PolicyParser::new();
+        parser
+            .parse("test.codexpolicy", policy_src)
+            .expect("parse policy");
+        let policy = parser.build();
+        let command = vec!["rm".to_string()];
+
+        let requirement = create_approval_requirement_for_command(
+            &policy,
+            &command,
+            AskForApproval::OnRequest,
+            &SandboxPolicy::DangerFullAccess,
+            SandboxPermissions::UseDefault,
+        );
+
+        assert_eq!(
+            requirement,
+            ApprovalRequirement::NeedsApproval {
+                reason: Some(PROMPT_REASON.to_string())
+            }
+        );
+    }
+
+    #[test]
+    fn approval_requirement_respects_approval_policy() {
+        let policy_src = r#"prefix_rule(pattern=["rm"], decision="prompt")"#;
+        let mut parser = PolicyParser::new();
+        parser
+            .parse("test.codexpolicy", policy_src)
+            .expect("parse policy");
+        let policy = parser.build();
+        let command = vec!["rm".to_string()];
+
+        let requirement = create_approval_requirement_for_command(
+            &policy,
+            &command,
+            AskForApproval::Never,
+            &SandboxPolicy::DangerFullAccess,
+            SandboxPermissions::UseDefault,
+        );
+
+        assert_eq!(
+            requirement,
+            ApprovalRequirement::Forbidden {
+                reason: PROMPT_REASON.to_string()
+            }
+        );
+    }
+
+    #[test]
+    fn approval_requirement_falls_back_to_heuristics() {
+        let command = vec!["python".to_string()];
+
+        let empty_policy = Policy::empty();
+        let requirement = create_approval_requirement_for_command(
+            &empty_policy,
+            &command,
+            AskForApproval::UnlessTrusted,
+            &SandboxPolicy::ReadOnly,
+            SandboxPermissions::UseDefault,
+        );
+
+        assert_eq!(
+            requirement,
+            ApprovalRequirement::NeedsApproval { reason: None }
+        );
+    }
+}

codex-rs/core/src/features.rs

@@ -27,6 +27,8 @@ pub enum Stage {
 /// Unique features toggled via configuration.
 #[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
 pub enum Feature {
+    /// Create a ghost commit at each turn.
+    GhostCommit,
     /// Use the single unified PTY-backed exec tool.
     UnifiedExec,
     /// Use the shell command tool that takes `command` as a single string of
@@ -40,12 +42,18 @@ pub enum Feature {
     ViewImageTool,
     /// Allow the model to request web searches.
     WebSearchRequest,
+    /// Gate the execpolicy enforcement for shell/unified exec.
+    ExecPolicy,
     /// Enable the model-based risk assessments for sandboxed commands.
     SandboxCommandAssessment,
-    /// Create a ghost commit at each turn.
-    GhostCommit,
     /// Enable Windows sandbox (restricted token) on Windows.
     WindowsSandbox,
+    /// Remote compaction enabled (only for ChatGPT auth)
+    RemoteCompaction,
+    /// Enable the default shell tool.
+    ShellTool,
+    /// Allow model to call multiple tools in parallel (only for models supporting it).
+    ParallelToolCalls,
 }
 
 impl Feature {
@@ -247,6 +255,14 @@ pub struct FeatureSpec {
 }
 
 pub const FEATURES: &[FeatureSpec] = &[
+    // Stable features.
+    FeatureSpec {
+        id: Feature::GhostCommit,
+        key: "undo",
+        stage: Stage::Stable,
+        default_enabled: true,
+    },
+    // Unstable features.
     FeatureSpec {
         id: Feature::UnifiedExec,
         key: "unified_exec",
@@ -283,22 +299,40 @@ pub const FEATURES: &[FeatureSpec] = &[
         stage: Stage::Stable,
         default_enabled: false,
     },
+    FeatureSpec {
+        id: Feature::ExecPolicy,
+        key: "exec_policy",
+        stage: Stage::Experimental,
+        default_enabled: true,
+    },
     FeatureSpec {
         id: Feature::SandboxCommandAssessment,
         key: "experimental_sandbox_command_assessment",
         stage: Stage::Experimental,
         default_enabled: false,
     },
-    FeatureSpec {
-        id: Feature::GhostCommit,
-        key: "ghost_commit",
-        stage: Stage::Experimental,
-        default_enabled: true,
-    },
     FeatureSpec {
         id: Feature::WindowsSandbox,
         key: "enable_experimental_windows_sandbox",
         stage: Stage::Experimental,
         default_enabled: false,
     },
+    FeatureSpec {
+        id: Feature::RemoteCompaction,
+        key: "remote_compaction",
+        stage: Stage::Experimental,
+        default_enabled: false,
+    },
+    FeatureSpec {
+        id: Feature::ParallelToolCalls,
+        key: "parallel",
+        stage: Stage::Experimental,
+        default_enabled: false,
+    },
+    FeatureSpec {
+        id: Feature::ShellTool,
+        key: "shell_tool",
+        stage: Stage::Stable,
+        default_enabled: true,
+    },
 ];

codex-rs/core/src/lib.rs

@@ -13,6 +13,7 @@ mod client;
 mod client_common;
 pub mod codex;
 mod codex_conversation;
+mod compact_remote;
 pub use codex_conversation::CodexConversation;
 mod codex_delegate;
 mod command_safety;
@@ -24,6 +25,7 @@ mod environment_context;
 pub mod error;
 pub mod exec;
 pub mod exec_env;
+mod exec_policy;
 pub mod features;
 mod flags;
 pub mod git_info;
@@ -34,14 +36,18 @@ mod mcp_tool_call;
 mod message_history;
 mod model_provider_info;
 pub mod parse_command;
+pub mod powershell;
 mod response_processing;
 pub mod sandboxing;
 pub mod token_data;
 mod truncate;
 mod unified_exec;
 mod user_instructions;
-pub use model_provider_info::BUILT_IN_OSS_MODEL_PROVIDER_ID;
+pub use model_provider_info::DEFAULT_LMSTUDIO_PORT;
+pub use model_provider_info::DEFAULT_OLLAMA_PORT;
+pub use model_provider_info::LMSTUDIO_OSS_PROVIDER_ID;
 pub use model_provider_info::ModelProviderInfo;
+pub use model_provider_info::OLLAMA_OSS_PROVIDER_ID;
 pub use model_provider_info::WireApi;
 pub use model_provider_info::built_in_model_providers;
 pub use model_provider_info::create_oss_provider_with_base_url;

codex-rs/core/src/model_family.rs

@@ -1,11 +1,17 @@
+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;
+use crate::truncate::TruncationPolicy;
 
 /// 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)]
@@ -27,6 +33,9 @@ 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,
 
@@ -53,8 +62,13 @@ 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,
+
+    pub truncation_policy: TruncationPolicy,
 }
 
 macro_rules! model_family {
@@ -76,7 +90,11 @@ macro_rules! model_family {
             effective_context_window_percent: 95,
             support_verbosity: false,
             shell_type: ConfigShellToolType::Default,
+            default_verbosity: None,
+            default_reasoning_effort: None,
+            truncation_policy: TruncationPolicy::Bytes(10_000),
         };
+
         // apply overrides
         $(
             mf.$key = $value;
@@ -118,9 +136,7 @@ pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
         model_family!(slug, "gpt-4o", needs_special_apply_patch_instructions: true)
     } else if slug.starts_with("gpt-3.5") {
         model_family!(slug, "gpt-3.5", needs_special_apply_patch_instructions: true)
-    } else if slug.starts_with("porcupine") {
-        model_family!(slug, "porcupine", shell_type: ConfigShellToolType::UnifiedExec)
-    } else if slug.starts_with("test-gpt-5-codex") {
+    } else if slug.starts_with("test-gpt-5") {
         model_family!(
             slug, slug,
             supports_reasoning_summaries: true,
@@ -134,6 +150,7 @@ pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
             ],
             supports_parallel_tool_calls: true,
             support_verbosity: true,
+            truncation_policy: TruncationPolicy::Tokens(10_000),
         )
 
     // Internal models.
@@ -149,19 +166,39 @@ pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
                 "list_dir".to_string(),
                 "read_file".to_string(),
             ],
+            shell_type: if cfg!(windows) { ConfigShellToolType::ShellCommand } else { ConfigShellToolType::Default },
             supports_parallel_tool_calls: true,
             support_verbosity: true,
+            truncation_policy: TruncationPolicy::Tokens(10_000),
         )
 
     // 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,
             reasoning_summary_format: ReasoningSummaryFormat::Experimental,
             base_instructions: GPT_5_CODEX_INSTRUCTIONS.to_string(),
             apply_patch_tool_type: Some(ApplyPatchToolType::Freeform),
+            shell_type: if cfg!(windows) { ConfigShellToolType::ShellCommand } else { ConfigShellToolType::Default },
+            supports_parallel_tool_calls: true,
             support_verbosity: false,
+            truncation_policy: TruncationPolicy::Tokens(10_000),
+        )
+    } 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),
+            truncation_policy: TruncationPolicy::Bytes(10_000),
+            supports_parallel_tool_calls: true,
         )
     } else if slug.starts_with("gpt-5") {
         model_family!(
@@ -169,6 +206,7 @@ pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
             supports_reasoning_summaries: true,
             needs_special_apply_patch_instructions: true,
             support_verbosity: true,
+            truncation_policy: TruncationPolicy::Bytes(10_000),
         )
     } else {
         None
@@ -189,5 +227,8 @@ pub fn derive_default_model_family(model: &str) -> ModelFamily {
         effective_context_window_percent: 95,
         support_verbosity: false,
         shell_type: ConfigShellToolType::Default,
+        default_verbosity: None,
+        default_reasoning_effort: None,
+        truncation_policy: TruncationPolicy::Bytes(10_000),
     }
 }

codex-rs/core/src/model_provider_info.rs

@@ -8,6 +8,7 @@
 use crate::CodexAuth;
 use crate::default_client::CodexHttpClient;
 use crate::default_client::CodexRequestBuilder;
+use crate::error::CodexErr;
 use codex_app_server_protocol::AuthMode;
 use serde::Deserialize;
 use serde::Serialize;
@@ -109,21 +110,7 @@ impl ModelProviderInfo {
         client: &'a CodexHttpClient,
         auth: &Option<CodexAuth>,
     ) -> crate::error::Result<CodexRequestBuilder> {
-        let effective_auth = if let Some(secret_key) = &self.experimental_bearer_token {
-            Some(CodexAuth::from_api_key(secret_key))
-        } else {
-            match self.api_key() {
-                Ok(Some(key)) => Some(CodexAuth::from_api_key(&key)),
-                Ok(None) => auth.clone(),
-                Err(err) => {
-                    if auth.is_some() {
-                        auth.clone()
-                    } else {
-                        return Err(err);
-                    }
-                }
-            }
-        };
+        let effective_auth = self.effective_auth(auth)?;
 
         let url = self.get_full_url(&effective_auth);
 
@@ -136,6 +123,51 @@ impl ModelProviderInfo {
         Ok(self.apply_http_headers(builder))
     }
 
+    pub async fn create_compact_request_builder<'a>(
+        &'a self,
+        client: &'a CodexHttpClient,
+        auth: &Option<CodexAuth>,
+    ) -> crate::error::Result<CodexRequestBuilder> {
+        if self.wire_api != WireApi::Responses {
+            return Err(CodexErr::UnsupportedOperation(
+                "Compaction endpoint requires Responses API providers".to_string(),
+            ));
+        }
+        let effective_auth = self.effective_auth(auth)?;
+
+        let url = self.get_compact_url(&effective_auth).ok_or_else(|| {
+            CodexErr::UnsupportedOperation(
+                "Compaction endpoint requires Responses API providers".to_string(),
+            )
+        })?;
+
+        let mut builder = client.post(url);
+
+        if let Some(auth) = effective_auth.as_ref() {
+            builder = builder.bearer_auth(auth.get_token().await?);
+        }
+
+        Ok(self.apply_http_headers(builder))
+    }
+
+    fn effective_auth(&self, auth: &Option<CodexAuth>) -> crate::error::Result<Option<CodexAuth>> {
+        if let Some(secret_key) = &self.experimental_bearer_token {
+            return Ok(Some(CodexAuth::from_api_key(secret_key)));
+        }
+
+        match self.api_key() {
+            Ok(Some(key)) => Ok(Some(CodexAuth::from_api_key(&key))),
+            Ok(None) => Ok(auth.clone()),
+            Err(err) => {
+                if auth.is_some() {
+                    Ok(auth.clone())
+                } else {
+                    Err(err)
+                }
+            }
+        }
+    }
+
     fn get_query_string(&self) -> String {
         self.query_params
             .as_ref()
@@ -173,6 +205,18 @@ impl ModelProviderInfo {
         }
     }
 
+    pub(crate) fn get_compact_url(&self, auth: &Option<CodexAuth>) -> Option<String> {
+        if self.wire_api != WireApi::Responses {
+            return None;
+        }
+        let full = self.get_full_url(auth);
+        if let Some((path, query)) = full.split_once('?') {
+            Some(format!("{path}/compact?{query}"))
+        } else {
+            Some(format!("{full}/compact"))
+        }
+    }
+
     pub(crate) fn is_azure_responses_endpoint(&self) -> bool {
         if self.wire_api != WireApi::Responses {
             return false;
@@ -258,9 +302,11 @@ impl ModelProviderInfo {
     }
 }
 
-const DEFAULT_OLLAMA_PORT: u32 = 11434;
+pub const DEFAULT_LMSTUDIO_PORT: u16 = 1234;
+pub const DEFAULT_OLLAMA_PORT: u16 = 11434;
 
-pub const BUILT_IN_OSS_MODEL_PROVIDER_ID: &str = "oss";
+pub const LMSTUDIO_OSS_PROVIDER_ID: &str = "lmstudio";
+pub const OLLAMA_OSS_PROVIDER_ID: &str = "ollama";
 
 /// Built-in default provider list.
 pub fn built_in_model_providers() -> HashMap<String, ModelProviderInfo> {
@@ -311,14 +357,21 @@ pub fn built_in_model_providers() -> HashMap<String, ModelProviderInfo> {
                 requires_openai_auth: true,
             },
         ),
-        (BUILT_IN_OSS_MODEL_PROVIDER_ID, create_oss_provider()),
+        (
+            OLLAMA_OSS_PROVIDER_ID,
+            create_oss_provider(DEFAULT_OLLAMA_PORT, WireApi::Chat),
+        ),
+        (
+            LMSTUDIO_OSS_PROVIDER_ID,
+            create_oss_provider(DEFAULT_LMSTUDIO_PORT, WireApi::Responses),
+        ),
     ]
     .into_iter()
     .map(|(k, v)| (k.to_string(), v))
     .collect()
 }
 
-pub fn create_oss_provider() -> ModelProviderInfo {
+pub fn create_oss_provider(default_provider_port: u16, wire_api: WireApi) -> ModelProviderInfo {
     // These CODEX_OSS_ environment variables are experimental: we may
     // switch to reading values from config.toml instead.
     let codex_oss_base_url = match std::env::var("CODEX_OSS_BASE_URL")
@@ -331,22 +384,21 @@ pub fn create_oss_provider() -> ModelProviderInfo {
             port = std::env::var("CODEX_OSS_PORT")
                 .ok()
                 .filter(|v| !v.trim().is_empty())
-                .and_then(|v| v.parse::<u32>().ok())
-                .unwrap_or(DEFAULT_OLLAMA_PORT)
+                .and_then(|v| v.parse::<u16>().ok())
+                .unwrap_or(default_provider_port)
         ),
     };
-
-    create_oss_provider_with_base_url(&codex_oss_base_url)
+    create_oss_provider_with_base_url(&codex_oss_base_url, wire_api)
 }
 
-pub fn create_oss_provider_with_base_url(base_url: &str) -> ModelProviderInfo {
+pub fn create_oss_provider_with_base_url(base_url: &str, wire_api: WireApi) -> ModelProviderInfo {
     ModelProviderInfo {
         name: "gpt-oss".into(),
         base_url: Some(base_url.into()),
         env_key: None,
         env_key_instructions: None,
         experimental_bearer_token: None,
-        wire_api: WireApi::Chat,
+        wire_api,
         query_params: None,
         http_headers: None,
         env_http_headers: 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

@@ -1,16 +1,22 @@
 use crate::bash::extract_bash_command;
 use crate::bash::try_parse_shell;
 use crate::bash::try_parse_word_only_commands_sequence;
+use crate::powershell::extract_powershell_command;
 use codex_protocol::parse_command::ParsedCommand;
 use shlex::split as shlex_split;
 use shlex::try_join as shlex_try_join;
 use std::path::PathBuf;
 
-fn shlex_join(tokens: &[String]) -> String {
+pub fn shlex_join(tokens: &[String]) -> String {
     shlex_try_join(tokens.iter().map(String::as_str))
         .unwrap_or_else(|_| "<command included NUL byte>".to_string())
 }
 
+/// Extracts the shell and script from a command, regardless of platform
+pub fn extract_shell_command(command: &[String]) -> Option<(&str, &str)> {
+    extract_bash_command(command).or_else(|| extract_powershell_command(command))
+}
+
 /// DO NOT REVIEW THIS CODE BY HAND
 /// This parsing code is quite complex and not easy to hand-modify.
 /// The easiest way to iterate is to add unit tests and have Codex fix the implementation.
@@ -877,6 +883,42 @@ mod tests {
             }],
         );
     }
+
+    #[test]
+    fn powershell_command_is_stripped() {
+        assert_parsed(
+            &vec_str(&["powershell", "-Command", "Get-ChildItem"]),
+            vec![ParsedCommand::Unknown {
+                cmd: "Get-ChildItem".to_string(),
+            }],
+        );
+    }
+
+    #[test]
+    fn pwsh_with_noprofile_and_c_alias_is_stripped() {
+        assert_parsed(
+            &vec_str(&["pwsh", "-NoProfile", "-c", "Write-Host hi"]),
+            vec![ParsedCommand::Unknown {
+                cmd: "Write-Host hi".to_string(),
+            }],
+        );
+    }
+
+    #[test]
+    fn powershell_with_path_is_stripped() {
+        let command = if cfg!(windows) {
+            "C:\\windows\\System32\\WindowsPowerShell\\v1.0\\powershell.exe"
+        } else {
+            "/usr/local/bin/powershell.exe"
+        };
+
+        assert_parsed(
+            &vec_str(&[command, "-NoProfile", "-c", "Write-Host hi"]),
+            vec![ParsedCommand::Unknown {
+                cmd: "Write-Host hi".to_string(),
+            }],
+        );
+    }
 }
 
 pub fn parse_command_impl(command: &[String]) -> Vec<ParsedCommand> {
@@ -884,6 +926,12 @@ pub fn parse_command_impl(command: &[String]) -> Vec<ParsedCommand> {
         return commands;
     }
 
+    if let Some((_, script)) = extract_powershell_command(command) {
+        return vec![ParsedCommand::Unknown {
+            cmd: script.to_string(),
+        }];
+    }
+
     let normalized = normalize_tokens(command);
 
     let parts = if contains_connectors(&normalized) {
@@ -1190,6 +1238,7 @@ fn parse_find_query_and_path(tail: &[String]) -> (Option<String>, Option<String>
 }
 
 fn parse_shell_lc_commands(original: &[String]) -> Option<Vec<ParsedCommand>> {
+    // Only handle bash/zsh here; PowerShell is stripped separately without bash parsing.
     let (_, script) = extract_bash_command(original)?;
 
     if let Some(tree) = try_parse_shell(script)

codex-rs/core/src/powershell.rs

@@ -0,0 +1,93 @@
+use std::path::PathBuf;
+
+use crate::shell::ShellType;
+use crate::shell::detect_shell_type;
+
+const POWERSHELL_FLAGS: &[&str] = &["-nologo", "-noprofile", "-command", "-c"];
+
+/// Extract the PowerShell script body from an invocation such as:
+///
+/// - ["pwsh", "-NoProfile", "-Command", "Get-ChildItem -Recurse | Select-String foo"]
+/// - ["powershell.exe", "-Command", "Write-Host hi"]
+/// - ["powershell", "-NoLogo", "-NoProfile", "-Command", "...script..."]
+///
+/// Returns (`shell`, `script`) when the first arg is a PowerShell executable and a
+/// `-Command` (or `-c`) flag is present followed by a script string.
+pub fn extract_powershell_command(command: &[String]) -> Option<(&str, &str)> {
+    if command.len() < 3 {
+        return None;
+    }
+
+    let shell = &command[0];
+    if detect_shell_type(&PathBuf::from(shell)) != Some(ShellType::PowerShell) {
+        return None;
+    }
+
+    // Find the first occurrence of -Command (accept common short alias -c as well)
+    let mut i = 1usize;
+    while i + 1 < command.len() {
+        let flag = &command[i];
+        // Reject unknown flags
+        if !POWERSHELL_FLAGS.contains(&flag.to_ascii_lowercase().as_str()) {
+            return None;
+        }
+        if flag.eq_ignore_ascii_case("-Command") || flag.eq_ignore_ascii_case("-c") {
+            let script = &command[i + 1];
+            return Some((shell, script.as_str()));
+        }
+        i += 1;
+    }
+    None
+}
+
+#[cfg(test)]
+mod tests {
+    use super::extract_powershell_command;
+
+    #[test]
+    fn extracts_basic_powershell_command() {
+        let cmd = vec![
+            "powershell".to_string(),
+            "-Command".to_string(),
+            "Write-Host hi".to_string(),
+        ];
+        let (_shell, script) = extract_powershell_command(&cmd).expect("extract");
+        assert_eq!(script, "Write-Host hi");
+    }
+
+    #[test]
+    fn extracts_lowercase_flags() {
+        let cmd = vec![
+            "powershell".to_string(),
+            "-nologo".to_string(),
+            "-command".to_string(),
+            "Write-Host hi".to_string(),
+        ];
+        let (_shell, script) = extract_powershell_command(&cmd).expect("extract");
+        assert_eq!(script, "Write-Host hi");
+    }
+
+    #[test]
+    fn extracts_full_path_powershell_command() {
+        let command = if cfg!(windows) {
+            "C:\\windows\\System32\\WindowsPowerShell\\v1.0\\powershell.exe".to_string()
+        } else {
+            "/usr/local/bin/powershell.exe".to_string()
+        };
+        let cmd = vec![command, "-Command".to_string(), "Write-Host hi".to_string()];
+        let (_shell, script) = extract_powershell_command(&cmd).expect("extract");
+        assert_eq!(script, "Write-Host hi");
+    }
+
+    #[test]
+    fn extracts_with_noprofile_and_alias() {
+        let cmd = vec![
+            "pwsh".to_string(),
+            "-NoProfile".to_string(),
+            "-c".to_string(),
+            "Get-ChildItem | Select-String foo".to_string(),
+        ];
+        let (_shell, script) = extract_powershell_command(&cmd).expect("extract");
+        assert_eq!(script, "Get-ChildItem | Select-String foo");
+    }
+}

codex-rs/core/src/response_processing.rs

@@ -13,92 +13,58 @@ pub(crate) async fn process_items(
     sess: &Session,
     turn_context: &TurnContext,
 ) -> (Vec<ResponseInputItem>, Vec<ResponseItem>) {
-    let mut items_to_record_in_conversation_history = Vec::<ResponseItem>::new();
+    let mut outputs_to_record = Vec::<ResponseItem>::new();
+    let mut new_inputs_to_record = Vec::<ResponseItem>::new();
     let mut responses = Vec::<ResponseInputItem>::new();
     for processed_response_item in processed_items {
         let crate::codex::ProcessedResponseItem { item, response } = processed_response_item;
-        match (&item, &response) {
-            (ResponseItem::Message { role, .. }, None) if role == "assistant" => {
-                // If the model returned a message, we need to record it.
-                items_to_record_in_conversation_history.push(item);
-            }
-            (
-                ResponseItem::LocalShellCall { .. },
-                Some(ResponseInputItem::FunctionCallOutput { call_id, output }),
-            ) => {
-                items_to_record_in_conversation_history.push(item);
-                items_to_record_in_conversation_history.push(ResponseItem::FunctionCallOutput {
+
+        if let Some(response) = &response {
+            responses.push(response.clone());
+        }
+
+        match response {
+            Some(ResponseInputItem::FunctionCallOutput { call_id, output }) => {
+                new_inputs_to_record.push(ResponseItem::FunctionCallOutput {
                     call_id: call_id.clone(),
                     output: output.clone(),
                 });
             }
-            (
-                ResponseItem::FunctionCall { .. },
-                Some(ResponseInputItem::FunctionCallOutput { call_id, output }),
-            ) => {
-                items_to_record_in_conversation_history.push(item);
-                items_to_record_in_conversation_history.push(ResponseItem::FunctionCallOutput {
+
+            Some(ResponseInputItem::CustomToolCallOutput { call_id, output }) => {
+                new_inputs_to_record.push(ResponseItem::CustomToolCallOutput {
                     call_id: call_id.clone(),
                     output: output.clone(),
                 });
             }
-            (
-                ResponseItem::CustomToolCall { .. },
-                Some(ResponseInputItem::CustomToolCallOutput { call_id, output }),
-            ) => {
-                items_to_record_in_conversation_history.push(item);
-                items_to_record_in_conversation_history.push(ResponseItem::CustomToolCallOutput {
-                    call_id: call_id.clone(),
-                    output: output.clone(),
-                });
-            }
-            (
-                ResponseItem::FunctionCall { .. },
-                Some(ResponseInputItem::McpToolCallOutput { call_id, result }),
-            ) => {
-                items_to_record_in_conversation_history.push(item);
+            Some(ResponseInputItem::McpToolCallOutput { call_id, result }) => {
                 let output = match result {
-                    Ok(call_tool_result) => FunctionCallOutputPayload::from(call_tool_result),
+                    Ok(call_tool_result) => FunctionCallOutputPayload::from(&call_tool_result),
                     Err(err) => FunctionCallOutputPayload {
                         content: err.clone(),
                         success: Some(false),
                         ..Default::default()
                     },
                 };
-                items_to_record_in_conversation_history.push(ResponseItem::FunctionCallOutput {
+                new_inputs_to_record.push(ResponseItem::FunctionCallOutput {
                     call_id: call_id.clone(),
                     output,
                 });
             }
-            (
-                ResponseItem::Reasoning {
-                    id,
-                    summary,
-                    content,
-                    encrypted_content,
-                },
-                None,
-            ) => {
-                items_to_record_in_conversation_history.push(ResponseItem::Reasoning {
-                    id: id.clone(),
-                    summary: summary.clone(),
-                    content: content.clone(),
-                    encrypted_content: encrypted_content.clone(),
-                });
-            }
+            None => {}
             _ => {
                 warn!("Unexpected response item: {item:?} with response: {response:?}");
             }
         };
-        if let Some(response) = response {
-            responses.push(response);
-        }
+
+        outputs_to_record.push(item);
     }
 
+    let all_items_to_record = [outputs_to_record, new_inputs_to_record].concat();
     // Only attempt to take the lock if there is something to record.
-    if !items_to_record_in_conversation_history.is_empty() {
-        sess.record_conversation_items(turn_context, &items_to_record_in_conversation_history)
+    if !all_items_to_record.is_empty() {
+        sess.record_conversation_items(turn_context, &all_items_to_record)
             .await;
     }
-    (responses, items_to_record_in_conversation_history)
+    (responses, all_items_to_record)
 }

codex-rs/core/src/rollout/policy.rs

@@ -27,7 +27,8 @@ pub(crate) fn should_persist_response_item(item: &ResponseItem) -> bool {
         | ResponseItem::CustomToolCall { .. }
         | ResponseItem::CustomToolCallOutput { .. }
         | ResponseItem::WebSearchCall { .. }
-        | ResponseItem::GhostSnapshot { .. } => true,
+        | ResponseItem::GhostSnapshot { .. }
+        | ResponseItem::CompactionSummary { .. } => true,
         ResponseItem::Other => false,
     }
 }
@@ -72,6 +73,8 @@ pub(crate) fn should_persist_event_msg(ev: &EventMsg) -> bool {
         | EventMsg::GetHistoryEntryResponse(_)
         | EventMsg::UndoStarted(_)
         | EventMsg::McpListToolsResponse(_)
+        | EventMsg::McpStartupUpdate(_)
+        | EventMsg::McpStartupComplete(_)
         | EventMsg::ListCustomPromptsResponse(_)
         | EventMsg::PlanUpdate(_)
         | EventMsg::ShutdownComplete

codex-rs/core/src/rollout/tests.rs

@@ -814,6 +814,7 @@ async fn test_tail_skips_trailing_non_responses() -> Result<()> {
         timestamp: format!("{ts}-compacted"),
         item: RolloutItem::Compacted(CompactedItem {
             message: "compacted".into(),
+            replacement_history: None,
         }),
     };
     writeln!(file, "{}", serde_json::to_string(&compacted_line)?)?;

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

@@ -26,6 +26,28 @@ use std::collections::HashMap;
 use std::path::Path;
 use std::path::PathBuf;
 
+#[derive(Clone, Copy, Debug, Eq, PartialEq)]
+pub enum SandboxPermissions {
+    UseDefault,
+    RequireEscalated,
+}
+
+impl SandboxPermissions {
+    pub fn requires_escalated_permissions(self) -> bool {
+        matches!(self, SandboxPermissions::RequireEscalated)
+    }
+}
+
+impl From<bool> for SandboxPermissions {
+    fn from(with_escalated_permissions: bool) -> Self {
+        if with_escalated_permissions {
+            SandboxPermissions::RequireEscalated
+        } else {
+            SandboxPermissions::UseDefault
+        }
+    }
+}
+
 #[derive(Clone, Debug)]
 pub struct CommandSpec {
     pub program: String,

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)]
@@ -36,7 +40,10 @@ impl Shell {
                     .file_name()
                     .map(|s| s.to_string_lossy().to_string())
             }
-            Shell::PowerShell(ps) => Some(ps.exe.clone()),
+            Shell::PowerShell(ps) => ps
+                .shell_path
+                .file_stem()
+                .map(|s| s.to_string_lossy().to_string()),
             Shell::Unknown => None,
         }
     }
@@ -47,10 +54,14 @@ impl Shell {
         match self {
             Shell::Zsh(ZshShell { shell_path, .. }) | Shell::Bash(BashShell { shell_path, .. }) => {
                 let arg = if use_login_shell { "-lc" } else { "-c" };
-                vec![shell_path.clone(), arg.to_string(), command.to_string()]
+                vec![
+                    shell_path.to_string_lossy().to_string(),
+                    arg.to_string(),
+                    command.to_string(),
+                ]
             }
             Shell::PowerShell(ps) => {
-                let mut args = vec![ps.exe.clone(), "-NoLogo".to_string()];
+                let mut args = vec![ps.shell_path.to_string_lossy().to_string()];
                 if !use_login_shell {
                     args.push("-NoProfile".to_string());
                 }
@@ -65,7 +76,7 @@ impl Shell {
 }
 
 #[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;
@@ -78,75 +89,181 @@ 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(if cfg!(windows) {
+                "C:\\windows\\System32\\WindowsPowerShell\\v1.0\\powershell.exe"
+            } else {
+                "/usr/local/bin/pwsh"
+            })),
+            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)]
@@ -156,6 +273,34 @@ 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")
@@ -164,292 +309,44 @@ 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/state/service.rs

@@ -8,9 +8,12 @@ use crate::unified_exec::UnifiedExecSessionManager;
 use crate::user_notification::UserNotifier;
 use codex_otel::otel_event_manager::OtelEventManager;
 use tokio::sync::Mutex;
+use tokio::sync::RwLock;
+use tokio_util::sync::CancellationToken;
 
 pub(crate) struct SessionServices {
-    pub(crate) mcp_connection_manager: McpConnectionManager,
+    pub(crate) mcp_connection_manager: Arc<RwLock<McpConnectionManager>>,
+    pub(crate) mcp_startup_cancellation_token: CancellationToken,
     pub(crate) unified_exec_manager: UnifiedExecSessionManager,
     pub(crate) notifier: UserNotifier,
     pub(crate) rollout: Mutex<Option<RolloutRecorder>>,

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

@@ -7,6 +7,7 @@ use crate::context_manager::ContextManager;
 use crate::protocol::RateLimitSnapshot;
 use crate::protocol::TokenUsage;
 use crate::protocol::TokenUsageInfo;
+use crate::truncate::TruncationPolicy;
 
 /// Persistent, session-scoped state previously stored directly on `Session`.
 pub(crate) struct SessionState {
@@ -18,20 +19,21 @@ pub(crate) struct SessionState {
 impl SessionState {
     /// Create a new session state mirroring previous `State::default()` semantics.
     pub(crate) fn new(session_configuration: SessionConfiguration) -> Self {
+        let history = ContextManager::new();
         Self {
             session_configuration,
-            history: ContextManager::new(),
+            history,
             latest_rate_limits: None,
         }
     }
 
     // History helpers
-    pub(crate) fn record_items<I>(&mut self, items: I)
+    pub(crate) fn record_items<I>(&mut self, items: I, policy: TruncationPolicy)
     where
         I: IntoIterator,
         I::Item: std::ops::Deref<Target = ResponseItem>,
     {
-        self.history.record_items(items)
+        self.history.record_items(items, policy);
     }
 
     pub(crate) fn clone_history(&self) -> ContextManager {
@@ -42,6 +44,10 @@ 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

@@ -1,15 +1,14 @@
 use std::sync::Arc;
 
-use async_trait::async_trait;
-use tokio_util::sync::CancellationToken;
-
-use crate::codex::TurnContext;
-use crate::compact;
-use crate::state::TaskKind;
-use codex_protocol::user_input::UserInput;
-
 use super::SessionTask;
 use super::SessionTaskContext;
+use crate::codex::TurnContext;
+use crate::features::Feature;
+use crate::state::TaskKind;
+use async_trait::async_trait;
+use codex_app_server_protocol::AuthMode;
+use codex_protocol::user_input::UserInput;
+use tokio_util::sync::CancellationToken;
 
 #[derive(Clone, Copy, Default)]
 pub(crate) struct CompactTask;
@@ -27,6 +26,17 @@ impl SessionTask for CompactTask {
         input: Vec<UserInput>,
         _cancellation_token: CancellationToken,
     ) -> Option<String> {
-        compact::run_compact_task(session.clone_session(), ctx, input).await
+        let session = session.clone_session();
+        if session
+            .services
+            .auth_manager
+            .auth()
+            .is_some_and(|auth| auth.mode == AuthMode::ChatGPT)
+            && session.enabled(Feature::RemoteCompaction).await
+        {
+            crate::compact_remote::run_remote_compact_task(session, ctx, input).await
+        } else {
+            crate::compact::run_compact_task(session, ctx, input).await
+        }
     }
 }

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

@@ -20,6 +20,7 @@ use crate::parse_command::parse_command;
 use crate::protocol::EventMsg;
 use crate::protocol::ExecCommandBeginEvent;
 use crate::protocol::ExecCommandEndEvent;
+use crate::protocol::ExecCommandSource;
 use crate::protocol::SandboxPolicy;
 use crate::protocol::TaskStartedEvent;
 use crate::sandboxing::ExecEnv;
@@ -64,30 +65,33 @@ impl SessionTask for UserShellCommandTask {
         // allows commands that use shell features (pipes, &&, redirects, etc.).
         // We do not source rc files or otherwise reformat the script.
         let use_login_shell = true;
-        let shell_invocation = session
+        let command = session
             .user_shell()
             .derive_exec_args(&self.command, use_login_shell);
 
         let call_id = Uuid::new_v4().to_string();
         let raw_command = self.command.clone();
+        let cwd = turn_context.cwd.clone();
 
-        let parsed_cmd = parse_command(&shell_invocation);
+        let parsed_cmd = parse_command(&command);
         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,
+                    turn_id: turn_context.sub_id.clone(),
+                    command: command.clone(),
+                    cwd: cwd.clone(),
+                    parsed_cmd: parsed_cmd.clone(),
+                    source: ExecCommandSource::UserShell,
+                    interaction_input: None,
                 }),
             )
             .await;
 
         let exec_env = ExecEnv {
-            command: shell_invocation,
-            cwd: turn_context.cwd.clone(),
+            command: command.clone(),
+            cwd: cwd.clone(),
             env: create_env(&turn_context.shell_environment_policy),
             timeout_ms: None,
             sandbox: SandboxType::None,
@@ -127,6 +131,12 @@ impl SessionTask for UserShellCommandTask {
                         turn_context.as_ref(),
                         EventMsg::ExecCommandEnd(ExecCommandEndEvent {
                             call_id,
+                            turn_id: turn_context.sub_id.clone(),
+                            command: command.clone(),
+                            cwd: cwd.clone(),
+                            parsed_cmd: parsed_cmd.clone(),
+                            source: ExecCommandSource::UserShell,
+                            interaction_input: None,
                             stdout: String::new(),
                             stderr: aborted_message.clone(),
                             aggregated_output: aborted_message.clone(),
@@ -143,6 +153,12 @@ impl SessionTask for UserShellCommandTask {
                         turn_context.as_ref(),
                         EventMsg::ExecCommandEnd(ExecCommandEndEvent {
                             call_id: call_id.clone(),
+                            turn_id: turn_context.sub_id.clone(),
+                            command: command.clone(),
+                            cwd: cwd.clone(),
+                            parsed_cmd: parsed_cmd.clone(),
+                            source: ExecCommandSource::UserShell,
+                            interaction_input: None,
                             stdout: output.stdout.text.clone(),
                             stderr: output.stderr.text.clone(),
                             aggregated_output: output.aggregated_output.text.clone(),
@@ -174,6 +190,12 @@ impl SessionTask for UserShellCommandTask {
                         turn_context.as_ref(),
                         EventMsg::ExecCommandEnd(ExecCommandEndEvent {
                             call_id,
+                            turn_id: turn_context.sub_id.clone(),
+                            command,
+                            cwd,
+                            parsed_cmd,
+                            source: ExecCommandSource::UserShell,
+                            interaction_input: None,
                             stdout: exec_output.stdout.text.clone(),
                             stderr: exec_output.stderr.text.clone(),
                             aggregated_output: exec_output.aggregated_output.text.clone(),

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

@@ -4,17 +4,13 @@ use crate::tools::TELEMETRY_PREVIEW_MAX_BYTES;
 use crate::tools::TELEMETRY_PREVIEW_MAX_LINES;
 use crate::tools::TELEMETRY_PREVIEW_TRUNCATION_NOTICE;
 use crate::turn_diff_tracker::TurnDiffTracker;
-use codex_otel::otel_event_manager::OtelEventManager;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::ResponseInputItem;
 use codex_protocol::models::ShellToolCallParams;
-use codex_protocol::protocol::FileChange;
 use codex_utils_string::take_bytes_at_char_boundary;
 use mcp_types::CallToolResult;
 use std::borrow::Cow;
-use std::collections::HashMap;
-use std::path::PathBuf;
 use std::sync::Arc;
 use tokio::sync::Mutex;
 
@@ -244,25 +240,3 @@ mod tests {
         assert_eq!(lines.last(), Some(&TELEMETRY_PREVIEW_TRUNCATION_NOTICE));
     }
 }
-
-#[derive(Clone, Debug)]
-#[allow(dead_code)]
-pub(crate) struct ExecCommandContext {
-    pub(crate) turn: Arc<TurnContext>,
-    pub(crate) call_id: String,
-    pub(crate) command_for_display: Vec<String>,
-    pub(crate) cwd: PathBuf,
-    pub(crate) apply_patch: Option<ApplyPatchCommandContext>,
-    pub(crate) tool_name: String,
-    pub(crate) otel_event_manager: OtelEventManager,
-    // TODO(abhisek-oai): Find a better way to track this.
-    // https://github.com/openai/codex/pull/2471/files#r2470352242
-    pub(crate) is_user_shell_command: bool,
-}
-
-#[derive(Clone, Debug)]
-#[allow(dead_code)]
-pub(crate) struct ApplyPatchCommandContext {
-    pub(crate) user_explicitly_approved_this_action: bool,
-    pub(crate) changes: HashMap<PathBuf, FileChange>,
-}

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

@@ -8,12 +8,14 @@ use crate::parse_command::parse_command;
 use crate::protocol::EventMsg;
 use crate::protocol::ExecCommandBeginEvent;
 use crate::protocol::ExecCommandEndEvent;
+use crate::protocol::ExecCommandSource;
 use crate::protocol::FileChange;
 use crate::protocol::PatchApplyBeginEvent;
 use crate::protocol::PatchApplyEndEvent;
 use crate::protocol::TurnDiffEvent;
 use crate::tools::context::SharedTurnDiffTracker;
 use crate::tools::sandboxing::ToolError;
+use codex_protocol::parse_command::ParsedCommand;
 use std::collections::HashMap;
 use std::path::Path;
 use std::path::PathBuf;
@@ -60,17 +62,21 @@ pub(crate) async fn emit_exec_command_begin(
     ctx: ToolEventCtx<'_>,
     command: &[String],
     cwd: &Path,
-    is_user_shell_command: bool,
+    parsed_cmd: &[ParsedCommand],
+    source: ExecCommandSource,
+    interaction_input: Option<String>,
 ) {
     ctx.session
         .send_event(
             ctx.turn,
             EventMsg::ExecCommandBegin(ExecCommandBeginEvent {
                 call_id: ctx.call_id.to_string(),
+                turn_id: ctx.turn.sub_id.clone(),
                 command: command.to_vec(),
                 cwd: cwd.to_path_buf(),
-                parsed_cmd: parse_command(command),
-                is_user_shell_command,
+                parsed_cmd: parsed_cmd.to_vec(),
+                source,
+                interaction_input,
             }),
         )
         .await;
@@ -80,27 +86,30 @@ pub(crate) enum ToolEmitter {
     Shell {
         command: Vec<String>,
         cwd: PathBuf,
-        is_user_shell_command: bool,
+        source: ExecCommandSource,
+        parsed_cmd: Vec<ParsedCommand>,
     },
     ApplyPatch {
         changes: HashMap<PathBuf, FileChange>,
         auto_approved: bool,
     },
     UnifiedExec {
-        command: String,
+        command: Vec<String>,
         cwd: PathBuf,
-        // True for `exec_command` and false for `write_stdin`.
-        #[allow(dead_code)]
-        is_startup_command: bool,
+        source: ExecCommandSource,
+        interaction_input: Option<String>,
+        parsed_cmd: Vec<ParsedCommand>,
     },
 }
 
 impl ToolEmitter {
-    pub fn shell(command: Vec<String>, cwd: PathBuf, is_user_shell_command: bool) -> Self {
+    pub fn shell(command: Vec<String>, cwd: PathBuf, source: ExecCommandSource) -> Self {
+        let parsed_cmd = parse_command(&command);
         Self::Shell {
             command,
             cwd,
-            is_user_shell_command,
+            source,
+            parsed_cmd,
         }
     }
 
@@ -111,11 +120,19 @@ impl ToolEmitter {
         }
     }
 
-    pub fn unified_exec(command: String, cwd: PathBuf, is_startup_command: bool) -> Self {
+    pub fn unified_exec(
+        command: &[String],
+        cwd: PathBuf,
+        source: ExecCommandSource,
+        interaction_input: Option<String>,
+    ) -> Self {
+        let parsed_cmd = parse_command(command);
         Self::UnifiedExec {
-            command,
+            command: command.to_vec(),
             cwd,
-            is_startup_command,
+            source,
+            interaction_input,
+            parsed_cmd,
         }
     }
 
@@ -125,45 +142,15 @@ impl ToolEmitter {
                 Self::Shell {
                     command,
                     cwd,
-                    is_user_shell_command,
+                    source,
+                    parsed_cmd,
                 },
-                ToolEventStage::Begin,
+                stage,
             ) => {
-                emit_exec_command_begin(ctx, command, cwd.as_path(), *is_user_shell_command).await;
-            }
-            (Self::Shell { .. }, ToolEventStage::Success(output)) => {
-                emit_exec_end(
+                emit_exec_stage(
                     ctx,
-                    output.stdout.text.clone(),
-                    output.stderr.text.clone(),
-                    output.aggregated_output.text.clone(),
-                    output.exit_code,
-                    output.duration,
-                    format_exec_output_str(&output),
-                )
-                .await;
-            }
-            (Self::Shell { .. }, ToolEventStage::Failure(ToolEventFailure::Output(output))) => {
-                emit_exec_end(
-                    ctx,
-                    output.stdout.text.clone(),
-                    output.stderr.text.clone(),
-                    output.aggregated_output.text.clone(),
-                    output.exit_code,
-                    output.duration,
-                    format_exec_output_str(&output),
-                )
-                .await;
-            }
-            (Self::Shell { .. }, ToolEventStage::Failure(ToolEventFailure::Message(message))) => {
-                emit_exec_end(
-                    ctx,
-                    String::new(),
-                    (*message).to_string(),
-                    (*message).to_string(),
-                    -1,
-                    Duration::ZERO,
-                    message.clone(),
+                    ExecCommandInput::new(command, cwd.as_path(), parsed_cmd, *source, None),
+                    stage,
                 )
                 .await;
             }
@@ -217,48 +204,26 @@ 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;
-            }
-            (Self::UnifiedExec { .. }, ToolEventStage::Success(output)) => {
-                emit_exec_end(
-                    ctx,
-                    output.stdout.text.clone(),
-                    output.stderr.text.clone(),
-                    output.aggregated_output.text.clone(),
-                    output.exit_code,
-                    output.duration,
-                    format_exec_output_str(&output),
-                )
-                .await;
-            }
             (
-                Self::UnifiedExec { .. },
-                ToolEventStage::Failure(ToolEventFailure::Output(output)),
+                Self::UnifiedExec {
+                    command,
+                    cwd,
+                    source,
+                    interaction_input,
+                    parsed_cmd,
+                },
+                stage,
             ) => {
-                emit_exec_end(
+                emit_exec_stage(
                     ctx,
-                    output.stdout.text.clone(),
-                    output.stderr.text.clone(),
-                    output.aggregated_output.text.clone(),
-                    output.exit_code,
-                    output.duration,
-                    format_exec_output_str(&output),
-                )
-                .await;
-            }
-            (
-                Self::UnifiedExec { .. },
-                ToolEventStage::Failure(ToolEventFailure::Message(message)),
-            ) => {
-                emit_exec_end(
-                    ctx,
-                    String::new(),
-                    (*message).to_string(),
-                    (*message).to_string(),
-                    -1,
-                    Duration::ZERO,
-                    message.clone(),
+                    ExecCommandInput::new(
+                        command,
+                        cwd.as_path(),
+                        parsed_cmd,
+                        *source,
+                        interaction_input.as_deref(),
+                    ),
+                    stage,
                 )
                 .await;
             }
@@ -317,26 +282,107 @@ impl ToolEmitter {
     }
 }
 
-async fn emit_exec_end(
-    ctx: ToolEventCtx<'_>,
+struct ExecCommandInput<'a> {
+    command: &'a [String],
+    cwd: &'a Path,
+    parsed_cmd: &'a [ParsedCommand],
+    source: ExecCommandSource,
+    interaction_input: Option<&'a str>,
+}
+
+impl<'a> ExecCommandInput<'a> {
+    fn new(
+        command: &'a [String],
+        cwd: &'a Path,
+        parsed_cmd: &'a [ParsedCommand],
+        source: ExecCommandSource,
+        interaction_input: Option<&'a str>,
+    ) -> Self {
+        Self {
+            command,
+            cwd,
+            parsed_cmd,
+            source,
+            interaction_input,
+        }
+    }
+}
+
+struct ExecCommandResult {
     stdout: String,
     stderr: String,
     aggregated_output: String,
     exit_code: i32,
     duration: Duration,
     formatted_output: String,
+}
+
+async fn emit_exec_stage(
+    ctx: ToolEventCtx<'_>,
+    exec_input: ExecCommandInput<'_>,
+    stage: ToolEventStage,
+) {
+    match stage {
+        ToolEventStage::Begin => {
+            emit_exec_command_begin(
+                ctx,
+                exec_input.command,
+                exec_input.cwd,
+                exec_input.parsed_cmd,
+                exec_input.source,
+                exec_input.interaction_input.map(str::to_owned),
+            )
+            .await;
+        }
+        ToolEventStage::Success(output)
+        | ToolEventStage::Failure(ToolEventFailure::Output(output)) => {
+            let exec_result = ExecCommandResult {
+                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),
+            };
+            emit_exec_end(ctx, exec_input, exec_result).await;
+        }
+        ToolEventStage::Failure(ToolEventFailure::Message(message)) => {
+            let text = message.to_string();
+            let exec_result = ExecCommandResult {
+                stdout: String::new(),
+                stderr: text.clone(),
+                aggregated_output: text.clone(),
+                exit_code: -1,
+                duration: Duration::ZERO,
+                formatted_output: text,
+            };
+            emit_exec_end(ctx, exec_input, exec_result).await;
+        }
+    }
+}
+
+async fn emit_exec_end(
+    ctx: ToolEventCtx<'_>,
+    exec_input: ExecCommandInput<'_>,
+    exec_result: ExecCommandResult,
 ) {
     ctx.session
         .send_event(
             ctx.turn,
             EventMsg::ExecCommandEnd(ExecCommandEndEvent {
                 call_id: ctx.call_id.to_string(),
-                stdout,
-                stderr,
-                aggregated_output,
-                exit_code,
-                duration,
-                formatted_output,
+                turn_id: ctx.turn.sub_id.clone(),
+                command: exec_input.command.to_vec(),
+                cwd: exec_input.cwd.to_path_buf(),
+                parsed_cmd: exec_input.parsed_cmd.to_vec(),
+                source: exec_input.source,
+                interaction_input: exec_input.interaction_input.map(str::to_owned),
+                stdout: exec_result.stdout,
+                stderr: exec_result.stderr,
+                aggregated_output: exec_result.aggregated_output,
+                exit_code: exec_result.exit_code,
+                duration: exec_result.duration,
+                formatted_output: exec_result.formatted_output,
             }),
         )
         .await;

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/mcp_resource.rs

@@ -287,6 +287,8 @@ async fn handle_list_resources(
             let resources = session
                 .services
                 .mcp_connection_manager
+                .read()
+                .await
                 .list_all_resources()
                 .await;
             Ok(ListResourcesPayload::from_all_servers(resources))
@@ -396,6 +398,8 @@ async fn handle_list_resource_templates(
             let templates = session
                 .services
                 .mcp_connection_manager
+                .read()
+                .await
                 .list_all_resource_templates()
                 .await;
             Ok(ListResourceTemplatesPayload::from_all_servers(templates))

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

@@ -9,7 +9,11 @@ use crate::apply_patch::convert_apply_patch_to_protocol;
 use crate::codex::TurnContext;
 use crate::exec::ExecParams;
 use crate::exec_env::create_env;
+use crate::exec_policy::create_approval_requirement_for_command;
 use crate::function_tool::FunctionCallError;
+use crate::is_safe_command::is_known_safe_command;
+use crate::protocol::ExecCommandSource;
+use crate::sandboxing::SandboxPermissions;
 use crate::tools::context::ToolInvocation;
 use crate::tools::context::ToolOutput;
 use crate::tools::context::ToolPayload;
@@ -77,6 +81,18 @@ 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,
@@ -103,7 +119,6 @@ impl ToolHandler for ShellHandler {
                     turn,
                     tracker,
                     call_id,
-                    false,
                 )
                 .await
             }
@@ -116,7 +131,6 @@ impl ToolHandler for ShellHandler {
                     turn,
                     tracker,
                     call_id,
-                    true,
                 )
                 .await
             }
@@ -164,7 +178,6 @@ impl ToolHandler for ShellCommandHandler {
             turn,
             tracker,
             call_id,
-            false,
         )
         .await
     }
@@ -178,7 +191,6 @@ impl ShellHandler {
         turn: Arc<TurnContext>,
         tracker: crate::tools::context::SharedTurnDiffTracker,
         call_id: String,
-        is_user_shell_command: bool,
     ) -> Result<ToolOutput, FunctionCallError> {
         // Approval policy guard for explicit escalation in non-OnRequest modes.
         if exec_params.with_escalated_permissions.unwrap_or(false)
@@ -271,12 +283,9 @@ impl ShellHandler {
             }
         }
 
-        // Regular shell execution path.
-        let emitter = ToolEmitter::shell(
-            exec_params.command.clone(),
-            exec_params.cwd.clone(),
-            is_user_shell_command,
-        );
+        let source = ExecCommandSource::Agent;
+        let emitter =
+            ToolEmitter::shell(exec_params.command.clone(), exec_params.cwd.clone(), source);
         let event_ctx = ToolEventCtx::new(session.as_ref(), turn.as_ref(), &call_id, None);
         emitter.begin(event_ctx).await;
 
@@ -287,6 +296,13 @@ impl ShellHandler {
             env: exec_params.env.clone(),
             with_escalated_permissions: exec_params.with_escalated_permissions,
             justification: exec_params.justification.clone(),
+            approval_requirement: create_approval_requirement_for_command(
+                &turn.exec_policy,
+                &exec_params.command,
+                turn.approval_policy,
+                &turn.sandbox_policy,
+                SandboxPermissions::from(exec_params.with_escalated_permissions.unwrap_or(false)),
+            ),
         };
         let mut orchestrator = ToolOrchestrator::new();
         let mut runtime = ShellRuntime::new();
@@ -311,8 +327,11 @@ 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;
 
@@ -322,27 +341,19 @@ mod tests {
     #[test]
     fn commands_generated_by_shell_command_handler_can_be_matched_by_is_known_safe_command() {
         let bash_shell = Shell::Bash(BashShell {
-            shell_path: "/bin/bash".to_string(),
-            bashrc_path: "/home/user/.bashrc".to_string(),
+            shell_path: PathBuf::from("/bin/bash"),
         });
         assert_safe(&bash_shell, "ls -la");
 
         let zsh_shell = Shell::Zsh(ZshShell {
-            shell_path: "/bin/zsh".to_string(),
-            zshrc_path: "/home/user/.zshrc".to_string(),
+            shell_path: PathBuf::from("/bin/zsh"),
         });
         assert_safe(&zsh_shell, "ls -la");
 
-        #[cfg(target_os = "windows")]
-        {
-            use crate::shell::PowerShellConfig;
-
-            let powershell = Shell::PowerShell(PowerShellConfig {
-                exe: "pwsh.exe".to_string(),
-                bash_exe_fallback: None,
-            });
-            assert_safe(&powershell, "ls -Name");
-        }
+        let powershell = Shell::PowerShell(PowerShellConfig {
+            shell_path: PathBuf::from("pwsh.exe"),
+        });
+        assert_safe(&powershell, "ls -Name");
     }
 
     fn assert_safe(shell: &Shell, command: &str) {

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

@@ -1,12 +1,12 @@
 use std::path::PathBuf;
 
-use async_trait::async_trait;
-use serde::Deserialize;
-
 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::ExecCommandSource;
 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;
@@ -20,6 +20,8 @@ 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;
 
@@ -32,8 +34,8 @@ struct ExecCommandArgs {
     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)]
@@ -47,12 +49,20 @@ 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()
 }
@@ -74,6 +84,20 @@ 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,
@@ -104,15 +128,15 @@ impl ToolHandler for UnifiedExecHandler {
                         "failed to parse exec_command arguments: {err:?}"
                     ))
                 })?;
+
+                let command = get_command(&args);
                 let ExecCommandArgs {
-                    cmd,
                     workdir,
-                    shell,
-                    login,
                     yield_time_ms,
                     max_output_tokens,
                     with_escalated_permissions,
                     justification,
+                    ..
                 } = args;
 
                 if with_escalated_permissions.unwrap_or(false)
@@ -139,15 +163,18 @@ impl ToolHandler for UnifiedExecHandler {
                     &context.call_id,
                     None,
                 );
-                let emitter = ToolEmitter::unified_exec(cmd.clone(), cwd.clone(), true);
+                let emitter = ToolEmitter::unified_exec(
+                    &command,
+                    cwd.clone(),
+                    ExecCommandSource::UnifiedExecStartup,
+                    None,
+                );
                 emitter.emit(event_ctx, ToolEventStage::Begin).await;
 
                 manager
                     .exec_command(
                         ExecCommandRequest {
-                            command: &cmd,
-                            shell: &shell,
-                            login,
+                            command,
                             yield_time_ms,
                             max_output_tokens,
                             workdir,
@@ -169,6 +196,7 @@ impl ToolHandler for UnifiedExecHandler {
                 })?;
                 manager
                     .write_stdin(WriteStdinRequest {
+                        call_id: &call_id,
                         session_id: args.session_id,
                         input: &args.chars,
                         yield_time_ms: args.yield_time_ms,
@@ -208,6 +236,11 @@ impl ToolHandler for UnifiedExecHandler {
     }
 }
 
+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 format_response(response: &UnifiedExecResponse) -> String {
     let mut sections = Vec::new();
 

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

@@ -9,7 +9,7 @@ pub mod runtimes;
 pub mod sandboxing;
 pub mod spec;
 
-use crate::context_manager::format_output_for_model_body;
+use crate::context_manager::truncate_with_line_bytes_budget;
 use crate::exec::ExecToolCallOutput;
 pub use router::ToolRouter;
 use serde::Serialize;
@@ -20,6 +20,9 @@ pub(crate) const TELEMETRY_PREVIEW_MAX_LINES: usize = 64; // lines
 pub(crate) const TELEMETRY_PREVIEW_TRUNCATION_NOTICE: &str =
     "[... telemetry preview truncated ...]";
 
+// TODO(aibrahim): migrate shell tool to use truncate text and respect config value
+const SHELL_OUTPUT_MAX_BYTES: usize = 10_000;
+
 /// Format the combined exec output for sending back to the model.
 /// Includes exit code and duration metadata; truncates large bodies safely.
 pub fn format_exec_output_for_model(exec_output: &ExecToolCallOutput) -> String {
@@ -75,5 +78,5 @@ pub fn format_exec_output_str(exec_output: &ExecToolCallOutput) -> String {
     };
 
     // Truncate for model consumption before serialization.
-    format_output_for_model_body(&body)
+    truncate_with_line_bytes_budget(&body, SHELL_OUTPUT_MAX_BYTES)
 }

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

@@ -11,11 +11,13 @@ use crate::error::get_error_message_ui;
 use crate::exec::ExecToolCallOutput;
 use crate::sandboxing::SandboxManager;
 use crate::tools::sandboxing::ApprovalCtx;
+use crate::tools::sandboxing::ApprovalRequirement;
 use crate::tools::sandboxing::ProvidesSandboxRetryData;
 use crate::tools::sandboxing::SandboxAttempt;
 use crate::tools::sandboxing::ToolCtx;
 use crate::tools::sandboxing::ToolError;
 use crate::tools::sandboxing::ToolRuntime;
+use crate::tools::sandboxing::default_approval_requirement;
 use codex_protocol::protocol::AskForApproval;
 use codex_protocol::protocol::ReviewDecision;
 
@@ -49,40 +51,52 @@ impl ToolOrchestrator {
         let otel_cfg = codex_otel::otel_event_manager::ToolDecisionSource::Config;
 
         // 1) Approval
-        let needs_initial_approval =
-            tool.wants_initial_approval(req, approval_policy, &turn_ctx.sandbox_policy);
         let mut already_approved = false;
 
-        if needs_initial_approval {
-            let mut risk = None;
-
-            if let Some(metadata) = req.sandbox_retry_data() {
-                risk = tool_ctx
-                    .session
-                    .assess_sandbox_command(turn_ctx, &tool_ctx.call_id, &metadata.command, None)
-                    .await;
+        let requirement = tool.approval_requirement(req).unwrap_or_else(|| {
+            default_approval_requirement(approval_policy, &turn_ctx.sandbox_policy)
+        });
+        match requirement {
+            ApprovalRequirement::Skip => {
+                otel.tool_decision(otel_tn, otel_ci, ReviewDecision::Approved, otel_cfg);
             }
+            ApprovalRequirement::Forbidden { reason } => {
+                return Err(ToolError::Rejected(reason));
+            }
+            ApprovalRequirement::NeedsApproval { reason } => {
+                let mut risk = None;
 
-            let approval_ctx = ApprovalCtx {
-                session: tool_ctx.session,
-                turn: turn_ctx,
-                call_id: &tool_ctx.call_id,
-                retry_reason: None,
-                risk,
-            };
-            let decision = tool.start_approval_async(req, approval_ctx).await;
-
-            otel.tool_decision(otel_tn, otel_ci, decision, otel_user.clone());
-
-            match decision {
-                ReviewDecision::Denied | ReviewDecision::Abort => {
-                    return Err(ToolError::Rejected("rejected by user".to_string()));
+                if let Some(metadata) = req.sandbox_retry_data() {
+                    risk = tool_ctx
+                        .session
+                        .assess_sandbox_command(
+                            turn_ctx,
+                            &tool_ctx.call_id,
+                            &metadata.command,
+                            None,
+                        )
+                        .await;
                 }
-                ReviewDecision::Approved | ReviewDecision::ApprovedForSession => {}
+
+                let approval_ctx = ApprovalCtx {
+                    session: tool_ctx.session,
+                    turn: turn_ctx,
+                    call_id: &tool_ctx.call_id,
+                    retry_reason: reason,
+                    risk,
+                };
+                let decision = tool.start_approval_async(req, approval_ctx).await;
+
+                otel.tool_decision(otel_tn, otel_ci, decision, otel_user.clone());
+
+                match decision {
+                    ReviewDecision::Denied | ReviewDecision::Abort => {
+                        return Err(ToolError::Rejected("rejected by user".to_string()));
+                    }
+                    ReviewDecision::Approved | ReviewDecision::ApprovedForSession => {}
+                }
+                already_approved = true;
             }
-            already_approved = true;
-        } else {
-            otel.tool_decision(otel_tn, otel_ci, ReviewDecision::Approved, otel_cfg);
         }
 
         // 2) First attempt under the selected sandbox.

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::trace!("waiting for tool gate");
-                        readiness.wait_ready().await;
-                        tracing::trace!("tool gate released");
                         let _guard = if supports_parallel {
                             Either::Left(lock.read().await)
                         } else {
@@ -117,7 +112,7 @@ impl ToolCallRuntime {
 
     fn abort_message(call: &ToolCall, secs: f32) -> String {
         match call.tool_name.as_str() {
-            "shell" | "container.exec" | "local_shell" | "unified_exec" => {
+            "shell" | "container.exec" | "local_shell" | "shell_command" | "unified_exec" => {
                 format!("Wall time: {secs:.1} seconds\naborted by user")
             }
             _ => format!("aborted by user after {secs:.1}s"),

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/router.rs

@@ -54,7 +54,7 @@ impl ToolRouter {
             .any(|config| config.spec.name() == tool_name)
     }
 
-    pub fn build_tool_call(
+    pub async fn build_tool_call(
         session: &Session,
         item: ResponseItem,
     ) -> Result<Option<ToolCall>, FunctionCallError> {
@@ -65,7 +65,7 @@ impl ToolRouter {
                 call_id,
                 ..
             } => {
-                if let Some((server, tool)) = session.parse_mcp_tool_name(&name) {
+                if let Some((server, tool)) = session.parse_mcp_tool_name(&name).await {
                     Ok(Some(ToolCall {
                         tool_name: name,
                         call_id,

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

@@ -4,13 +4,12 @@ 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::requires_initial_appoval;
 use crate::exec::ExecToolCallOutput;
-use crate::protocol::SandboxPolicy;
 use crate::sandboxing::execute_env;
 use crate::tools::runtimes::build_command_spec;
 use crate::tools::sandboxing::Approvable;
 use crate::tools::sandboxing::ApprovalCtx;
+use crate::tools::sandboxing::ApprovalRequirement;
 use crate::tools::sandboxing::ProvidesSandboxRetryData;
 use crate::tools::sandboxing::SandboxAttempt;
 use crate::tools::sandboxing::SandboxRetryData;
@@ -20,7 +19,6 @@ use crate::tools::sandboxing::ToolCtx;
 use crate::tools::sandboxing::ToolError;
 use crate::tools::sandboxing::ToolRuntime;
 use crate::tools::sandboxing::with_cached_approval;
-use codex_protocol::protocol::AskForApproval;
 use codex_protocol::protocol::ReviewDecision;
 use futures::future::BoxFuture;
 use std::path::PathBuf;
@@ -33,6 +31,7 @@ pub struct ShellRequest {
     pub env: std::collections::HashMap<String, String>,
     pub with_escalated_permissions: Option<bool>,
     pub justification: Option<String>,
+    pub approval_requirement: ApprovalRequirement,
 }
 
 impl ProvidesSandboxRetryData for ShellRequest {
@@ -114,18 +113,8 @@ impl Approvable<ShellRequest> for ShellRuntime {
         })
     }
 
-    fn wants_initial_approval(
-        &self,
-        req: &ShellRequest,
-        policy: AskForApproval,
-        sandbox_policy: &SandboxPolicy,
-    ) -> bool {
-        requires_initial_appoval(
-            policy,
-            sandbox_policy,
-            &req.command,
-            req.with_escalated_permissions.unwrap_or(false),
-        )
+    fn approval_requirement(&self, req: &ShellRequest) -> Option<ApprovalRequirement> {
+        Some(req.approval_requirement.clone())
     }
 
     fn wants_escalated_first_attempt(&self, req: &ShellRequest) -> bool {

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

@@ -1,4 +1,3 @@
-use crate::command_safety::is_dangerous_command::requires_initial_appoval;
 /*
 Runtime: unified exec
 
@@ -10,6 +9,7 @@ use crate::error::SandboxErr;
 use crate::tools::runtimes::build_command_spec;
 use crate::tools::sandboxing::Approvable;
 use crate::tools::sandboxing::ApprovalCtx;
+use crate::tools::sandboxing::ApprovalRequirement;
 use crate::tools::sandboxing::ProvidesSandboxRetryData;
 use crate::tools::sandboxing::SandboxAttempt;
 use crate::tools::sandboxing::SandboxRetryData;
@@ -22,9 +22,7 @@ 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;
@@ -36,6 +34,7 @@ pub struct UnifiedExecRequest {
     pub env: HashMap<String, String>,
     pub with_escalated_permissions: Option<bool>,
     pub justification: Option<String>,
+    pub approval_requirement: ApprovalRequirement,
 }
 
 impl ProvidesSandboxRetryData for UnifiedExecRequest {
@@ -65,6 +64,7 @@ impl UnifiedExecRequest {
         env: HashMap<String, String>,
         with_escalated_permissions: Option<bool>,
         justification: Option<String>,
+        approval_requirement: ApprovalRequirement,
     ) -> Self {
         Self {
             command,
@@ -72,6 +72,7 @@ impl UnifiedExecRequest {
             env,
             with_escalated_permissions,
             justification,
+            approval_requirement,
         }
     }
 }
@@ -129,18 +130,8 @@ impl Approvable<UnifiedExecRequest> for UnifiedExecRuntime<'_> {
         })
     }
 
-    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 approval_requirement(&self, req: &UnifiedExecRequest) -> Option<ApprovalRequirement> {
+        Some(req.approval_requirement.clone())
     }
 
     fn wants_escalated_first_attempt(&self, req: &UnifiedExecRequest) -> bool {

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

@@ -86,6 +86,37 @@ pub(crate) struct ApprovalCtx<'a> {
     pub risk: Option<SandboxCommandAssessment>,
 }
 
+// Specifies what tool orchestrator should do with a given tool call.
+#[derive(Clone, Debug, PartialEq, Eq)]
+pub(crate) enum ApprovalRequirement {
+    /// No approval required for this tool call
+    Skip,
+    /// Approval required for this tool call
+    NeedsApproval { reason: Option<String> },
+    /// Execution forbidden for this tool call
+    Forbidden { reason: String },
+}
+
+/// - Never, OnFailure: do not ask
+/// - OnRequest: ask unless sandbox policy is DangerFullAccess
+/// - UnlessTrusted: always ask
+pub(crate) fn default_approval_requirement(
+    policy: AskForApproval,
+    sandbox_policy: &SandboxPolicy,
+) -> ApprovalRequirement {
+    let needs_approval = match policy {
+        AskForApproval::Never | AskForApproval::OnFailure => false,
+        AskForApproval::OnRequest => !matches!(sandbox_policy, SandboxPolicy::DangerFullAccess),
+        AskForApproval::UnlessTrusted => true,
+    };
+
+    if needs_approval {
+        ApprovalRequirement::NeedsApproval { reason: None }
+    } else {
+        ApprovalRequirement::Skip
+    }
+}
+
 pub(crate) trait Approvable<Req> {
     type ApprovalKey: Hash + Eq + Clone + Debug + Serialize;
 
@@ -106,22 +137,11 @@ pub(crate) trait Approvable<Req> {
         matches!(policy, AskForApproval::Never)
     }
 
-    /// Decide whether an initial user approval should be requested before the
-    /// first attempt. Defaults to the orchestrator's behavior (pre‑refactor):
-    /// - Never, OnFailure: do not ask
-    /// - OnRequest: ask unless sandbox policy is DangerFullAccess
-    /// - UnlessTrusted: always ask
-    fn wants_initial_approval(
-        &self,
-        _req: &Req,
-        policy: AskForApproval,
-        sandbox_policy: &SandboxPolicy,
-    ) -> bool {
-        match policy {
-            AskForApproval::Never | AskForApproval::OnFailure => false,
-            AskForApproval::OnRequest => !matches!(sandbox_policy, SandboxPolicy::DangerFullAccess),
-            AskForApproval::UnlessTrusted => true,
-        }
+    /// Override the default approval requirement. Return `Some(_)` to specify
+    /// a custom requirement, or `None` to fall back to
+    /// policy-based default.
+    fn approval_requirement(&self, _req: &Req) -> Option<ApprovalRequirement> {
+        None
     }
 
     /// Decide we can request an approval for no-sandbox execution.