fix: retire Feature.RmcpClient

I noticed that `features: Features` was defined on `struct
SessionConfiguration`, which is commonly owned by `SessionState`, which
is in turn owned by `Session`.

Though I do not believe that `Features` should be allowed to be modified
over the course of a session (if the feature state is not invariant, it
makes it harder to reason about), which argues that it should live on
`Session` rather than `SessionState` or `SessionConfiguration`.

This PR moves `Features` to `Session` and updates all call sites. It
appears the only place we were mutating `Features` was:

- in tests
- the sub-agent config for a review task:


3ef76ff29d/codex-rs/core/src/tasks/review.rs (L86-L89)

Note this change also means it is no longer an `async` call to check the
state of a feature, eliminating the possibility of a
[TOCTTOU](https://en.wikipedia.org/wiki/Time-of-check_to_time-of-use)
error between checking the state of a feature and acting on it:


3ef76ff29d/codex-rs/core/src/codex.rs (L1069-L1076)

.github/codex/home/config.toml

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

.github/workflows/cargo-deny.yml

@@ -0,0 +1,26 @@
+name: cargo-deny
+
+on:
+  pull_request:
+  push:
+    branches:
+      - main
+
+jobs:
+  cargo-deny:
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: ./codex-rs
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v6
+
+      - name: Install Rust toolchain
+        uses: dtolnay/rust-toolchain@stable
+
+      - name: Run cargo-deny
+        uses: EmbarkStudios/cargo-deny-action@v1
+        with:
+          rust-version: stable
+          manifest-path: ./codex-rs/Cargo.toml

.github/workflows/ci.yml

@@ -12,7 +12,7 @@ jobs:
       NODE_OPTIONS: --max-old-space-size=4096
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v5
+        uses: actions/checkout@v6
 
       - name: Setup pnpm
         uses: pnpm/action-setup@v4

.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 +46,4 @@ 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,dependabot[bot],github-actions[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/codespell.yml

@@ -18,7 +18,7 @@ jobs:
 
     steps:
       - name: Checkout
-        uses: actions/checkout@v5
+        uses: actions/checkout@v6
       - name: Annotate locations with typos
         uses: codespell-project/codespell-problem-matcher@b80729f885d32f78a716c2f107b4db1025001c42 # v1
       - name: Codespell

.github/workflows/issue-deduplicator.yml

@@ -16,7 +16,7 @@ jobs:
     outputs:
       codex_output: ${{ steps.codex.outputs.final-message }}
     steps:
-      - uses: actions/checkout@v5
+      - uses: actions/checkout@v6
 
       - name: Prepare Codex inputs
         env:
@@ -46,7 +46,6 @@ jobs:
         with:
           openai-api-key: ${{ secrets.CODEX_OPENAI_API_KEY }}
           allow-users: "*"
-          model: gpt-5
           prompt: |
             You are an assistant that triages new GitHub issues by identifying potential duplicates.
 

.github/workflows/issue-labeler.yml

@@ -16,7 +16,7 @@ jobs:
     outputs:
       codex_output: ${{ steps.codex.outputs.final-message }}
     steps:
-      - uses: actions/checkout@v5
+      - uses: actions/checkout@v6
 
       - id: codex
         uses: openai/codex-action@main

.github/workflows/rust-ci.yml

@@ -17,7 +17,7 @@ jobs:
       codex: ${{ steps.detect.outputs.codex }}
       workflows: ${{ steps.detect.outputs.workflows }}
     steps:
-      - uses: actions/checkout@v5
+      - uses: actions/checkout@v6
         with:
           fetch-depth: 0
       - name: Detect changed paths (no external action)
@@ -56,7 +56,7 @@ jobs:
       run:
         working-directory: codex-rs
     steps:
-      - uses: actions/checkout@v5
+      - uses: actions/checkout@v6
       - uses: dtolnay/rust-toolchain@1.90
         with:
           components: rustfmt
@@ -74,7 +74,7 @@ jobs:
       run:
         working-directory: codex-rs
     steps:
-      - uses: actions/checkout@v5
+      - uses: actions/checkout@v6
       - uses: dtolnay/rust-toolchain@1.90
       - uses: taiki-e/install-action@44c6d64aa62cd779e873306675c7a58e86d6d532 # v2
         with:
@@ -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
 
@@ -147,12 +147,21 @@ jobs:
             profile: release
 
     steps:
-      - uses: actions/checkout@v5
+      - uses: actions/checkout@v6
       - uses: dtolnay/rust-toolchain@1.90
         with:
           targets: ${{ matrix.target }}
           components: clippy
 
+      - name: Compute lockfile hash
+        id: lockhash
+        working-directory: codex-rs
+        shell: bash
+        run: |
+          set -euo pipefail
+          echo "hash=$(sha256sum Cargo.lock | cut -d' ' -f1)" >> "$GITHUB_OUTPUT"
+          echo "toolchain_hash=$(sha256sum rust-toolchain.toml | cut -d' ' -f1)" >> "$GITHUB_OUTPUT"
+
       # Explicit cache restore: split cargo home vs target, so we can
       # avoid caching the large target dir on the gnu-dev job.
       - name: Restore cargo home cache
@@ -164,18 +173,20 @@ jobs:
             ~/.cargo/registry/index/
             ~/.cargo/registry/cache/
             ~/.cargo/git/db/
-          key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ hashFiles('codex-rs/rust-toolchain.toml') }}
+          key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ steps.lockhash.outputs.hash }}-${{ steps.lockhash.outputs.toolchain_hash }}
           restore-keys: |
             cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-
 
       # 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
@@ -188,15 +199,20 @@ 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:
           path: ${{ github.workspace }}/.sccache/
-          key: sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ github.run_id }}
+          key: sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ steps.lockhash.outputs.hash }}-${{ github.run_id }}
           restore-keys: |
-            sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-
+            sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ steps.lockhash.outputs.hash }}-
             sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-
 
       - if: ${{ matrix.target == 'x86_64-unknown-linux-musl' || matrix.target == 'aarch64-unknown-linux-musl'}}
@@ -271,23 +287,23 @@ jobs:
             ~/.cargo/registry/index/
             ~/.cargo/registry/cache/
             ~/.cargo/git/db/
-          key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ hashFiles('codex-rs/rust-toolchain.toml') }}
+          key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ steps.lockhash.outputs.hash }}-${{ steps.lockhash.outputs.toolchain_hash }}
 
       - 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:
           path: ${{ github.workspace }}/.sccache/
-          key: sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ github.run_id }}
+          key: sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ steps.lockhash.outputs.hash }}-${{ 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 +342,8 @@ 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
 
@@ -351,11 +368,20 @@ jobs:
             profile: dev
 
     steps:
-      - uses: actions/checkout@v5
+      - uses: actions/checkout@v6
       - uses: dtolnay/rust-toolchain@1.90
         with:
           targets: ${{ matrix.target }}
 
+      - name: Compute lockfile hash
+        id: lockhash
+        working-directory: codex-rs
+        shell: bash
+        run: |
+          set -euo pipefail
+          echo "hash=$(sha256sum Cargo.lock | cut -d' ' -f1)" >> "$GITHUB_OUTPUT"
+          echo "toolchain_hash=$(sha256sum rust-toolchain.toml | cut -d' ' -f1)" >> "$GITHUB_OUTPUT"
+
       - name: Restore cargo home cache
         id: cache_cargo_home_restore
         uses: actions/cache/restore@v4
@@ -365,17 +391,19 @@ jobs:
             ~/.cargo/registry/index/
             ~/.cargo/registry/cache/
             ~/.cargo/git/db/
-          key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ hashFiles('codex-rs/rust-toolchain.toml') }}
+          key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ steps.lockhash.outputs.hash }}-${{ steps.lockhash.outputs.toolchain_hash }}
           restore-keys: |
             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
@@ -388,15 +416,20 @@ 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:
           path: ${{ github.workspace }}/.sccache/
-          key: sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ github.run_id }}
+          key: sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ steps.lockhash.outputs.hash }}-${{ github.run_id }}
           restore-keys: |
-            sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-
+            sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ steps.lockhash.outputs.hash }}-
             sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-
 
       - uses: taiki-e/install-action@44c6d64aa62cd779e873306675c7a58e86d6d532 # v2
@@ -406,10 +439,10 @@ 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'
@@ -421,23 +454,23 @@ jobs:
             ~/.cargo/registry/index/
             ~/.cargo/registry/cache/
             ~/.cargo/git/db/
-          key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ hashFiles('codex-rs/rust-toolchain.toml') }}
+          key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ steps.lockhash.outputs.hash }}-${{ steps.lockhash.outputs.toolchain_hash }}
 
       - 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:
           path: ${{ github.workspace }}/.sccache/
-          key: sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ github.run_id }}
+          key: sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ steps.lockhash.outputs.hash }}-${{ github.run_id }}
 
       - name: sccache stats
-        if: always()
+        if: always() && env.USE_SCCACHE == 'true'
         continue-on-error: true
         run: sccache --show-stats || true
 
       - name: sccache summary
-        if: always()
+        if: always() && env.USE_SCCACHE == 'true'
         shell: bash
         run: |
           {

.github/workflows/rust-release.yml

@@ -19,7 +19,7 @@ jobs:
   tag-check:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v5
+      - uses: actions/checkout@v6
 
       - name: Validate tag matches Cargo.toml version
         shell: bash
@@ -76,7 +76,7 @@ jobs:
             target: aarch64-pc-windows-msvc
 
     steps:
-      - uses: actions/checkout@v5
+      - uses: actions/checkout@v6
       - uses: dtolnay/rust-toolchain@1.90
         with:
           targets: ${{ matrix.target }}
@@ -295,6 +295,15 @@ jobs:
           # ${{ matrix.target }}
           dest="dist/${{ matrix.target }}"
 
+          # We want to ship the raw Windows executables in the GitHub Release
+          # in addition to the compressed archives. Keep the originals for
+          # Windows targets; remove them elsewhere to limit the number of
+          # artifacts that end up in the GitHub Release.
+          keep_originals=false
+          if [[ "${{ matrix.runner }}" == windows* ]]; then
+            keep_originals=true
+          fi
+
           # For compatibility with environments that lack the `zstd` tool we
           # additionally create a `.tar.gz` for all platforms and `.zip` for
           # Windows alongside every single binary that we publish. The end result is:
@@ -324,7 +333,11 @@ jobs:
 
             # Also create .zst (existing behaviour) *and* remove the original
             # uncompressed binary to keep the directory small.
-            zstd -T0 -19 --rm "$dest/$base"
+            zstd_args=(-T0 -19)
+            if [[ "${keep_originals}" == false ]]; then
+              zstd_args+=(--rm)
+            fi
+            zstd "${zstd_args[@]}" "$dest/$base"
           done
 
       - name: Remove signing keychain
@@ -358,8 +371,19 @@ jobs:
           path: |
             codex-rs/dist/${{ matrix.target }}/*
 
+  shell-tool-mcp:
+    name: shell-tool-mcp
+    needs: tag-check
+    uses: ./.github/workflows/shell-tool-mcp.yml
+    with:
+      release-tag: ${{ github.ref_name }}
+      publish: true
+    secrets: inherit
+
   release:
-    needs: build
+    needs:
+      - build
+      - shell-tool-mcp
     name: release
     runs-on: ubuntu-latest
     permissions:
@@ -373,7 +397,7 @@ jobs:
 
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v5
+        uses: actions/checkout@v6
 
       - uses: actions/download-artifact@v4
         with:
@@ -382,6 +406,14 @@ jobs:
       - name: List
         run: ls -R dist/
 
+      # This is a temporary fix: we should modify shell-tool-mcp.yml so these
+      # files do not end up in dist/ in the first place.
+      - name: Delete entries from dist/ that should not go in the release
+        run: |
+          rm -rf dist/shell-tool-mcp*
+
+          ls -R dist/
+
       - name: Define release name
         id: release_name
         run: |

.github/workflows/sdk.yml

@@ -11,7 +11,7 @@ jobs:
     timeout-minutes: 10
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v5
+        uses: actions/checkout@v6
 
       - name: Setup pnpm
         uses: pnpm/action-setup@v4

.github/workflows/shell-tool-mcp-ci.yml

@@ -0,0 +1,48 @@
+name: shell-tool-mcp CI
+
+on:
+  push:
+    paths:
+      - "shell-tool-mcp/**"
+      - ".github/workflows/shell-tool-mcp-ci.yml"
+      - "pnpm-lock.yaml"
+      - "pnpm-workspace.yaml"
+  pull_request:
+    paths:
+      - "shell-tool-mcp/**"
+      - ".github/workflows/shell-tool-mcp-ci.yml"
+      - "pnpm-lock.yaml"
+      - "pnpm-workspace.yaml"
+
+env:
+  NODE_VERSION: 22
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Setup pnpm
+        uses: pnpm/action-setup@v4
+        with:
+          run_install: false
+
+      - name: Setup Node.js
+        uses: actions/setup-node@v5
+        with:
+          node-version: ${{ env.NODE_VERSION }}
+          cache: "pnpm"
+
+      - name: Install dependencies
+        run: pnpm install --frozen-lockfile
+
+      - name: Format check
+        run: pnpm --filter @openai/codex-shell-tool-mcp run format
+
+      - name: Run tests
+        run: pnpm --filter @openai/codex-shell-tool-mcp test
+
+      - name: Build
+        run: pnpm --filter @openai/codex-shell-tool-mcp run build

.github/workflows/shell-tool-mcp.yml

@@ -0,0 +1,405 @@
+name: shell-tool-mcp
+
+on:
+  workflow_call:
+    inputs:
+      release-version:
+        description: Version to publish (x.y.z or x.y.z-alpha.N). Defaults to GITHUB_REF_NAME when it starts with rust-v.
+        required: false
+        type: string
+      release-tag:
+        description: Tag name to use when downloading release artifacts (defaults to rust-v<version>).
+        required: false
+        type: string
+      publish:
+        description: Whether to publish to npm when the version is releasable.
+        required: false
+        default: true
+        type: boolean
+
+env:
+  NODE_VERSION: 22
+
+jobs:
+  metadata:
+    runs-on: ubuntu-latest
+    outputs:
+      version: ${{ steps.compute.outputs.version }}
+      release_tag: ${{ steps.compute.outputs.release_tag }}
+      should_publish: ${{ steps.compute.outputs.should_publish }}
+      npm_tag: ${{ steps.compute.outputs.npm_tag }}
+    steps:
+      - name: Compute version and tags
+        id: compute
+        run: |
+          set -euo pipefail
+
+          version="${{ inputs.release-version }}"
+          release_tag="${{ inputs.release-tag }}"
+
+          if [[ -z "$version" ]]; then
+            if [[ -n "$release_tag" && "$release_tag" =~ ^rust-v.+ ]]; then
+              version="${release_tag#rust-v}"
+            elif [[ "${GITHUB_REF_NAME:-}" =~ ^rust-v.+ ]]; then
+              version="${GITHUB_REF_NAME#rust-v}"
+              release_tag="${GITHUB_REF_NAME}"
+            else
+              echo "release-version is required when GITHUB_REF_NAME is not a rust-v tag."
+              exit 1
+            fi
+          fi
+
+          if [[ -z "$release_tag" ]]; then
+            release_tag="rust-v${version}"
+          fi
+
+          npm_tag=""
+          should_publish="false"
+          if [[ "$version" =~ ^[0-9]+\.[0-9]+\.[0-9]+$ ]]; then
+            should_publish="true"
+          elif [[ "$version" =~ ^[0-9]+\.[0-9]+\.[0-9]+-alpha\.[0-9]+$ ]]; then
+            should_publish="true"
+            npm_tag="alpha"
+          fi
+
+          echo "version=${version}" >> "$GITHUB_OUTPUT"
+          echo "release_tag=${release_tag}" >> "$GITHUB_OUTPUT"
+          echo "npm_tag=${npm_tag}" >> "$GITHUB_OUTPUT"
+          echo "should_publish=${should_publish}" >> "$GITHUB_OUTPUT"
+
+  rust-binaries:
+    name: Build Rust - ${{ matrix.target }}
+    needs: metadata
+    runs-on: ${{ matrix.runner }}
+    timeout-minutes: 30
+    defaults:
+      run:
+        working-directory: codex-rs
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - runner: macos-15-xlarge
+            target: aarch64-apple-darwin
+          - runner: macos-15-xlarge
+            target: x86_64-apple-darwin
+          - runner: ubuntu-24.04
+            target: x86_64-unknown-linux-musl
+            install_musl: true
+          - runner: ubuntu-24.04-arm
+            target: aarch64-unknown-linux-musl
+            install_musl: true
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - uses: dtolnay/rust-toolchain@1.90
+        with:
+          targets: ${{ matrix.target }}
+
+      - if: ${{ matrix.install_musl }}
+        name: Install musl build dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y musl-tools pkg-config
+
+      - name: Build exec server binaries
+        run: cargo build --release --target ${{ matrix.target }} --bin codex-exec-mcp-server --bin codex-execve-wrapper
+
+      - name: Stage exec server binaries
+        run: |
+          dest="${GITHUB_WORKSPACE}/artifacts/vendor/${{ matrix.target }}"
+          mkdir -p "$dest"
+          cp "target/${{ matrix.target }}/release/codex-exec-mcp-server" "$dest/"
+          cp "target/${{ matrix.target }}/release/codex-execve-wrapper" "$dest/"
+
+      - uses: actions/upload-artifact@v5
+        with:
+          name: shell-tool-mcp-rust-${{ matrix.target }}
+          path: artifacts/**
+          if-no-files-found: error
+
+  bash-linux:
+    name: Build Bash (Linux) - ${{ matrix.variant }} - ${{ matrix.target }}
+    needs: metadata
+    runs-on: ${{ matrix.runner }}
+    timeout-minutes: 30
+    container:
+      image: ${{ matrix.image }}
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - runner: ubuntu-24.04
+            target: x86_64-unknown-linux-musl
+            variant: ubuntu-24.04
+            image: ubuntu:24.04
+          - runner: ubuntu-24.04
+            target: x86_64-unknown-linux-musl
+            variant: ubuntu-22.04
+            image: ubuntu:22.04
+          - runner: ubuntu-24.04
+            target: x86_64-unknown-linux-musl
+            variant: debian-12
+            image: debian:12
+          - runner: ubuntu-24.04
+            target: x86_64-unknown-linux-musl
+            variant: debian-11
+            image: debian:11
+          - runner: ubuntu-24.04
+            target: x86_64-unknown-linux-musl
+            variant: centos-9
+            image: quay.io/centos/centos:stream9
+          - runner: ubuntu-24.04-arm
+            target: aarch64-unknown-linux-musl
+            variant: ubuntu-24.04
+            image: arm64v8/ubuntu:24.04
+          - runner: ubuntu-24.04-arm
+            target: aarch64-unknown-linux-musl
+            variant: ubuntu-22.04
+            image: arm64v8/ubuntu:22.04
+          - runner: ubuntu-24.04-arm
+            target: aarch64-unknown-linux-musl
+            variant: ubuntu-20.04
+            image: arm64v8/ubuntu:20.04
+          - runner: ubuntu-24.04-arm
+            target: aarch64-unknown-linux-musl
+            variant: debian-12
+            image: arm64v8/debian:12
+          - runner: ubuntu-24.04-arm
+            target: aarch64-unknown-linux-musl
+            variant: debian-11
+            image: arm64v8/debian:11
+          - runner: ubuntu-24.04-arm
+            target: aarch64-unknown-linux-musl
+            variant: centos-9
+            image: quay.io/centos/centos:stream9
+    steps:
+      - name: Install build prerequisites
+        shell: bash
+        run: |
+          set -euo pipefail
+          if command -v apt-get >/dev/null 2>&1; then
+            apt-get update
+            DEBIAN_FRONTEND=noninteractive apt-get install -y git build-essential bison autoconf gettext
+          elif command -v dnf >/dev/null 2>&1; then
+            dnf install -y git gcc gcc-c++ make bison autoconf gettext
+          elif command -v yum >/dev/null 2>&1; then
+            yum install -y git gcc gcc-c++ make bison autoconf gettext
+          else
+            echo "Unsupported package manager in container"
+            exit 1
+          fi
+
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Build patched Bash
+        shell: bash
+        run: |
+          set -euo pipefail
+          git clone --depth 1 https://github.com/bminor/bash /tmp/bash
+          cd /tmp/bash
+          git fetch --depth 1 origin a8a1c2fac029404d3f42cd39f5a20f24b6e4fe4b
+          git checkout a8a1c2fac029404d3f42cd39f5a20f24b6e4fe4b
+          git apply "${GITHUB_WORKSPACE}/shell-tool-mcp/patches/bash-exec-wrapper.patch"
+          ./configure --without-bash-malloc
+          cores="$(command -v nproc >/dev/null 2>&1 && nproc || getconf _NPROCESSORS_ONLN)"
+          make -j"${cores}"
+
+          dest="${GITHUB_WORKSPACE}/artifacts/vendor/${{ matrix.target }}/bash/${{ matrix.variant }}"
+          mkdir -p "$dest"
+          cp bash "$dest/bash"
+
+      - uses: actions/upload-artifact@v5
+        with:
+          name: shell-tool-mcp-bash-${{ matrix.target }}-${{ matrix.variant }}
+          path: artifacts/**
+          if-no-files-found: error
+
+  bash-darwin:
+    name: Build Bash (macOS) - ${{ matrix.variant }} - ${{ matrix.target }}
+    needs: metadata
+    runs-on: ${{ matrix.runner }}
+    timeout-minutes: 30
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - runner: macos-15-xlarge
+            target: aarch64-apple-darwin
+            variant: macos-15
+          - runner: macos-14
+            target: aarch64-apple-darwin
+            variant: macos-14
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Build patched Bash
+        shell: bash
+        run: |
+          set -euo pipefail
+          git clone --depth 1 https://github.com/bminor/bash /tmp/bash
+          cd /tmp/bash
+          git fetch --depth 1 origin a8a1c2fac029404d3f42cd39f5a20f24b6e4fe4b
+          git checkout a8a1c2fac029404d3f42cd39f5a20f24b6e4fe4b
+          git apply "${GITHUB_WORKSPACE}/shell-tool-mcp/patches/bash-exec-wrapper.patch"
+          ./configure --without-bash-malloc
+          cores="$(getconf _NPROCESSORS_ONLN)"
+          make -j"${cores}"
+
+          dest="${GITHUB_WORKSPACE}/artifacts/vendor/${{ matrix.target }}/bash/${{ matrix.variant }}"
+          mkdir -p "$dest"
+          cp bash "$dest/bash"
+
+      - uses: actions/upload-artifact@v5
+        with:
+          name: shell-tool-mcp-bash-${{ matrix.target }}-${{ matrix.variant }}
+          path: artifacts/**
+          if-no-files-found: error
+
+  package:
+    name: Package npm module
+    needs:
+      - metadata
+      - rust-binaries
+      - bash-linux
+      - bash-darwin
+    runs-on: ubuntu-latest
+    env:
+      PACKAGE_VERSION: ${{ needs.metadata.outputs.version }}
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Setup pnpm
+        uses: pnpm/action-setup@v4
+        with:
+          version: 10.8.1
+          run_install: false
+
+      - name: Setup Node.js
+        uses: actions/setup-node@v5
+        with:
+          node-version: ${{ env.NODE_VERSION }}
+
+      - name: Install JavaScript dependencies
+        run: pnpm install --frozen-lockfile
+
+      - name: Build (shell-tool-mcp)
+        run: pnpm --filter @openai/codex-shell-tool-mcp run build
+
+      - name: Download build artifacts
+        uses: actions/download-artifact@v4
+        with:
+          path: artifacts
+
+      - name: Assemble staging directory
+        id: staging
+        shell: bash
+        run: |
+          set -euo pipefail
+          staging="${STAGING_DIR}"
+          mkdir -p "$staging" "$staging/vendor"
+          cp shell-tool-mcp/README.md "$staging/"
+          cp shell-tool-mcp/package.json "$staging/"
+          cp -R shell-tool-mcp/bin "$staging/"
+
+          found_vendor="false"
+          shopt -s nullglob
+          for vendor_dir in artifacts/*/vendor; do
+            rsync -av "$vendor_dir/" "$staging/vendor/"
+            found_vendor="true"
+          done
+          if [[ "$found_vendor" == "false" ]]; then
+            echo "No vendor payloads were downloaded."
+            exit 1
+          fi
+
+          node - <<'NODE'
+            import fs from "node:fs";
+            import path from "node:path";
+
+            const stagingDir = process.env.STAGING_DIR;
+            const version = process.env.PACKAGE_VERSION;
+            const pkgPath = path.join(stagingDir, "package.json");
+            const pkg = JSON.parse(fs.readFileSync(pkgPath, "utf8"));
+            pkg.version = version;
+            fs.writeFileSync(pkgPath, JSON.stringify(pkg, null, 2) + "\n");
+          NODE
+
+          echo "dir=$staging" >> "$GITHUB_OUTPUT"
+        env:
+          STAGING_DIR: ${{ runner.temp }}/shell-tool-mcp
+
+      - name: Ensure binaries are executable
+        run: |
+          set -euo pipefail
+          staging="${{ steps.staging.outputs.dir }}"
+          chmod +x \
+            "$staging"/vendor/*/codex-exec-mcp-server \
+            "$staging"/vendor/*/codex-execve-wrapper \
+            "$staging"/vendor/*/bash/*/bash
+
+      - name: Create npm tarball
+        shell: bash
+        run: |
+          set -euo pipefail
+          mkdir -p dist/npm
+          staging="${{ steps.staging.outputs.dir }}"
+          pack_info=$(cd "$staging" && npm pack --ignore-scripts --json --pack-destination "${GITHUB_WORKSPACE}/dist/npm")
+          filename=$(PACK_INFO="$pack_info" node -e 'const data = JSON.parse(process.env.PACK_INFO); console.log(data[0].filename);')
+          mv "dist/npm/${filename}" "dist/npm/codex-shell-tool-mcp-npm-${PACKAGE_VERSION}.tgz"
+
+      - uses: actions/upload-artifact@v5
+        with:
+          name: codex-shell-tool-mcp-npm
+          path: dist/npm/codex-shell-tool-mcp-npm-${{ env.PACKAGE_VERSION }}.tgz
+          if-no-files-found: error
+
+  publish:
+    name: Publish npm package
+    needs:
+      - metadata
+      - package
+    if: ${{ inputs.publish && needs.metadata.outputs.should_publish == 'true' }}
+    runs-on: ubuntu-latest
+    permissions:
+      id-token: write
+      contents: read
+    steps:
+      - name: Setup pnpm
+        uses: pnpm/action-setup@v4
+        with:
+          version: 10.8.1
+          run_install: false
+
+      - name: Setup Node.js
+        uses: actions/setup-node@v5
+        with:
+          node-version: ${{ env.NODE_VERSION }}
+          registry-url: https://registry.npmjs.org
+          scope: "@openai"
+
+      - name: Update npm
+        run: npm install -g npm@latest
+
+      - name: Download npm tarball
+        uses: actions/download-artifact@v4
+        with:
+          name: codex-shell-tool-mcp-npm
+          path: dist/npm
+
+      - name: Publish to npm
+        env:
+          NPM_TAG: ${{ needs.metadata.outputs.npm_tag }}
+          VERSION: ${{ needs.metadata.outputs.version }}
+        shell: bash
+        run: |
+          set -euo pipefail
+          tag_args=()
+          if [[ -n "${NPM_TAG}" ]]; then
+            tag_args+=(--tag "${NPM_TAG}")
+          fi
+          npm publish "dist/npm/codex-shell-tool-mcp-npm-${VERSION}.tgz" "${tag_args[@]}"

.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

@@ -69,7 +69,9 @@ 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
+
+See the [Execpolicy quickstart](./docs/execpolicy.md) to set up rules that govern what commands Codex can execute.
 
 ### Docs & FAQ
 
@@ -83,6 +85,7 @@ Codex CLI supports a rich set of configuration options, with preferences stored
 - [**Configuration**](./docs/config.md)
   - [Example config](./docs/example-config.md)
 - [**Sandbox & approvals**](./docs/sandbox.md)
+- [**Execpolicy quickstart**](./docs/execpolicy.md)
 - [**Authentication**](./docs/authentication.md)
   - [Auth methods](./docs/authentication.md#forcing-a-specific-auth-method-advanced)
   - [Login on a "Headless" machine](./docs/authentication.md#connecting-on-a-headless-machine)

codex-cli/bin/codex.js

@@ -95,14 +95,6 @@ function detectPackageManager() {
     return "bun";
   }
 
-  if (
-    process.env.BUN_INSTALL ||
-    process.env.BUN_INSTALL_GLOBAL_DIR ||
-    process.env.BUN_INSTALL_BIN_DIR
-  ) {
-    return "bun";
-  }
-
   return userAgent ? "npm" : null;
 }
 

codex-rs/.cargo/audit.toml

@@ -0,0 +1,6 @@
+[advisories]
+ignore = [
+    "RUSTSEC-2024-0388", # derivative 2.2.0 via starlark; upstream crate is unmaintained
+    "RUSTSEC-2025-0057", # fxhash 0.2.1 via starlark_map; upstream crate is unmaintained
+    "RUSTSEC-2024-0436", # paste 1.0.15 via starlark/ratatui; upstream crate is unmaintained
+]

codex-rs/.config/nextest.toml

@@ -0,0 +1,13 @@
+[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 }
+
+[[profile.default.overrides]]
+filter = 'test(approval_matrix_covers_all_modes)'
+slow-timeout = { period = "30s", terminate-after = 2 }

codex-rs/.github/workflows/cargo-audit.yml

@@ -0,0 +1,26 @@
+name: Cargo audit
+
+on:
+  pull_request:
+  push:
+    branches:
+      - main
+
+permissions:
+  contents: read
+
+jobs:
+  audit:
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: codex-rs
+    steps:
+      - uses: actions/checkout@v4
+      - uses: dtolnay/rust-toolchain@stable
+      - name: Install cargo-audit
+        uses: taiki-e/install-action@v2
+        with:
+          tool: cargo-audit
+      - name: Run cargo audit
+        run: cargo audit --deny warnings

codex-rs/Cargo.toml

@@ -5,6 +5,7 @@ members = [
     "async-utils",
     "app-server",
     "app-server-protocol",
+    "app-server-test-client",
     "apply-patch",
     "arg0",
     "feedback",
@@ -15,10 +16,13 @@ members = [
     "common",
     "core",
     "exec",
+    "exec-server",
     "execpolicy",
+    "execpolicy-legacy",
     "keyring-store",
     "file-search",
     "linux-sandbox",
+    "lmstudio",
     "login",
     "mcp-server",
     "mcp-types",
@@ -37,7 +41,8 @@ members = [
     "utils/pty",
     "utils/readiness",
     "utils/string",
-    "utils/tokenizer",
+    "codex-client",
+    "codex-api",
 ]
 resolver = "2"
 
@@ -48,11 +53,13 @@ version = "0.0.0"
 # crates created with `cargo new -w ...` automatically inherit the 2024
 # edition.
 edition = "2024"
+license = "Apache-2.0"
 
 [workspace.dependencies]
 # Internal
 app_test_support = { path = "app-server/tests/common" }
 codex-ansi-escape = { path = "ansi-escape" }
+codex-api = { path = "codex-api" }
 codex-app-server = { path = "app-server" }
 codex-app-server-protocol = { path = "app-server-protocol" }
 codex-apply-patch = { path = "apply-patch" }
@@ -60,14 +67,17 @@ codex-arg0 = { path = "arg0" }
 codex-async-utils = { path = "async-utils" }
 codex-backend-client = { path = "backend-client" }
 codex-chatgpt = { path = "chatgpt" }
+codex-client = { path = "codex-client" }
 codex-common = { path = "common" }
 codex-core = { path = "core" }
 codex-exec = { path = "exec" }
+codex-execpolicy = { path = "execpolicy" }
 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" }
@@ -84,7 +94,6 @@ codex-utils-json-to-toml = { path = "utils/json-to-toml" }
 codex-utils-pty = { path = "utils/pty" }
 codex-utils-readiness = { path = "utils/readiness" }
 codex-utils-string = { path = "utils/string" }
-codex-utils-tokenizer = { path = "utils/tokenizer" }
 codex-windows-sandbox = { path = "windows-sandbox-rs" }
 core_test_support = { path = "core/tests/common" }
 mcp-types = { path = "mcp-types" }
@@ -94,7 +103,7 @@ mcp_test_support = { path = "mcp-server/tests/common" }
 allocative = "0.3.3"
 ansi-to-tui = "7.0.0"
 anyhow = "1"
-arboard = "3"
+arboard = { version = "3", features = ["wayland-data-control"] }
 askama = "0.14"
 assert_cmd = "2"
 assert_matches = "1.5.0"
@@ -104,6 +113,7 @@ async-trait = "0.1.89"
 axum = { version = "0.8", default-features = false }
 base64 = "0.22.1"
 bytes = "1.10.1"
+chardetng = "0.1.17"
 chrono = "0.4.42"
 clap = "4"
 clap_complete = "4"
@@ -115,6 +125,7 @@ diffy = "0.4.2"
 dirs = "6"
 dotenvy = "0.15.7"
 dunce = "1.0.4"
+encoding_rs = "0.8.35"
 env-flags = "0.1.1"
 env_logger = "0.11.5"
 escargot = "0.5"
@@ -122,17 +133,17 @@ 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 }
+image = { version = "^0.25.9", 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"
+libc = "0.2.177"
 log = "0.4"
 lru = "0.12.5"
 maplit = "1.0.2"
@@ -140,6 +151,7 @@ mime_guess = "2.0.5"
 multimap = "0.10.0"
 notify = "8.2.0"
 nucleo-matcher = "0.3.1"
+once_cell = "1.20.2"
 openssl-sys = "*"
 opentelemetry = "0.30.0"
 opentelemetry-appender-tracing = "0.30.0"
@@ -148,7 +160,6 @@ 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"
@@ -158,20 +169,23 @@ pulldown-cmark = "0.10"
 rand = "0.9"
 ratatui = "0.29.0"
 ratatui-macros = "0.6.0"
+regex = "1.12.2"
 regex-lite = "0.1.7"
 reqwest = "0.12"
-rmcp = { version = "0.8.5", default-features = false }
+rmcp = { version = "0.10.0", default-features = false }
 schemars = "0.8.22"
 seccompiler = "0.5.0"
 sentry = "0.34.0"
 serde = "1"
 serde_json = "1"
-serde_with = "3.14"
+serde_yaml = "0.9"
+serde_with = "3.16"
 serial_test = "3.2.0"
 sha1 = "0.10.6"
 sha2 = "0.10"
 shlex = "1.3.0"
 similar = "2.7.0"
+socket2 = "0.6.0"
 starlark = "0.13.0"
 strum = "0.27.2"
 strum_macros = "0.27.2"
@@ -188,9 +202,9 @@ tokio-stream = "0.1.17"
 tokio-test = "0.4"
 tokio-util = "0.7.16"
 toml = "0.9.5"
-toml_edit = "0.23.4"
+toml_edit = "0.23.5"
 tonic = "0.13.1"
-tracing = "0.1.41"
+tracing = "0.1.43"
 tracing-appender = "0.2.3"
 tracing-subscriber = "0.3.20"
 tracing-test = "0.2.5"
@@ -254,12 +268,7 @@ unwrap_used = "deny"
 # cargo-shear cannot see the platform-specific openssl-sys usage, so we
 # silence the false positive here instead of deleting a real dependency.
 [workspace.metadata.cargo-shear]
-ignored = [
-    "icu_provider",
-    "openssl-sys",
-    "codex-utils-readiness",
-    "codex-utils-tokenizer",
-]
+ignored = ["icu_provider", "openssl-sys", "codex-utils-readiness"]
 
 [profile.release]
 lto = "fat"
@@ -278,8 +287,8 @@ opt-level = 0
 [patch.crates-io]
 # Uncomment to debug local changes.
 # ratatui = { path = "../../ratatui" }
-ratatui = { git = "https://github.com/nornagon/ratatui", branch = "nornagon-v0.29.0-patch" }
 crossterm = { git = "https://github.com/nornagon/crossterm", branch = "nornagon/color-query" }
+ratatui = { git = "https://github.com/nornagon/ratatui", branch = "nornagon-v0.29.0-patch" }
 
 # Uncomment to debug local changes.
 # rmcp = { path = "../../rust-sdk/crates/rmcp" }

codex-rs/README.md

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

codex-rs/ansi-escape/Cargo.toml

@@ -1,7 +1,8 @@
 [package]
-edition = "2024"
 name = "codex-ansi-escape"
-version = { workspace = true }
+version.workspace = true
+edition.workspace = true
+license.workspace = true
 
 [lib]
 name = "codex_ansi_escape"

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

@@ -1,7 +1,8 @@
 [package]
-edition = "2024"
 name = "codex-app-server-protocol"
-version = { workspace = true }
+version.workspace = true
+edition.workspace = true
+license.workspace = true
 
 [lib]
 name = "codex_app_server_protocol"
@@ -15,11 +16,11 @@ 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 }
 strum_macros = { workspace = true }
+thiserror = { workspace = true }
 ts-rs = { workspace = true }
 uuid = { workspace = true, features = ["serde", "v7"] }
 

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;
@@ -64,7 +61,32 @@ pub fn generate_types(out_dir: &Path, prettier: Option<&Path>) -> Result<()> {
     Ok(())
 }
 
+#[derive(Clone, Copy, Debug)]
+pub struct GenerateTsOptions {
+    pub generate_indices: bool,
+    pub ensure_headers: bool,
+    pub run_prettier: bool,
+}
+
+impl Default for GenerateTsOptions {
+    fn default() -> Self {
+        Self {
+            generate_indices: true,
+            ensure_headers: true,
+            run_prettier: true,
+        }
+    }
+}
+
 pub fn generate_ts(out_dir: &Path, prettier: Option<&Path>) -> Result<()> {
+    generate_ts_with_options(out_dir, prettier, GenerateTsOptions::default())
+}
+
+pub fn generate_ts_with_options(
+    out_dir: &Path,
+    prettier: Option<&Path>,
+    options: GenerateTsOptions,
+) -> Result<()> {
     let v2_out_dir = out_dir.join("v2");
     ensure_dir(out_dir)?;
     ensure_dir(&v2_out_dir)?;
@@ -77,21 +99,34 @@ pub fn generate_ts(out_dir: &Path, prettier: Option<&Path>) -> Result<()> {
     export_server_responses(out_dir)?;
     ServerNotification::export_all_to(out_dir)?;
 
-    generate_index_ts(out_dir)?;
-    generate_index_ts(&v2_out_dir)?;
+    if options.generate_indices {
+        generate_index_ts(out_dir)?;
+        generate_index_ts(&v2_out_dir)?;
+    }
 
     // Ensure our header is present on all TS files (root + subdirs like v2/).
-    let ts_files = ts_files_in_recursive(out_dir)?;
-    for file in &ts_files {
-        prepend_header_if_missing(file)?;
+    let mut ts_files = Vec::new();
+    let should_collect_ts_files =
+        options.ensure_headers || (options.run_prettier && prettier.is_some());
+    if should_collect_ts_files {
+        ts_files = ts_files_in_recursive(out_dir)?;
+    }
+
+    if options.ensure_headers {
+        for file in &ts_files {
+            prepend_header_if_missing(file)?;
+        }
     }
 
     // Optionally run Prettier on all generated TS files.
-    if let Some(prettier_bin) = prettier
+    if options.run_prettier
+        && let Some(prettier_bin) = prettier
         && !ts_files.is_empty()
     {
         let status = Command::new(prettier_bin)
             .arg("--write")
+            .arg("--log-level")
+            .arg("warn")
             .args(ts_files.iter().map(|p| p.as_os_str()))
             .status()
             .with_context(|| format!("Failed to invoke Prettier at {}", prettier_bin.display()))?;
@@ -118,10 +153,6 @@ pub fn generate_json(out_dir: &Path) -> Result<()> {
         |d| write_json_schema_with_return::<crate::ClientNotification>(d, "ClientNotification"),
         |d| write_json_schema_with_return::<crate::ServerNotification>(d, "ServerNotification"),
         |d| write_json_schema_with_return::<EventMsg>(d, "EventMsg"),
-        |d| write_json_schema_with_return::<FileChange>(d, "FileChange"),
-        |d| write_json_schema_with_return::<crate::protocol::v1::InputItem>(d, "InputItem"),
-        |d| write_json_schema_with_return::<ParsedCommand>(d, "ParsedCommand"),
-        |d| write_json_schema_with_return::<SandboxPolicy>(d, "SandboxPolicy"),
     ];
 
     let mut schemas: Vec<GeneratedSchema> = Vec::new();
@@ -150,13 +181,10 @@ fn build_schema_bundle(schemas: Vec<GeneratedSchema>) -> Result<Value> {
         "ClientNotification",
         "ClientRequest",
         "EventMsg",
-        "FileChange",
-        "InputItem",
-        "ParsedCommand",
-        "SandboxPolicy",
         "ServerNotification",
         "ServerRequest",
     ];
+    const IGNORED_DEFINITIONS: &[&str] = &["Option<()>"];
 
     let namespaced_types = collect_namespaced_types(&schemas);
     let mut definitions = Map::new();
@@ -169,6 +197,10 @@ fn build_schema_bundle(schemas: Vec<GeneratedSchema>) -> Result<Value> {
             in_v1_dir,
         } = schema;
 
+        if IGNORED_DEFINITIONS.contains(&logical_name.as_str()) {
+            continue;
+        }
+
         if let Some(ref ns) = namespace {
             rewrite_refs_to_namespace(&mut value, ns);
         }
@@ -179,6 +211,9 @@ fn build_schema_bundle(schemas: Vec<GeneratedSchema>) -> Result<Value> {
             && let Value::Object(defs_obj) = defs
         {
             for (def_name, mut def_schema) in defs_obj {
+                if IGNORED_DEFINITIONS.contains(&def_name.as_str()) {
+                    continue;
+                }
                 if SPECIAL_DEFINITIONS.contains(&def_name.as_str()) {
                     continue;
                 }
@@ -384,14 +419,6 @@ 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()
         {
@@ -732,7 +759,13 @@ mod tests {
 
         let _guard = TempDirGuard(output_dir.clone());
 
-        generate_ts(&output_dir, None)?;
+        // Avoid doing more work than necessary to keep the test from timing out.
+        let options = GenerateTsOptions {
+            generate_indices: false,
+            ensure_headers: false,
+            run_prettier: false,
+        };
+        generate_ts_with_options(&output_dir, None, options)?;
 
         let mut undefined_offenders = Vec::new();
         let mut optional_nullable_offenders = BTreeSet::new();

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

@@ -7,5 +7,6 @@ pub use export::generate_ts;
 pub use export::generate_types;
 pub use jsonrpc_lite::*;
 pub use protocol::common::*;
+pub use protocol::thread_history::*;
 pub use protocol::v1::*;
 pub use protocol::v2::*;

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;
@@ -46,7 +38,7 @@ macro_rules! client_request_definitions {
     (
         $(
             $(#[$variant_meta:meta])*
-            $variant:ident {
+            $variant:ident $(=> $wire:literal)? {
                 params: $(#[$params_meta:meta])* $params:ty,
                 response: $response:ty,
             }
@@ -58,6 +50,7 @@ macro_rules! client_request_definitions {
         pub enum ClientRequest {
             $(
                 $(#[$variant_meta])*
+                $(#[serde(rename = $wire)] #[ts(rename = $wire)])?
                 $variant {
                     #[serde(rename = "id")]
                     request_id: RequestId,
@@ -101,105 +94,106 @@ macro_rules! client_request_definitions {
 }
 
 client_request_definitions! {
+    Initialize {
+        params: v1::InitializeParams,
+        response: v1::InitializeResponse,
+    },
+
     /// NEW APIs
     // Thread lifecycle
-    #[serde(rename = "thread/start")]
-    #[ts(rename = "thread/start")]
-    ThreadStart {
+    ThreadStart => "thread/start" {
         params: v2::ThreadStartParams,
         response: v2::ThreadStartResponse,
     },
-    #[serde(rename = "thread/resume")]
-    #[ts(rename = "thread/resume")]
-    ThreadResume {
+    ThreadResume => "thread/resume" {
         params: v2::ThreadResumeParams,
         response: v2::ThreadResumeResponse,
     },
-    #[serde(rename = "thread/archive")]
-    #[ts(rename = "thread/archive")]
-    ThreadArchive {
+    ThreadArchive => "thread/archive" {
         params: v2::ThreadArchiveParams,
         response: v2::ThreadArchiveResponse,
     },
-    #[serde(rename = "thread/list")]
-    #[ts(rename = "thread/list")]
-    ThreadList {
+    ThreadList => "thread/list" {
         params: v2::ThreadListParams,
         response: v2::ThreadListResponse,
     },
-    #[serde(rename = "thread/compact")]
-    #[ts(rename = "thread/compact")]
-    ThreadCompact {
+    ThreadCompact => "thread/compact" {
         params: v2::ThreadCompactParams,
         response: v2::ThreadCompactResponse,
     },
-    #[serde(rename = "turn/start")]
-    #[ts(rename = "turn/start")]
-    TurnStart {
+    TurnStart => "turn/start" {
         params: v2::TurnStartParams,
         response: v2::TurnStartResponse,
     },
-    #[serde(rename = "turn/interrupt")]
-    #[ts(rename = "turn/interrupt")]
-    TurnInterrupt {
+    TurnInterrupt => "turn/interrupt" {
         params: v2::TurnInterruptParams,
         response: v2::TurnInterruptResponse,
     },
+    ReviewStart => "review/start" {
+        params: v2::ReviewStartParams,
+        response: v2::ReviewStartResponse,
+    },
 
-    #[serde(rename = "model/list")]
-    #[ts(rename = "model/list")]
-    ModelList {
+    ModelList => "model/list" {
         params: v2::ModelListParams,
         response: v2::ModelListResponse,
     },
 
-    #[serde(rename = "account/login/start")]
-    #[ts(rename = "account/login/start")]
-    LoginAccount {
+    McpServersList => "mcpServers/list" {
+        params: v2::ListMcpServersParams,
+        response: v2::ListMcpServersResponse,
+    },
+
+    LoginAccount => "account/login/start" {
         params: v2::LoginAccountParams,
         response: v2::LoginAccountResponse,
     },
 
-    #[serde(rename = "account/login/cancel")]
-    #[ts(rename = "account/login/cancel")]
-    CancelLoginAccount {
+    CancelLoginAccount => "account/login/cancel" {
         params: v2::CancelLoginAccountParams,
         response: v2::CancelLoginAccountResponse,
     },
 
-    #[serde(rename = "account/logout")]
-    #[ts(rename = "account/logout")]
-    LogoutAccount {
+    LogoutAccount => "account/logout" {
         params: #[ts(type = "undefined")] #[serde(skip_serializing_if = "Option::is_none")] Option<()>,
         response: v2::LogoutAccountResponse,
     },
 
-    #[serde(rename = "account/rateLimits/read")]
-    #[ts(rename = "account/rateLimits/read")]
-    GetAccountRateLimits {
+    GetAccountRateLimits => "account/rateLimits/read" {
         params: #[ts(type = "undefined")] #[serde(skip_serializing_if = "Option::is_none")] Option<()>,
         response: v2::GetAccountRateLimitsResponse,
     },
 
-    #[serde(rename = "feedback/upload")]
-    #[ts(rename = "feedback/upload")]
-    FeedbackUpload {
+    FeedbackUpload => "feedback/upload" {
         params: v2::FeedbackUploadParams,
         response: v2::FeedbackUploadResponse,
     },
 
-    #[serde(rename = "account/read")]
-    #[ts(rename = "account/read")]
-    GetAccount {
+    /// Execute a command (argv vector) under the server's sandbox.
+    OneOffCommandExec => "command/exec" {
+        params: v2::CommandExecParams,
+        response: v2::CommandExecResponse,
+    },
+
+    ConfigRead => "config/read" {
+        params: v2::ConfigReadParams,
+        response: v2::ConfigReadResponse,
+    },
+    ConfigValueWrite => "config/value/write" {
+        params: v2::ConfigValueWriteParams,
+        response: v2::ConfigWriteResponse,
+    },
+    ConfigBatchWrite => "config/batchWrite" {
+        params: v2::ConfigBatchWriteParams,
+        response: v2::ConfigWriteResponse,
+    },
+
+    GetAccount => "account/read" {
         params: v2::GetAccountParams,
         response: v2::GetAccountResponse,
     },
 
     /// DEPRECATED APIs below
-    Initialize {
-        params: v1::InitializeParams,
-        response: v1::InitializeResponse,
-    },
     NewConversation {
         params: v1::NewConversationParams,
         response: v1::NewConversationResponse,
@@ -303,34 +297,36 @@ 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 },)*
                 }
             }
         }
@@ -338,9 +334,9 @@ 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(())
         }
 
@@ -349,9 +345,12 @@ 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)
         }
 
@@ -360,9 +359,12 @@ 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)
         }
     };
@@ -400,7 +402,7 @@ macro_rules! server_notification_definitions {
         impl TryFrom<JSONRPCNotification> for ServerNotification {
             type Error = serde_json::Error;
 
-            fn try_from(value: JSONRPCNotification) -> Result<Self, Self::Error> {
+            fn try_from(value: JSONRPCNotification) -> Result<Self, serde_json::Error> {
                 serde_json::from_value(serde_json::to_value(value)?)
             }
         }
@@ -452,49 +454,34 @@ 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,
+    },
+
+    /// Sent when approval is requested for a specific file change.
+    /// This request is used for Turns started via turn/start.
+    FileChangeRequestApproval => "item/fileChange/requestApproval" {
+        params: v2::FileChangeRequestApprovalParams,
+        response: v2::FileChangeRequestApprovalResponse,
+    },
+
+    /// 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)]
@@ -524,16 +511,28 @@ pub struct FuzzyFileSearchResponse {
 
 server_notification_definitions! {
     /// NEW NOTIFICATIONS
+    Error => "error" (v2::ErrorNotification),
     ThreadStarted => "thread/started" (v2::ThreadStartedNotification),
+    ThreadTokenUsageUpdated => "thread/tokenUsage/updated" (v2::ThreadTokenUsageUpdatedNotification),
     TurnStarted => "turn/started" (v2::TurnStartedNotification),
     TurnCompleted => "turn/completed" (v2::TurnCompletedNotification),
+    TurnDiffUpdated => "turn/diff/updated" (v2::TurnDiffUpdatedNotification),
+    TurnPlanUpdated => "turn/plan/updated" (v2::TurnPlanUpdatedNotification),
     ItemStarted => "item/started" (v2::ItemStartedNotification),
     ItemCompleted => "item/completed" (v2::ItemCompletedNotification),
     AgentMessageDelta => "item/agentMessage/delta" (v2::AgentMessageDeltaNotification),
     CommandExecutionOutputDelta => "item/commandExecution/outputDelta" (v2::CommandExecutionOutputDeltaNotification),
+    FileChangeOutputDelta => "item/fileChange/outputDelta" (v2::FileChangeOutputDeltaNotification),
     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),
+    ContextCompacted => "thread/compacted" (v2::ContextCompactedNotification),
+
+    /// Notifies the user of world-writable directories on Windows, which cannot be protected by the sandbox.
+    WindowsWorldWritableWarning => "windows/worldWritableWarning" (v2::WindowsWorldWritableWarningNotification),
 
     #[serde(rename = "account/login/completed")]
     #[ts(rename = "account/login/completed")]
@@ -556,17 +555,20 @@ 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-max".to_string()),
                 model_provider: None,
                 profile: None,
                 cwd: None,
@@ -584,7 +586,7 @@ mod tests {
                 "method": "newConversation",
                 "id": 42,
                 "params": {
-                    "model": "gpt-5-codex",
+                    "model": "gpt-5.1-codex-max",
                     "modelProvider": null,
                     "profile": null,
                     "cwd": null,
@@ -639,7 +641,7 @@ 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/mappers.rs

@@ -0,0 +1,15 @@
+use crate::protocol::v1;
+use crate::protocol::v2;
+
+impl From<v1::ExecOneOffCommandParams> for v2::CommandExecParams {
+    fn from(value: v1::ExecOneOffCommandParams) -> Self {
+        Self {
+            command: value.command,
+            timeout_ms: value
+                .timeout_ms
+                .map(|timeout| i64::try_from(timeout).unwrap_or(60_000)),
+            cwd: value.cwd,
+            sandbox_policy: value.sandbox_policy.map(std::convert::Into::into),
+        }
+    }
+}

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

@@ -2,5 +2,7 @@
 // Exposes protocol pieces used by `lib.rs` via `pub use protocol::common::*;`.
 
 pub mod common;
+mod mappers;
+pub mod thread_history;
 pub mod v1;
 pub mod v2;

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

@@ -0,0 +1,413 @@
+use crate::protocol::v2::ThreadItem;
+use crate::protocol::v2::Turn;
+use crate::protocol::v2::TurnError;
+use crate::protocol::v2::TurnStatus;
+use crate::protocol::v2::UserInput;
+use codex_protocol::protocol::AgentReasoningEvent;
+use codex_protocol::protocol::AgentReasoningRawContentEvent;
+use codex_protocol::protocol::EventMsg;
+use codex_protocol::protocol::TurnAbortedEvent;
+use codex_protocol::protocol::UserMessageEvent;
+
+/// Convert persisted [`EventMsg`] entries into a sequence of [`Turn`] values.
+///
+/// The purpose of this is to convert the EventMsgs persisted in a rollout file
+/// into a sequence of Turns and ThreadItems, which allows the client to render
+/// the historical messages when resuming a thread.
+pub fn build_turns_from_event_msgs(events: &[EventMsg]) -> Vec<Turn> {
+    let mut builder = ThreadHistoryBuilder::new();
+    for event in events {
+        builder.handle_event(event);
+    }
+    builder.finish()
+}
+
+struct ThreadHistoryBuilder {
+    turns: Vec<Turn>,
+    current_turn: Option<PendingTurn>,
+    next_turn_index: i64,
+    next_item_index: i64,
+}
+
+impl ThreadHistoryBuilder {
+    fn new() -> Self {
+        Self {
+            turns: Vec::new(),
+            current_turn: None,
+            next_turn_index: 1,
+            next_item_index: 1,
+        }
+    }
+
+    fn finish(mut self) -> Vec<Turn> {
+        self.finish_current_turn();
+        self.turns
+    }
+
+    /// This function should handle all EventMsg variants that can be persisted in a rollout file.
+    /// See `should_persist_event_msg` in `codex-rs/core/rollout/policy.rs`.
+    fn handle_event(&mut self, event: &EventMsg) {
+        match event {
+            EventMsg::UserMessage(payload) => self.handle_user_message(payload),
+            EventMsg::AgentMessage(payload) => self.handle_agent_message(payload.message.clone()),
+            EventMsg::AgentReasoning(payload) => self.handle_agent_reasoning(payload),
+            EventMsg::AgentReasoningRawContent(payload) => {
+                self.handle_agent_reasoning_raw_content(payload)
+            }
+            EventMsg::TokenCount(_) => {}
+            EventMsg::EnteredReviewMode(_) => {}
+            EventMsg::ExitedReviewMode(_) => {}
+            EventMsg::UndoCompleted(_) => {}
+            EventMsg::TurnAborted(payload) => self.handle_turn_aborted(payload),
+            _ => {}
+        }
+    }
+
+    fn handle_user_message(&mut self, payload: &UserMessageEvent) {
+        self.finish_current_turn();
+        let mut turn = self.new_turn();
+        let id = self.next_item_id();
+        let content = self.build_user_inputs(payload);
+        turn.items.push(ThreadItem::UserMessage { id, content });
+        self.current_turn = Some(turn);
+    }
+
+    fn handle_agent_message(&mut self, text: String) {
+        if text.is_empty() {
+            return;
+        }
+
+        let id = self.next_item_id();
+        self.ensure_turn()
+            .items
+            .push(ThreadItem::AgentMessage { id, text });
+    }
+
+    fn handle_agent_reasoning(&mut self, payload: &AgentReasoningEvent) {
+        if payload.text.is_empty() {
+            return;
+        }
+
+        // If the last item is a reasoning item, add the new text to the summary.
+        if let Some(ThreadItem::Reasoning { summary, .. }) = self.ensure_turn().items.last_mut() {
+            summary.push(payload.text.clone());
+            return;
+        }
+
+        // Otherwise, create a new reasoning item.
+        let id = self.next_item_id();
+        self.ensure_turn().items.push(ThreadItem::Reasoning {
+            id,
+            summary: vec![payload.text.clone()],
+            content: Vec::new(),
+        });
+    }
+
+    fn handle_agent_reasoning_raw_content(&mut self, payload: &AgentReasoningRawContentEvent) {
+        if payload.text.is_empty() {
+            return;
+        }
+
+        // If the last item is a reasoning item, add the new text to the content.
+        if let Some(ThreadItem::Reasoning { content, .. }) = self.ensure_turn().items.last_mut() {
+            content.push(payload.text.clone());
+            return;
+        }
+
+        // Otherwise, create a new reasoning item.
+        let id = self.next_item_id();
+        self.ensure_turn().items.push(ThreadItem::Reasoning {
+            id,
+            summary: Vec::new(),
+            content: vec![payload.text.clone()],
+        });
+    }
+
+    fn handle_turn_aborted(&mut self, _payload: &TurnAbortedEvent) {
+        let Some(turn) = self.current_turn.as_mut() else {
+            return;
+        };
+        turn.status = TurnStatus::Interrupted;
+    }
+
+    fn finish_current_turn(&mut self) {
+        if let Some(turn) = self.current_turn.take() {
+            if turn.items.is_empty() {
+                return;
+            }
+            self.turns.push(turn.into());
+        }
+    }
+
+    fn new_turn(&mut self) -> PendingTurn {
+        PendingTurn {
+            id: self.next_turn_id(),
+            items: Vec::new(),
+            error: None,
+            status: TurnStatus::Completed,
+        }
+    }
+
+    fn ensure_turn(&mut self) -> &mut PendingTurn {
+        if self.current_turn.is_none() {
+            let turn = self.new_turn();
+            return self.current_turn.insert(turn);
+        }
+
+        if let Some(turn) = self.current_turn.as_mut() {
+            return turn;
+        }
+
+        unreachable!("current turn must exist after initialization");
+    }
+
+    fn next_turn_id(&mut self) -> String {
+        let id = format!("turn-{}", self.next_turn_index);
+        self.next_turn_index += 1;
+        id
+    }
+
+    fn next_item_id(&mut self) -> String {
+        let id = format!("item-{}", self.next_item_index);
+        self.next_item_index += 1;
+        id
+    }
+
+    fn build_user_inputs(&self, payload: &UserMessageEvent) -> Vec<UserInput> {
+        let mut content = Vec::new();
+        if !payload.message.trim().is_empty() {
+            content.push(UserInput::Text {
+                text: payload.message.clone(),
+            });
+        }
+        if let Some(images) = &payload.images {
+            for image in images {
+                content.push(UserInput::Image { url: image.clone() });
+            }
+        }
+        content
+    }
+}
+
+struct PendingTurn {
+    id: String,
+    items: Vec<ThreadItem>,
+    error: Option<TurnError>,
+    status: TurnStatus,
+}
+
+impl From<PendingTurn> for Turn {
+    fn from(value: PendingTurn) -> Self {
+        Self {
+            id: value.id,
+            items: value.items,
+            error: value.error,
+            status: value.status,
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use codex_protocol::protocol::AgentMessageEvent;
+    use codex_protocol::protocol::AgentReasoningEvent;
+    use codex_protocol::protocol::AgentReasoningRawContentEvent;
+    use codex_protocol::protocol::TurnAbortReason;
+    use codex_protocol::protocol::TurnAbortedEvent;
+    use codex_protocol::protocol::UserMessageEvent;
+    use pretty_assertions::assert_eq;
+
+    #[test]
+    fn builds_multiple_turns_with_reasoning_items() {
+        let events = vec![
+            EventMsg::UserMessage(UserMessageEvent {
+                message: "First turn".into(),
+                images: Some(vec!["https://example.com/one.png".into()]),
+            }),
+            EventMsg::AgentMessage(AgentMessageEvent {
+                message: "Hi there".into(),
+            }),
+            EventMsg::AgentReasoning(AgentReasoningEvent {
+                text: "thinking".into(),
+            }),
+            EventMsg::AgentReasoningRawContent(AgentReasoningRawContentEvent {
+                text: "full reasoning".into(),
+            }),
+            EventMsg::UserMessage(UserMessageEvent {
+                message: "Second turn".into(),
+                images: None,
+            }),
+            EventMsg::AgentMessage(AgentMessageEvent {
+                message: "Reply two".into(),
+            }),
+        ];
+
+        let turns = build_turns_from_event_msgs(&events);
+        assert_eq!(turns.len(), 2);
+
+        let first = &turns[0];
+        assert_eq!(first.id, "turn-1");
+        assert_eq!(first.status, TurnStatus::Completed);
+        assert_eq!(first.items.len(), 3);
+        assert_eq!(
+            first.items[0],
+            ThreadItem::UserMessage {
+                id: "item-1".into(),
+                content: vec![
+                    UserInput::Text {
+                        text: "First turn".into(),
+                    },
+                    UserInput::Image {
+                        url: "https://example.com/one.png".into(),
+                    }
+                ],
+            }
+        );
+        assert_eq!(
+            first.items[1],
+            ThreadItem::AgentMessage {
+                id: "item-2".into(),
+                text: "Hi there".into(),
+            }
+        );
+        assert_eq!(
+            first.items[2],
+            ThreadItem::Reasoning {
+                id: "item-3".into(),
+                summary: vec!["thinking".into()],
+                content: vec!["full reasoning".into()],
+            }
+        );
+
+        let second = &turns[1];
+        assert_eq!(second.id, "turn-2");
+        assert_eq!(second.items.len(), 2);
+        assert_eq!(
+            second.items[0],
+            ThreadItem::UserMessage {
+                id: "item-4".into(),
+                content: vec![UserInput::Text {
+                    text: "Second turn".into()
+                }],
+            }
+        );
+        assert_eq!(
+            second.items[1],
+            ThreadItem::AgentMessage {
+                id: "item-5".into(),
+                text: "Reply two".into(),
+            }
+        );
+    }
+
+    #[test]
+    fn splits_reasoning_when_interleaved() {
+        let events = vec![
+            EventMsg::UserMessage(UserMessageEvent {
+                message: "Turn start".into(),
+                images: None,
+            }),
+            EventMsg::AgentReasoning(AgentReasoningEvent {
+                text: "first summary".into(),
+            }),
+            EventMsg::AgentReasoningRawContent(AgentReasoningRawContentEvent {
+                text: "first content".into(),
+            }),
+            EventMsg::AgentMessage(AgentMessageEvent {
+                message: "interlude".into(),
+            }),
+            EventMsg::AgentReasoning(AgentReasoningEvent {
+                text: "second summary".into(),
+            }),
+        ];
+
+        let turns = build_turns_from_event_msgs(&events);
+        assert_eq!(turns.len(), 1);
+        let turn = &turns[0];
+        assert_eq!(turn.items.len(), 4);
+
+        assert_eq!(
+            turn.items[1],
+            ThreadItem::Reasoning {
+                id: "item-2".into(),
+                summary: vec!["first summary".into()],
+                content: vec!["first content".into()],
+            }
+        );
+        assert_eq!(
+            turn.items[3],
+            ThreadItem::Reasoning {
+                id: "item-4".into(),
+                summary: vec!["second summary".into()],
+                content: Vec::new(),
+            }
+        );
+    }
+
+    #[test]
+    fn marks_turn_as_interrupted_when_aborted() {
+        let events = vec![
+            EventMsg::UserMessage(UserMessageEvent {
+                message: "Please do the thing".into(),
+                images: None,
+            }),
+            EventMsg::AgentMessage(AgentMessageEvent {
+                message: "Working...".into(),
+            }),
+            EventMsg::TurnAborted(TurnAbortedEvent {
+                reason: TurnAbortReason::Replaced,
+            }),
+            EventMsg::UserMessage(UserMessageEvent {
+                message: "Let's try again".into(),
+                images: None,
+            }),
+            EventMsg::AgentMessage(AgentMessageEvent {
+                message: "Second attempt complete.".into(),
+            }),
+        ];
+
+        let turns = build_turns_from_event_msgs(&events);
+        assert_eq!(turns.len(), 2);
+
+        let first_turn = &turns[0];
+        assert_eq!(first_turn.status, TurnStatus::Interrupted);
+        assert_eq!(first_turn.items.len(), 2);
+        assert_eq!(
+            first_turn.items[0],
+            ThreadItem::UserMessage {
+                id: "item-1".into(),
+                content: vec![UserInput::Text {
+                    text: "Please do the thing".into()
+                }],
+            }
+        );
+        assert_eq!(
+            first_turn.items[1],
+            ThreadItem::AgentMessage {
+                id: "item-2".into(),
+                text: "Working...".into(),
+            }
+        );
+
+        let second_turn = &turns[1];
+        assert_eq!(second_turn.status, TurnStatus::Completed);
+        assert_eq!(second_turn.items.len(), 2);
+        assert_eq!(
+            second_turn.items[0],
+            ThreadItem::UserMessage {
+                id: "item-3".into(),
+                content: vec![UserInput::Text {
+                    text: "Let's try again".into()
+                }],
+            }
+        );
+        assert_eq!(
+            second_turn.items[1],
+            ThreadItem::AgentMessage {
+                id: "item-4".into(),
+                text: "Second attempt complete.".into(),
+            }
+        );
+    }
+}

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

@@ -3,13 +3,17 @@ use std::path::PathBuf;
 
 use codex_protocol::ConversationId;
 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::Verbosity;
 use codex_protocol::models::ResponseItem;
+use codex_protocol::openai_models::ReasoningEffort;
+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-test-client/Cargo.toml

@@ -0,0 +1,17 @@
+[package]
+name = "codex-app-server-test-client"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[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 +1,862 @@
+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::ApprovalDecision;
+use codex_app_server_protocol::AskForApproval;
+use codex_app_server_protocol::ClientInfo;
+use codex_app_server_protocol::ClientRequest;
+use codex_app_server_protocol::CommandExecutionRequestAcceptSettings;
+use codex_app_server_protocol::CommandExecutionRequestApprovalParams;
+use codex_app_server_protocol::CommandExecutionRequestApprovalResponse;
+use codex_app_server_protocol::FileChangeRequestApprovalParams;
+use codex_app_server_protocol::FileChangeRequestApprovalResponse;
+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::JSONRPCRequest;
+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::ServerRequest;
+use codex_app_server_protocol::ThreadStartParams;
+use codex_app_server_protocol::ThreadStartResponse;
+use codex_app_server_protocol::TurnStartParams;
+use codex_app_server_protocol::TurnStartResponse;
+use codex_app_server_protocol::TurnStatus;
+use codex_app_server_protocol::UserInput as V2UserInput;
+use codex_protocol::ConversationId;
+use codex_protocol::protocol::Event;
+use codex_protocol::protocol::EventMsg;
+use serde::Serialize;
+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,
+    /// Send two sequential V2 turns in the same thread to test follow-up behavior.
+    SendFollowUpV2 {
+        /// Initial user message for the first turn.
+        #[arg()]
+        first_message: String,
+        /// Follow-up user message for the second turn.
+        #[arg()]
+        follow_up_message: String,
+    },
+    /// 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::SendFollowUpV2 {
+            first_message,
+            follow_up_message,
+        } => send_follow_up_v2(codex_bin, first_message, follow_up_message),
+        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 send_follow_up_v2(
+    codex_bin: String,
+    first_message: String,
+    follow_up_message: String,
+) -> 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 first_turn_params = TurnStartParams {
+        thread_id: thread_response.thread.id.clone(),
+        input: vec![V2UserInput::Text {
+            text: first_message,
+        }],
+        ..Default::default()
+    };
+    let first_turn_response = client.turn_start(first_turn_params)?;
+    println!("< turn/start response (initial): {first_turn_response:?}");
+    client.stream_turn(&thread_response.thread.id, &first_turn_response.turn.id)?;
+
+    let follow_up_params = TurnStartParams {
+        thread_id: thread_response.thread.id.clone(),
+        input: vec![V2UserInput::Text {
+            text: follow_up_message,
+        }],
+        ..Default::default()
+    };
+    let follow_up_response = client.turn_start(follow_up_params)?;
+    println!("< turn/start response (follow-up): {follow_up_response:?}");
+    client.stream_turn(&thread_response.thread.id, &follow_up_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 payload.turn.status == TurnStatus::Failed
+                            && let Some(error) = payload.turn.error
+                        {
+                            println!("[turn error] {}", error.message);
+                        }
+                        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(request) => {
+                    self.handle_server_request(request)?;
+                }
+            }
+        }
+    }
+
+    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(request) => {
+                    self.handle_server_request(request)?;
+                }
+            }
+        }
+    }
+
+    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 handle_server_request(&mut self, request: JSONRPCRequest) -> Result<()> {
+        let server_request = ServerRequest::try_from(request)
+            .context("failed to deserialize ServerRequest from JSONRPCRequest")?;
+
+        match server_request {
+            ServerRequest::CommandExecutionRequestApproval { request_id, params } => {
+                self.handle_command_execution_request_approval(request_id, params)?;
+            }
+            ServerRequest::FileChangeRequestApproval { request_id, params } => {
+                self.approve_file_change_request(request_id, params)?;
+            }
+            other => {
+                bail!("received unsupported server request: {other:?}");
+            }
+        }
+
+        Ok(())
+    }
+
+    fn handle_command_execution_request_approval(
+        &mut self,
+        request_id: RequestId,
+        params: CommandExecutionRequestApprovalParams,
+    ) -> Result<()> {
+        let CommandExecutionRequestApprovalParams {
+            thread_id,
+            turn_id,
+            item_id,
+            reason,
+            risk,
+        } = params;
+
+        println!(
+            "\n< commandExecution approval requested for thread {thread_id}, turn {turn_id}, item {item_id}"
+        );
+        if let Some(reason) = reason.as_deref() {
+            println!("< reason: {reason}");
+        }
+        if let Some(risk) = risk.as_ref() {
+            println!("< risk assessment: {risk:?}");
+        }
+
+        let response = CommandExecutionRequestApprovalResponse {
+            decision: ApprovalDecision::Accept,
+            accept_settings: Some(CommandExecutionRequestAcceptSettings { for_session: false }),
+        };
+        self.send_server_request_response(request_id, &response)?;
+        println!("< approved commandExecution request for item {item_id}");
+        Ok(())
+    }
+
+    fn approve_file_change_request(
+        &mut self,
+        request_id: RequestId,
+        params: FileChangeRequestApprovalParams,
+    ) -> Result<()> {
+        let FileChangeRequestApprovalParams {
+            thread_id,
+            turn_id,
+            item_id,
+            reason,
+            grant_root,
+        } = params;
+
+        println!(
+            "\n< fileChange approval requested for thread {thread_id}, turn {turn_id}, item {item_id}"
+        );
+        if let Some(reason) = reason.as_deref() {
+            println!("< reason: {reason}");
+        }
+        if let Some(grant_root) = grant_root.as_deref() {
+            println!("< grant root: {}", grant_root.display());
+        }
+
+        let response = FileChangeRequestApprovalResponse {
+            decision: ApprovalDecision::Accept,
+        };
+        self.send_server_request_response(request_id, &response)?;
+        println!("< approved fileChange request for item {item_id}");
+        Ok(())
+    }
+
+    fn send_server_request_response<T>(&mut self, request_id: RequestId, response: &T) -> Result<()>
+    where
+        T: Serialize,
+    {
+        let message = JSONRPCMessage::Response(JSONRPCResponse {
+            id: request_id,
+            result: serde_json::to_value(response)?,
+        });
+        self.write_jsonrpc_message(message)
+    }
+
+    fn write_jsonrpc_message(&mut self, message: JSONRPCMessage) -> Result<()> {
+        let payload = serde_json::to_string(&message)?;
+        let pretty = serde_json::to_string_pretty(&message)?;
+        print_multiline_with_prefix("> ", &pretty);
+
+        if let Some(stdin) = self.stdin.as_mut() {
+            writeln!(stdin, "{payload}")?;
+            stdin
+                .flush()
+                .context("failed to flush response to codex app-server")?;
+            return Ok(());
+        }
+
+        bail!("codex app-server stdin closed")
+    }
+}
+
+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

@@ -1,7 +1,8 @@
 [package]
-edition = "2024"
 name = "codex-app-server"
-version = { workspace = true }
+version.workspace = true
+edition.workspace = true
+license.workspace = true
 
 [[bin]]
 name = "codex-app-server"
@@ -29,6 +30,10 @@ codex-utils-json-to-toml = { workspace = true }
 chrono = { workspace = true }
 serde = { workspace = true, features = ["derive"] }
 serde_json = { workspace = true }
+sha2 = { workspace = true }
+mcp-types = { workspace = true }
+tempfile = { workspace = true }
+toml = { workspace = true }
 tokio = { workspace = true, features = [
     "io-std",
     "macros",
@@ -46,9 +51,9 @@ 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 }
-tempfile = { workspace = true }
-toml = { workspace = true }
 wiremock = { workspace = true }
+shlex = { workspace = true }

codex-rs/app-server/README.md

@@ -1,6 +1,16 @@
 # codex-app-server
 
-`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.
+`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).
+
+## Table of Contents
+- [Protocol](#protocol)
+- [Message Schema](#message-schema)
+- [Core Primitives](#core-primitives)
+- [Lifecycle Overview](#lifecycle-overview)
+- [Initialization](#initialization)
+- [API Overview](#api-overview)
+- [Events](#events)
+- [Auth endpoints](#auth-endpoints)
 
 ## Protocol
 
@@ -15,40 +25,53 @@ codex app-server generate-ts --out DIR
 codex app-server generate-json-schema --out DIR
 ```
 
+## Core Primitives
+
+The API exposes three top level primitives representing an interaction between a user and Codex:
+- **Thread**: A conversation between a user and the Codex agent. Each thread contains multiple turns.
+- **Turn**: One turn of the conversation, typically starting with a user message and finishing with an agent message. Each turn contains multiple items.
+- **Item**: Represents user inputs and agent outputs as part of the turn, persisted and used as the context for future conversations. Example items include user message, agent reasoning, agent message, shell command, file edit, etc.
+
+Use the thread APIs to create, list, or archive conversations. Drive a conversation with turn APIs and stream progress via turn notifications.
+
+## 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.
 
-Example:
+Applications building on top of `codex app-server` should identify themselves via the `clientInfo` parameter.
 
+Example (from OpenAI's official VSCode extension):
 ```json
 { "method": "initialize", "id": 0, "params": {
     "clientInfo": { "name": "codex-vscode", "title": "Codex VS Code Extension", "version": "0.1.0" }
 } }
-{ "id": 0, "result": { "userAgent": "codex-app-server/0.1.0 codex-vscode/0.1.0" } }
-{ "method": "initialized" }
 ```
 
-## Core primitives
-
-We have 3 top level primitives:
-- Thread - a conversation between the Codex agent and a user. Each thread contains multiple turns.
-- Turn - one turn of the conversation, typically starting with a user message and finishing with an agent message. Each turn contains multiple items.
-- Item - represents user inputs and agent outputs as part of the turn, persisted and used as the context for future conversations.
-
-## Thread & turn endpoints
-
-The JSON-RPC API exposes dedicated methods for managing Codex conversations. Threads store long-lived conversation metadata, and turns store the per-message exchange (input → Codex output, including streamed items). Use the thread APIs to create, list, or archive sessions, then drive the conversation with turn APIs and notifications.
-
-### Quick reference
+## API Overview
 - `thread/start` — create a new thread; emits `thread/started` and auto-subscribes you to turn/item events for that thread.
 - `thread/resume` — reopen an existing thread by id so subsequent `turn/start` calls append to it.
 - `thread/list` — page through stored rollouts; supports cursor-based pagination and optional `modelProviders` filtering.
 - `thread/archive` — move a thread’s rollout file into the archived directory; returns `{}` on success.
 - `turn/start` — add user input to a thread and begin Codex generation; responds with the initial `turn` object and streams `turn/started`, `item/*`, and `turn/completed` notifications.
 - `turn/interrupt` — request cancellation of an in-flight turn by `(thread_id, turn_id)`; success is an empty `{}` response and the turn finishes with `status: "interrupted"`.
+- `review/start` — kick off Codex’s automated reviewer for a thread; responds like `turn/start` and emits `item/started`/`item/completed` notifications with `enteredReviewMode` and `exitedReviewMode` items, plus a final assistant `agentMessage` containing the review.
+- `command/exec` — run a single command under the server sandbox without starting a thread/turn (handy for utilities and validation).
+- `model/list` — list available models (with reasoning effort options).
+- `feedback/upload` — submit a feedback report (classification + optional reason/logs and conversation_id); returns the tracking thread id.
+- `command/exec` — run a single command under the server sandbox without starting a thread/turn (handy for utilities and validation).
+- `config/read` — fetch the effective config on disk after resolving config layering.
+- `config/value/write` — write a single config key/value to the user's config.toml on disk.
+- `config/batchWrite` — apply multiple config edits atomically to the user's config.toml on disk.
 
-### 1) Start or resume a thread
+### Example: Start or resume a thread
 
 Start a fresh thread when you need a new Codex conversation.
 
@@ -56,7 +79,7 @@ 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",
@@ -79,7 +102,7 @@ To continue a stored session, call `thread/resume` with the `thread.id` you prev
 { "id": 11, "result": { "thread": { "id": "thr_123", … } } }
 ```
 
-### 2) List threads (pagination & filters)
+### Example: List threads (with pagination & filters)
 
 `thread/list` lets you render a history UI. Pass any combination of:
 - `cursor` — opaque string from a prior response; omit for the first page.
@@ -104,7 +127,7 @@ Example:
 
 When `nextCursor` is `null`, you’ve reached the final page.
 
-### 3) Archive a thread
+### Example: Archive a thread
 
 Use `thread/archive` to move the persisted rollout (stored as a JSONL file on disk) into the archived sessions directory.
 
@@ -115,7 +138,7 @@ Use `thread/archive` to move the persisted rollout (stored as a JSONL file on di
 
 An archived thread will not appear in future calls to `thread/list`.
 
-### 4) Start a turn (send user input)
+### Example: Start a turn (send user input)
 
 Turns attach user input (text or images) to a thread and trigger Codex generation. The `input` field is a list of discriminated unions:
 
@@ -137,7 +160,7 @@ 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"
 } }
@@ -149,7 +172,7 @@ You can optionally specify config overrides on the new turn. If specified, these
 } } }
 ```
 
-### 5) Interrupt an active turn
+### Example: Interrupt an active turn
 
 You can cancel a running Turn with `turn/interrupt`.
 
@@ -163,11 +186,179 @@ You can cancel a running Turn with `turn/interrupt`.
 
 The server requests cancellations for running subprocesses, then emits a `turn/completed` event with `status: "interrupted"`. Rely on the `turn/completed` to know when Codex-side cleanup is done.
 
+### Example: Request a code review
+
+Use `review/start` to run Codex’s reviewer on the currently checked-out project. The request takes the thread id plus a `target` describing what should be reviewed:
+
+- `{"type":"uncommittedChanges"}` — staged, unstaged, and untracked files.
+- `{"type":"baseBranch","branch":"main"}` — diff against the provided branch’s upstream (see prompt for the exact `git merge-base`/`git diff` instructions Codex will run).
+- `{"type":"commit","sha":"abc1234","title":"Optional subject"}` — review a specific commit.
+- `{"type":"custom","instructions":"Free-form reviewer instructions"}` — fallback prompt equivalent to the legacy manual review request.
+- `delivery` (`"inline"` or `"detached"`, default `"inline"`) — where the review runs:
+  - `"inline"`: run the review as a new turn on the existing thread. The response’s `reviewThreadId` equals the original `threadId`, and no new `thread/started` notification is emitted.
+  - `"detached"`: fork a new review thread from the parent conversation and run the review there. The response’s `reviewThreadId` is the id of this new review thread, and the server emits a `thread/started` notification for it before streaming review items.
+
+Example request/response:
+
+```json
+{ "method": "review/start", "id": 40, "params": {
+    "threadId": "thr_123",
+    "delivery": "inline",
+    "target": { "type": "commit", "sha": "1234567deadbeef", "title": "Polish tui colors" }
+} }
+{ "id": 40, "result": {
+    "turn": {
+        "id": "turn_900",
+        "status": "inProgress",
+        "items": [
+            { "type": "userMessage", "id": "turn_900", "content": [ { "type": "text", "text": "Review commit 1234567: Polish tui colors" } ] }
+        ],
+        "error": null
+    },
+    "reviewThreadId": "thr_123"
+} }
+```
+
+For a detached review, use `"delivery": "detached"`. The response is the same shape, but `reviewThreadId` will be the id of the new review thread (different from the original `threadId`). The server also emits a `thread/started` notification for that new thread before streaming the review turn.
+
+Codex streams the usual `turn/started` notification followed by an `item/started`
+with an `enteredReviewMode` item so clients can show progress:
+
+```json
+{ "method": "item/started", "params": { "item": {
+    "type": "enteredReviewMode",
+    "id": "turn_900",
+    "review": "current changes"
+} } }
+```
+
+When the reviewer finishes, the server emits `item/started` and `item/completed`
+containing an `exitedReviewMode` item with the final review text:
+
+```json
+{ "method": "item/completed", "params": { "item": {
+    "type": "exitedReviewMode",
+    "id": "turn_900",
+    "review": "Looks solid overall...\n\n- Prefer Stylize helpers — app.rs:10-20\n  ..."
+} } }
+```
+
+The `review` string is plain text that already bundles the overall explanation plus a bullet list for each structured finding (matching `ThreadItem::ExitedReviewMode` in the generated schema). Use this notification to render the reviewer output in your client.
+
+### Example: One-off command execution
+
+Run a standalone command (argv vector) in the server’s sandbox without creating a thread or turn:
+
+```json
+{ "method": "command/exec", "id": 32, "params": {
+    "command": ["ls", "-la"],
+    "cwd": "/Users/me/project",                    // optional; defaults to server cwd
+    "sandboxPolicy": { "type": "workspaceWrite" }, // optional; defaults to user config
+    "timeoutMs": 10000                             // optional; ms timeout; defaults to server timeout
+} }
+{ "id": 32, "result": { "exitCode": 0, "stdout": "...", "stderr": "" } }
+```
+
+Notes:
+- Empty `command` arrays are rejected.
+- `sandboxPolicy` accepts the same shape used by `turn/start` (e.g., `dangerFullAccess`, `readOnly`, `workspaceWrite` with flags).
+- When omitted, `timeoutMs` falls back to the server default.
+
+## Events
+
+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` status). Token usage events stream separately via `thread/tokenUsage/updated`. 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`.
+
+- `turn/started` — `{ turn }` with the turn id, empty `items`, and `status: "inProgress"`.
+- `turn/completed` — `{ turn }` where `turn.status` is `completed`, `interrupted`, or `failed`; failures carry `{ error: { message, codexErrorInfo? } }`.
+- `turn/diff/updated` — `{ threadId, turnId, diff }` represents the up-to-date snapshot of the turn-level unified diff, emitted after every FileChange item. `diff` is the latest aggregated unified diff across every file change in the turn. UIs can render this to show the full "what changed" view without stitching individual `fileChange` items.
+- `turn/plan/updated` — `{ turnId, explanation?, plan }` whenever the agent shares or changes its plan; each `plan` entry is `{ step, status }` with `status` in `pending`, `inProgress`, or `completed`.
+
+Today both notifications carry an empty `items` array even when item events were streamed; rely on `item/*` notifications for the canonical item list until this is fixed.
+
+#### 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).
+- `commandExecution` — `{id, command, cwd, status, commandActions, aggregatedOutput?, exitCode?, durationMs?}` for sandboxed commands; `status` is `inProgress`, `completed`, `failed`, or `declined`.
+- `fileChange` — `{id, changes, status}` describing proposed edits; `changes` list `{path, kind, diff}` and `status` is `inProgress`, `completed`, `failed`, or `declined`.
+- `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.
+- `imageView` — `{id, path}` emitted when the agent invokes the image viewer tool.
+- `enteredReviewMode` — `{id, review}` sent when the reviewer starts; `review` is a short user-facing label such as `"current changes"` or the requested target description.
+- `exitedReviewMode` — `{id, review}` emitted when the reviewer finishes; `review` is the full plain-text review (usually, overall notes plus bullet point findings).
+- `compacted` - `{threadId, turnId}` when codex compacts the conversation history. This can happen automatically.
+
+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.
+#### commandExecution
+- `item/commandExecution/outputDelta` — streams stdout/stderr for the command; append deltas in order to render live output alongside `aggregatedOutput` in the final item.
+Final `commandExecution` items include parsed `commandActions`, `status`, `exitCode`, and `durationMs` so the UI can summarize what ran and whether it succeeded.
+#### fileChange
+- `item/fileChange/outputDelta` - contains the tool call response of the underlying `apply_patch` tool call.
+
+### Errors
+`error` event is emitted whenever the server hits an error mid-turn (for example, upstream model errors or quota limits). Carries the same `{ error: { message, codexErrorInfo? } }` payload as `turn.status: "failed"` and may precede that terminal notification.
+
+  `codexErrorInfo` maps to the `CodexErrorInfo` enum. Common values:
+  - `ContextWindowExceeded`
+  - `UsageLimitExceeded`
+  - `HttpConnectionFailed { httpStatusCode? }`: upstream HTTP failures including 4xx/5xx
+  - `ResponseStreamConnectionFailed { httpStatusCode? }`: failure to connect to the response SSE stream
+  - `ResponseStreamDisconnected { httpStatusCode? }`: disconnect of the response SSE stream in the middle of a turn before completion
+  - `ResponseTooManyFailedAttempts { httpStatusCode? }`
+  - `BadRequest`
+  - `Unauthorized`
+  - `SandboxError`
+  - `InternalServerError`
+  - `Other`: all unclassified errors
+
+When an upstream HTTP status is available (for example, from the Responses API or a provider), it is forwarded in `httpStatusCode` on the relevant `codexErrorInfo` variant.
+
+## Approvals
+
+Certain actions (shell commands or modifying files) may require explicit user approval depending on the user's config. When `turn/start` is used, the app-server drives an approval flow by sending a server-initiated JSON-RPC request to the client. The client must respond to tell Codex whether to proceed. UIs should present these requests inline with the active turn so users can review the proposed command or diff before choosing.
+
+- Requests include `threadId` and `turnId`—use them to scope UI state to the active conversation.
+- Respond with a single `{ "decision": "accept" | "decline" }` payload (plus optional `acceptSettings` on command executions). The server resumes or declines the work and ends the item with `item/completed`.
+
+### Command execution approvals
+
+Order of messages:
+1. `item/started` — shows the pending `commandExecution` item with `command`, `cwd`, and other fields so you can render the proposed action.
+2. `item/commandExecution/requestApproval` (request) — carries the same `itemId`, `threadId`, `turnId`, optionally `reason` or `risk`, plus `parsedCmd` for friendly display.
+3. Client response — `{ "decision": "accept", "acceptSettings": { "forSession": false } }` or `{ "decision": "decline" }`.
+4. `item/completed` — final `commandExecution` item with `status: "completed" | "failed" | "declined"` and execution output. Render this as the authoritative result.
+
+### File change approvals
+
+Order of messages:
+1. `item/started` — emits a `fileChange` item with `changes` (diff chunk summaries) and `status: "inProgress"`. Show the proposed edits and paths to the user.
+2. `item/fileChange/requestApproval` (request) — includes `itemId`, `threadId`, `turnId`, and an optional `reason`.
+3. Client response — `{ "decision": "accept" }` or `{ "decision": "decline" }`.
+4. `item/completed` — returns the same `fileChange` item with `status` updated to `completed`, `failed`, or `declined` after the patch attempt. Rely on this to show success/failure and finalize the diff state in your UI.
+
+UI guidance for IDEs: surface an approval dialog as soon as the request arrives. The turn will proceed after the server receives a response to the approval request. The terminal `item/completed` notification will be sent with the appropriate status.
+
 ## Auth endpoints
 
 The JSON-RPC auth/account surface exposes request/response methods plus server-initiated notifications (no `id`). Use these to determine auth state, start or cancel logins, logout, and inspect ChatGPT rate limits.
 
-### Quick reference
+### API Overview
 - `account/read` — fetch current account info; optionally refresh tokens.
 - `account/login/start` — begin login (`apiKey` or `chatgpt`).
 - `account/login/completed` (notify) — emitted when a login attempt finishes (success or error).
@@ -252,9 +443,3 @@ Field notes:
 - `usedPercent` is current usage within the OpenAI quota window.
 - `windowDurationMins` is the quota window length.
 - `resetsAt` is a Unix timestamp (seconds) for the next reset.
-
-### Dev notes
-
-- `codex app-server generate-ts --out <dir>` emits v2 types under `v2/`.
-- `codex app-server generate-json-schema --out <dir>` outputs `codex_app_server_protocol.schemas.json`.
-- See [“Authentication and authorization” in the config docs](../../docs/config.md#authentication-and-authorization) for configuration knobs.

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

@@ -0,0 +1,974 @@
+use crate::error_code::INTERNAL_ERROR_CODE;
+use crate::error_code::INVALID_REQUEST_ERROR_CODE;
+use anyhow::anyhow;
+use codex_app_server_protocol::ConfigBatchWriteParams;
+use codex_app_server_protocol::ConfigLayer;
+use codex_app_server_protocol::ConfigLayerMetadata;
+use codex_app_server_protocol::ConfigLayerName;
+use codex_app_server_protocol::ConfigReadParams;
+use codex_app_server_protocol::ConfigReadResponse;
+use codex_app_server_protocol::ConfigValueWriteParams;
+use codex_app_server_protocol::ConfigWriteErrorCode;
+use codex_app_server_protocol::ConfigWriteResponse;
+use codex_app_server_protocol::JSONRPCErrorError;
+use codex_app_server_protocol::MergeStrategy;
+use codex_app_server_protocol::OverriddenMetadata;
+use codex_app_server_protocol::WriteStatus;
+use codex_core::config::ConfigToml;
+use codex_core::config_loader::LoadedConfigLayers;
+use codex_core::config_loader::LoaderOverrides;
+use codex_core::config_loader::load_config_layers_with_overrides;
+use codex_core::config_loader::merge_toml_values;
+use serde_json::Value as JsonValue;
+use serde_json::json;
+use sha2::Digest;
+use sha2::Sha256;
+use std::collections::HashMap;
+use std::path::Path;
+use std::path::PathBuf;
+use tempfile::NamedTempFile;
+use tokio::task;
+use toml::Value as TomlValue;
+
+const SESSION_FLAGS_SOURCE: &str = "--config";
+const MDM_SOURCE: &str = "com.openai.codex/config_toml_base64";
+const CONFIG_FILE_NAME: &str = "config.toml";
+
+#[derive(Clone)]
+pub(crate) struct ConfigApi {
+    codex_home: PathBuf,
+    cli_overrides: Vec<(String, TomlValue)>,
+    loader_overrides: LoaderOverrides,
+}
+
+impl ConfigApi {
+    pub(crate) fn new(codex_home: PathBuf, cli_overrides: Vec<(String, TomlValue)>) -> Self {
+        Self {
+            codex_home,
+            cli_overrides,
+            loader_overrides: LoaderOverrides::default(),
+        }
+    }
+
+    #[cfg(test)]
+    fn with_overrides(
+        codex_home: PathBuf,
+        cli_overrides: Vec<(String, TomlValue)>,
+        loader_overrides: LoaderOverrides,
+    ) -> Self {
+        Self {
+            codex_home,
+            cli_overrides,
+            loader_overrides,
+        }
+    }
+
+    pub(crate) async fn read(
+        &self,
+        params: ConfigReadParams,
+    ) -> Result<ConfigReadResponse, JSONRPCErrorError> {
+        let layers = self
+            .load_layers_state()
+            .await
+            .map_err(|err| internal_error("failed to read configuration layers", err))?;
+
+        let effective = layers.effective_config();
+        validate_config(&effective).map_err(|err| internal_error("invalid configuration", err))?;
+
+        let response = ConfigReadResponse {
+            config: to_json_value(&effective),
+            origins: layers.origins(),
+            layers: params.include_layers.then(|| layers.layers_high_to_low()),
+        };
+
+        Ok(response)
+    }
+
+    pub(crate) async fn write_value(
+        &self,
+        params: ConfigValueWriteParams,
+    ) -> Result<ConfigWriteResponse, JSONRPCErrorError> {
+        let edits = vec![(params.key_path, params.value, params.merge_strategy)];
+        self.apply_edits(params.file_path, params.expected_version, edits)
+            .await
+    }
+
+    pub(crate) async fn batch_write(
+        &self,
+        params: ConfigBatchWriteParams,
+    ) -> Result<ConfigWriteResponse, JSONRPCErrorError> {
+        let edits = params
+            .edits
+            .into_iter()
+            .map(|edit| (edit.key_path, edit.value, edit.merge_strategy))
+            .collect();
+
+        self.apply_edits(params.file_path, params.expected_version, edits)
+            .await
+    }
+
+    async fn apply_edits(
+        &self,
+        file_path: String,
+        expected_version: Option<String>,
+        edits: Vec<(String, JsonValue, MergeStrategy)>,
+    ) -> Result<ConfigWriteResponse, JSONRPCErrorError> {
+        let allowed_path = self.codex_home.join(CONFIG_FILE_NAME);
+        if !paths_match(&allowed_path, &file_path) {
+            return Err(config_write_error(
+                ConfigWriteErrorCode::ConfigLayerReadonly,
+                "Only writes to the user config are allowed",
+            ));
+        }
+
+        let layers = self
+            .load_layers_state()
+            .await
+            .map_err(|err| internal_error("failed to load configuration", err))?;
+
+        if let Some(expected) = expected_version.as_deref()
+            && expected != layers.user.version
+        {
+            return Err(config_write_error(
+                ConfigWriteErrorCode::ConfigVersionConflict,
+                "Configuration was modified since last read. Fetch latest version and retry.",
+            ));
+        }
+
+        let mut user_config = layers.user.config.clone();
+        let mut mutated = false;
+        let mut parsed_segments = Vec::new();
+
+        for (key_path, value, strategy) in edits.into_iter() {
+            let segments = parse_key_path(&key_path).map_err(|message| {
+                config_write_error(ConfigWriteErrorCode::ConfigValidationError, message)
+            })?;
+            let parsed_value = parse_value(value).map_err(|message| {
+                config_write_error(ConfigWriteErrorCode::ConfigValidationError, message)
+            })?;
+
+            let changed = apply_merge(&mut user_config, &segments, parsed_value.as_ref(), strategy)
+                .map_err(|err| match err {
+                    MergeError::PathNotFound => config_write_error(
+                        ConfigWriteErrorCode::ConfigPathNotFound,
+                        "Path not found",
+                    ),
+                    MergeError::Validation(message) => {
+                        config_write_error(ConfigWriteErrorCode::ConfigValidationError, message)
+                    }
+                })?;
+
+            mutated |= changed;
+            parsed_segments.push(segments);
+        }
+
+        validate_config(&user_config).map_err(|err| {
+            config_write_error(
+                ConfigWriteErrorCode::ConfigValidationError,
+                format!("Invalid configuration: {err}"),
+            )
+        })?;
+
+        let updated_layers = layers.with_user_config(user_config.clone());
+        let effective = updated_layers.effective_config();
+        validate_config(&effective).map_err(|err| {
+            config_write_error(
+                ConfigWriteErrorCode::ConfigValidationError,
+                format!("Invalid configuration: {err}"),
+            )
+        })?;
+
+        if mutated {
+            self.persist_user_config(&user_config)
+                .await
+                .map_err(|err| internal_error("failed to persist config.toml", err))?;
+        }
+
+        let overridden = first_overridden_edit(&updated_layers, &effective, &parsed_segments);
+        let status = overridden
+            .as_ref()
+            .map(|_| WriteStatus::OkOverridden)
+            .unwrap_or(WriteStatus::Ok);
+
+        Ok(ConfigWriteResponse {
+            status,
+            version: updated_layers.user.version.clone(),
+            overridden_metadata: overridden,
+        })
+    }
+
+    async fn load_layers_state(&self) -> std::io::Result<LayersState> {
+        let LoadedConfigLayers {
+            base,
+            managed_config,
+            managed_preferences,
+        } = load_config_layers_with_overrides(&self.codex_home, self.loader_overrides.clone())
+            .await?;
+
+        let user = LayerState::new(
+            ConfigLayerName::User,
+            self.codex_home.join(CONFIG_FILE_NAME),
+            base,
+        );
+
+        let session_flags = LayerState::new(
+            ConfigLayerName::SessionFlags,
+            PathBuf::from(SESSION_FLAGS_SOURCE),
+            {
+                let mut root = TomlValue::Table(toml::map::Map::new());
+                for (path, value) in self.cli_overrides.iter() {
+                    apply_override(&mut root, path, value.clone());
+                }
+                root
+            },
+        );
+
+        let system = managed_config.map(|cfg| {
+            LayerState::new(
+                ConfigLayerName::System,
+                system_config_path(&self.codex_home),
+                cfg,
+            )
+        });
+
+        let mdm = managed_preferences
+            .map(|cfg| LayerState::new(ConfigLayerName::Mdm, PathBuf::from(MDM_SOURCE), cfg));
+
+        Ok(LayersState {
+            user,
+            session_flags,
+            system,
+            mdm,
+        })
+    }
+
+    async fn persist_user_config(&self, user_config: &TomlValue) -> anyhow::Result<()> {
+        let codex_home = self.codex_home.clone();
+        let serialized = toml::to_string_pretty(user_config)?;
+
+        task::spawn_blocking(move || -> anyhow::Result<()> {
+            std::fs::create_dir_all(&codex_home)?;
+
+            let target = codex_home.join(CONFIG_FILE_NAME);
+            let tmp = NamedTempFile::new_in(&codex_home)?;
+            std::fs::write(tmp.path(), serialized.as_bytes())?;
+            tmp.persist(&target)?;
+            Ok(())
+        })
+        .await
+        .map_err(|err| anyhow!("config persistence task panicked: {err}"))??;
+
+        Ok(())
+    }
+}
+
+fn parse_value(value: JsonValue) -> Result<Option<TomlValue>, String> {
+    if value.is_null() {
+        return Ok(None);
+    }
+
+    serde_json::from_value::<TomlValue>(value)
+        .map(Some)
+        .map_err(|err| format!("invalid value: {err}"))
+}
+
+fn parse_key_path(path: &str) -> Result<Vec<String>, String> {
+    if path.trim().is_empty() {
+        return Err("keyPath must not be empty".to_string());
+    }
+    Ok(path
+        .split('.')
+        .map(std::string::ToString::to_string)
+        .collect())
+}
+
+fn apply_override(target: &mut TomlValue, path: &str, value: TomlValue) {
+    use toml::value::Table;
+
+    let segments: Vec<&str> = path.split('.').collect();
+    let mut current = target;
+
+    for (idx, segment) in segments.iter().enumerate() {
+        let is_last = idx == segments.len() - 1;
+
+        if is_last {
+            match current {
+                TomlValue::Table(table) => {
+                    table.insert(segment.to_string(), value);
+                }
+                _ => {
+                    let mut table = Table::new();
+                    table.insert(segment.to_string(), value);
+                    *current = TomlValue::Table(table);
+                }
+            }
+            return;
+        }
+
+        match current {
+            TomlValue::Table(table) => {
+                current = table
+                    .entry((*segment).to_string())
+                    .or_insert_with(|| TomlValue::Table(Table::new()));
+            }
+            _ => {
+                *current = TomlValue::Table(Table::new());
+                if let TomlValue::Table(tbl) = current {
+                    current = tbl
+                        .entry((*segment).to_string())
+                        .or_insert_with(|| TomlValue::Table(Table::new()));
+                }
+            }
+        }
+    }
+}
+
+#[derive(Debug)]
+enum MergeError {
+    PathNotFound,
+    Validation(String),
+}
+
+fn apply_merge(
+    root: &mut TomlValue,
+    segments: &[String],
+    value: Option<&TomlValue>,
+    strategy: MergeStrategy,
+) -> Result<bool, MergeError> {
+    let Some(value) = value else {
+        return clear_path(root, segments);
+    };
+
+    let Some((last, parents)) = segments.split_last() else {
+        return Err(MergeError::Validation(
+            "keyPath must not be empty".to_string(),
+        ));
+    };
+
+    let mut current = root;
+
+    for segment in parents {
+        match current {
+            TomlValue::Table(table) => {
+                current = table
+                    .entry(segment.clone())
+                    .or_insert_with(|| TomlValue::Table(toml::map::Map::new()));
+            }
+            _ => {
+                *current = TomlValue::Table(toml::map::Map::new());
+                if let TomlValue::Table(table) = current {
+                    current = table
+                        .entry(segment.clone())
+                        .or_insert_with(|| TomlValue::Table(toml::map::Map::new()));
+                }
+            }
+        }
+    }
+
+    let table = current.as_table_mut().ok_or_else(|| {
+        MergeError::Validation("cannot set value on non-table parent".to_string())
+    })?;
+
+    if matches!(strategy, MergeStrategy::Upsert)
+        && let Some(existing) = table.get_mut(last)
+        && matches!(existing, TomlValue::Table(_))
+        && matches!(value, TomlValue::Table(_))
+    {
+        merge_toml_values(existing, value);
+        return Ok(true);
+    }
+
+    let changed = table
+        .get(last)
+        .map(|existing| Some(existing) != Some(value))
+        .unwrap_or(true);
+    table.insert(last.clone(), value.clone());
+    Ok(changed)
+}
+
+fn clear_path(root: &mut TomlValue, segments: &[String]) -> Result<bool, MergeError> {
+    let Some((last, parents)) = segments.split_last() else {
+        return Err(MergeError::Validation(
+            "keyPath must not be empty".to_string(),
+        ));
+    };
+
+    let mut current = root;
+    for segment in parents {
+        match current {
+            TomlValue::Table(table) => {
+                current = table.get_mut(segment).ok_or(MergeError::PathNotFound)?;
+            }
+            _ => return Err(MergeError::PathNotFound),
+        }
+    }
+
+    let Some(parent) = current.as_table_mut() else {
+        return Err(MergeError::PathNotFound);
+    };
+
+    Ok(parent.remove(last).is_some())
+}
+
+#[derive(Clone)]
+struct LayerState {
+    name: ConfigLayerName,
+    source: PathBuf,
+    config: TomlValue,
+    version: String,
+}
+
+impl LayerState {
+    fn new(name: ConfigLayerName, source: PathBuf, config: TomlValue) -> Self {
+        let version = version_for_toml(&config);
+        Self {
+            name,
+            source,
+            config,
+            version,
+        }
+    }
+
+    fn metadata(&self) -> ConfigLayerMetadata {
+        ConfigLayerMetadata {
+            name: self.name.clone(),
+            source: self.source.display().to_string(),
+            version: self.version.clone(),
+        }
+    }
+
+    fn as_layer(&self) -> ConfigLayer {
+        ConfigLayer {
+            name: self.name.clone(),
+            source: self.source.display().to_string(),
+            version: self.version.clone(),
+            config: to_json_value(&self.config),
+        }
+    }
+}
+
+#[derive(Clone)]
+struct LayersState {
+    user: LayerState,
+    session_flags: LayerState,
+    system: Option<LayerState>,
+    mdm: Option<LayerState>,
+}
+
+impl LayersState {
+    fn with_user_config(self, user_config: TomlValue) -> Self {
+        Self {
+            user: LayerState::new(self.user.name, self.user.source, user_config),
+            session_flags: self.session_flags,
+            system: self.system,
+            mdm: self.mdm,
+        }
+    }
+
+    fn effective_config(&self) -> TomlValue {
+        let mut merged = self.user.config.clone();
+        merge_toml_values(&mut merged, &self.session_flags.config);
+        if let Some(system) = &self.system {
+            merge_toml_values(&mut merged, &system.config);
+        }
+        if let Some(mdm) = &self.mdm {
+            merge_toml_values(&mut merged, &mdm.config);
+        }
+        merged
+    }
+
+    fn origins(&self) -> HashMap<String, ConfigLayerMetadata> {
+        let mut origins = HashMap::new();
+        let mut path = Vec::new();
+
+        record_origins(
+            &self.user.config,
+            &self.user.metadata(),
+            &mut path,
+            &mut origins,
+        );
+        record_origins(
+            &self.session_flags.config,
+            &self.session_flags.metadata(),
+            &mut path,
+            &mut origins,
+        );
+        if let Some(system) = &self.system {
+            record_origins(&system.config, &system.metadata(), &mut path, &mut origins);
+        }
+        if let Some(mdm) = &self.mdm {
+            record_origins(&mdm.config, &mdm.metadata(), &mut path, &mut origins);
+        }
+
+        origins
+    }
+
+    fn layers_high_to_low(&self) -> Vec<ConfigLayer> {
+        let mut layers = Vec::new();
+        if let Some(mdm) = &self.mdm {
+            layers.push(mdm.as_layer());
+        }
+        if let Some(system) = &self.system {
+            layers.push(system.as_layer());
+        }
+        layers.push(self.session_flags.as_layer());
+        layers.push(self.user.as_layer());
+        layers
+    }
+}
+
+fn record_origins(
+    value: &TomlValue,
+    meta: &ConfigLayerMetadata,
+    path: &mut Vec<String>,
+    origins: &mut HashMap<String, ConfigLayerMetadata>,
+) {
+    match value {
+        TomlValue::Table(table) => {
+            for (key, val) in table {
+                path.push(key.clone());
+                record_origins(val, meta, path, origins);
+                path.pop();
+            }
+        }
+        TomlValue::Array(items) => {
+            for (idx, item) in items.iter().enumerate() {
+                path.push(idx.to_string());
+                record_origins(item, meta, path, origins);
+                path.pop();
+            }
+        }
+        _ => {
+            if !path.is_empty() {
+                origins.insert(path.join("."), meta.clone());
+            }
+        }
+    }
+}
+
+fn to_json_value(value: &TomlValue) -> JsonValue {
+    serde_json::to_value(value).unwrap_or(JsonValue::Null)
+}
+
+fn validate_config(value: &TomlValue) -> Result<(), toml::de::Error> {
+    let _: ConfigToml = value.clone().try_into()?;
+    Ok(())
+}
+
+fn version_for_toml(value: &TomlValue) -> String {
+    let json = to_json_value(value);
+    let canonical = canonical_json(&json);
+    let serialized = serde_json::to_vec(&canonical).unwrap_or_default();
+    let mut hasher = Sha256::new();
+    hasher.update(serialized);
+    let hash = hasher.finalize();
+    let hex = hash
+        .iter()
+        .map(|byte| format!("{byte:02x}"))
+        .collect::<String>();
+    format!("sha256:{hex}")
+}
+
+fn canonical_json(value: &JsonValue) -> JsonValue {
+    match value {
+        JsonValue::Object(map) => {
+            let mut sorted = serde_json::Map::new();
+            let mut keys = map.keys().cloned().collect::<Vec<_>>();
+            keys.sort();
+            for key in keys {
+                if let Some(val) = map.get(&key) {
+                    sorted.insert(key, canonical_json(val));
+                }
+            }
+            JsonValue::Object(sorted)
+        }
+        JsonValue::Array(items) => JsonValue::Array(items.iter().map(canonical_json).collect()),
+        other => other.clone(),
+    }
+}
+
+fn paths_match(expected: &Path, provided: &str) -> bool {
+    let provided_path = PathBuf::from(provided);
+    if let (Ok(expanded_expected), Ok(expanded_provided)) =
+        (expected.canonicalize(), provided_path.canonicalize())
+    {
+        return expanded_expected == expanded_provided;
+    }
+
+    expected == provided_path
+}
+
+fn value_at_path<'a>(root: &'a TomlValue, segments: &[String]) -> Option<&'a TomlValue> {
+    let mut current = root;
+    for segment in segments {
+        match current {
+            TomlValue::Table(table) => {
+                current = table.get(segment)?;
+            }
+            TomlValue::Array(items) => {
+                let idx: usize = segment.parse().ok()?;
+                current = items.get(idx)?;
+            }
+            _ => return None,
+        }
+    }
+    Some(current)
+}
+
+fn override_message(layer: &ConfigLayerName) -> String {
+    match layer {
+        ConfigLayerName::Mdm => "Overridden by managed policy (mdm)".to_string(),
+        ConfigLayerName::System => "Overridden by managed config (system)".to_string(),
+        ConfigLayerName::SessionFlags => "Overridden by session flags".to_string(),
+        ConfigLayerName::User => "Overridden by user config".to_string(),
+    }
+}
+
+fn compute_override_metadata(
+    layers: &LayersState,
+    effective: &TomlValue,
+    segments: &[String],
+) -> Option<OverriddenMetadata> {
+    let user_value = value_at_path(&layers.user.config, segments);
+    let effective_value = value_at_path(effective, segments);
+
+    if user_value.is_some() && user_value == effective_value {
+        return None;
+    }
+
+    if user_value.is_none() && effective_value.is_none() {
+        return None;
+    }
+
+    let effective_layer = find_effective_layer(layers, segments);
+    let overriding_layer = effective_layer.unwrap_or_else(|| layers.user.metadata());
+    let message = override_message(&overriding_layer.name);
+
+    Some(OverriddenMetadata {
+        message,
+        overriding_layer,
+        effective_value: effective_value
+            .map(to_json_value)
+            .unwrap_or(JsonValue::Null),
+    })
+}
+
+fn first_overridden_edit(
+    layers: &LayersState,
+    effective: &TomlValue,
+    edits: &[Vec<String>],
+) -> Option<OverriddenMetadata> {
+    for segments in edits {
+        if let Some(meta) = compute_override_metadata(layers, effective, segments) {
+            return Some(meta);
+        }
+    }
+    None
+}
+
+fn find_effective_layer(layers: &LayersState, segments: &[String]) -> Option<ConfigLayerMetadata> {
+    let check =
+        |state: &LayerState| value_at_path(&state.config, segments).map(|_| state.metadata());
+
+    if let Some(mdm) = &layers.mdm
+        && let Some(meta) = check(mdm)
+    {
+        return Some(meta);
+    }
+    if let Some(system) = &layers.system
+        && let Some(meta) = check(system)
+    {
+        return Some(meta);
+    }
+    if let Some(meta) = check(&layers.session_flags) {
+        return Some(meta);
+    }
+    check(&layers.user)
+}
+
+fn system_config_path(codex_home: &Path) -> PathBuf {
+    if let Ok(path) = std::env::var("CODEX_MANAGED_CONFIG_PATH") {
+        return PathBuf::from(path);
+    }
+
+    #[cfg(unix)]
+    {
+        let _ = codex_home;
+        PathBuf::from("/etc/codex/managed_config.toml")
+    }
+
+    #[cfg(not(unix))]
+    {
+        codex_home.join("managed_config.toml")
+    }
+}
+
+fn internal_error<E: std::fmt::Display>(context: &str, err: E) -> JSONRPCErrorError {
+    JSONRPCErrorError {
+        code: INTERNAL_ERROR_CODE,
+        message: format!("{context}: {err}"),
+        data: None,
+    }
+}
+
+fn config_write_error(code: ConfigWriteErrorCode, message: impl Into<String>) -> JSONRPCErrorError {
+    JSONRPCErrorError {
+        code: INVALID_REQUEST_ERROR_CODE,
+        message: message.into(),
+        data: Some(json!({
+            "config_write_error_code": code,
+        })),
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use pretty_assertions::assert_eq;
+    use tempfile::tempdir;
+
+    #[tokio::test]
+    async fn read_includes_origins_and_layers() {
+        let tmp = tempdir().expect("tempdir");
+        std::fs::write(tmp.path().join(CONFIG_FILE_NAME), "model = \"user\"").unwrap();
+
+        let managed_path = tmp.path().join("managed_config.toml");
+        std::fs::write(&managed_path, "approval_policy = \"never\"").unwrap();
+
+        let api = ConfigApi::with_overrides(
+            tmp.path().to_path_buf(),
+            vec![],
+            LoaderOverrides {
+                managed_config_path: Some(managed_path),
+                #[cfg(target_os = "macos")]
+                managed_preferences_base64: None,
+            },
+        );
+
+        let response = api
+            .read(ConfigReadParams {
+                include_layers: true,
+            })
+            .await
+            .expect("response");
+
+        assert_eq!(
+            response.config.get("approval_policy"),
+            Some(&json!("never"))
+        );
+
+        assert_eq!(
+            response
+                .origins
+                .get("approval_policy")
+                .expect("origin")
+                .name,
+            ConfigLayerName::System
+        );
+        let layers = response.layers.expect("layers present");
+        assert_eq!(layers.first().unwrap().name, ConfigLayerName::System);
+        assert_eq!(layers.get(1).unwrap().name, ConfigLayerName::SessionFlags);
+        assert_eq!(layers.last().unwrap().name, ConfigLayerName::User);
+    }
+
+    #[tokio::test]
+    async fn write_value_reports_override() {
+        let tmp = tempdir().expect("tempdir");
+        std::fs::write(
+            tmp.path().join(CONFIG_FILE_NAME),
+            "approval_policy = \"on-request\"",
+        )
+        .unwrap();
+
+        let managed_path = tmp.path().join("managed_config.toml");
+        std::fs::write(&managed_path, "approval_policy = \"never\"").unwrap();
+
+        let api = ConfigApi::with_overrides(
+            tmp.path().to_path_buf(),
+            vec![],
+            LoaderOverrides {
+                managed_config_path: Some(managed_path),
+                #[cfg(target_os = "macos")]
+                managed_preferences_base64: None,
+            },
+        );
+
+        let result = api
+            .write_value(ConfigValueWriteParams {
+                file_path: tmp.path().join(CONFIG_FILE_NAME).display().to_string(),
+                key_path: "approval_policy".to_string(),
+                value: json!("never"),
+                merge_strategy: MergeStrategy::Replace,
+                expected_version: None,
+            })
+            .await
+            .expect("result");
+
+        let read_after = api
+            .read(ConfigReadParams {
+                include_layers: true,
+            })
+            .await
+            .expect("read");
+        let config_object = read_after.config.as_object().expect("object");
+        assert_eq!(config_object.get("approval_policy"), Some(&json!("never")));
+        assert_eq!(
+            read_after
+                .origins
+                .get("approval_policy")
+                .expect("origin")
+                .name,
+            ConfigLayerName::System
+        );
+        assert_eq!(result.status, WriteStatus::Ok);
+        assert!(result.overridden_metadata.is_none());
+    }
+
+    #[tokio::test]
+    async fn version_conflict_rejected() {
+        let tmp = tempdir().expect("tempdir");
+        std::fs::write(tmp.path().join(CONFIG_FILE_NAME), "model = \"user\"").unwrap();
+
+        let api = ConfigApi::new(tmp.path().to_path_buf(), vec![]);
+        let error = api
+            .write_value(ConfigValueWriteParams {
+                file_path: tmp.path().join(CONFIG_FILE_NAME).display().to_string(),
+                key_path: "model".to_string(),
+                value: json!("gpt-5"),
+                merge_strategy: MergeStrategy::Replace,
+                expected_version: Some("sha256:bogus".to_string()),
+            })
+            .await
+            .expect_err("should fail");
+
+        assert_eq!(error.code, INVALID_REQUEST_ERROR_CODE);
+        assert_eq!(
+            error
+                .data
+                .as_ref()
+                .and_then(|d| d.get("config_write_error_code"))
+                .and_then(serde_json::Value::as_str),
+            Some("configVersionConflict")
+        );
+    }
+
+    #[tokio::test]
+    async fn invalid_user_value_rejected_even_if_overridden_by_managed() {
+        let tmp = tempdir().expect("tempdir");
+        std::fs::write(tmp.path().join(CONFIG_FILE_NAME), "model = \"user\"").unwrap();
+
+        let managed_path = tmp.path().join("managed_config.toml");
+        std::fs::write(&managed_path, "approval_policy = \"never\"").unwrap();
+
+        let api = ConfigApi::with_overrides(
+            tmp.path().to_path_buf(),
+            vec![],
+            LoaderOverrides {
+                managed_config_path: Some(managed_path),
+                #[cfg(target_os = "macos")]
+                managed_preferences_base64: None,
+            },
+        );
+
+        let error = api
+            .write_value(ConfigValueWriteParams {
+                file_path: tmp.path().join(CONFIG_FILE_NAME).display().to_string(),
+                key_path: "approval_policy".to_string(),
+                value: json!("bogus"),
+                merge_strategy: MergeStrategy::Replace,
+                expected_version: None,
+            })
+            .await
+            .expect_err("should fail validation");
+
+        assert_eq!(error.code, INVALID_REQUEST_ERROR_CODE);
+        assert_eq!(
+            error
+                .data
+                .as_ref()
+                .and_then(|d| d.get("config_write_error_code"))
+                .and_then(serde_json::Value::as_str),
+            Some("configValidationError")
+        );
+
+        let contents =
+            std::fs::read_to_string(tmp.path().join(CONFIG_FILE_NAME)).expect("read config");
+        assert_eq!(contents.trim(), "model = \"user\"");
+    }
+
+    #[tokio::test]
+    async fn read_reports_managed_overrides_user_and_session_flags() {
+        let tmp = tempdir().expect("tempdir");
+        std::fs::write(tmp.path().join(CONFIG_FILE_NAME), "model = \"user\"").unwrap();
+
+        let managed_path = tmp.path().join("managed_config.toml");
+        std::fs::write(&managed_path, "model = \"system\"").unwrap();
+
+        let cli_overrides = vec![(
+            "model".to_string(),
+            TomlValue::String("session".to_string()),
+        )];
+
+        let api = ConfigApi::with_overrides(
+            tmp.path().to_path_buf(),
+            cli_overrides,
+            LoaderOverrides {
+                managed_config_path: Some(managed_path),
+                #[cfg(target_os = "macos")]
+                managed_preferences_base64: None,
+            },
+        );
+
+        let response = api
+            .read(ConfigReadParams {
+                include_layers: true,
+            })
+            .await
+            .expect("response");
+
+        assert_eq!(response.config.get("model"), Some(&json!("system")));
+        assert_eq!(
+            response.origins.get("model").expect("origin").name,
+            ConfigLayerName::System
+        );
+        let layers = response.layers.expect("layers");
+        assert_eq!(layers.first().unwrap().name, ConfigLayerName::System);
+        assert_eq!(layers.get(1).unwrap().name, ConfigLayerName::SessionFlags);
+        assert_eq!(layers.get(2).unwrap().name, ConfigLayerName::User);
+    }
+
+    #[tokio::test]
+    async fn write_value_reports_managed_override() {
+        let tmp = tempdir().expect("tempdir");
+        std::fs::write(tmp.path().join(CONFIG_FILE_NAME), "").unwrap();
+
+        let managed_path = tmp.path().join("managed_config.toml");
+        std::fs::write(&managed_path, "approval_policy = \"never\"").unwrap();
+
+        let api = ConfigApi::with_overrides(
+            tmp.path().to_path_buf(),
+            vec![],
+            LoaderOverrides {
+                managed_config_path: Some(managed_path),
+                #[cfg(target_os = "macos")]
+                managed_preferences_base64: None,
+            },
+        );
+
+        let result = api
+            .write_value(ConfigValueWriteParams {
+                file_path: tmp.path().join(CONFIG_FILE_NAME).display().to_string(),
+                key_path: "approval_policy".to_string(),
+                value: json!("on-request"),
+                merge_strategy: MergeStrategy::Replace,
+                expected_version: None,
+            })
+            .await
+            .expect("result");
+
+        assert_eq!(result.status, WriteStatus::OkOverridden);
+        let overridden = result.overridden_metadata.expect("overridden metadata");
+        assert_eq!(overridden.overriding_layer.name, ConfigLayerName::System);
+        assert_eq!(overridden.effective_value, json!("never"));
+    }
+}

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

@@ -18,6 +18,7 @@ use tokio::io::AsyncWriteExt;
 use tokio::io::BufReader;
 use tokio::io::{self};
 use tokio::sync::mpsc;
+use toml::Value as TomlValue;
 use tracing::Level;
 use tracing::debug;
 use tracing::error;
@@ -28,7 +29,9 @@ use tracing_subscriber::filter::Targets;
 use tracing_subscriber::layer::SubscriberExt;
 use tracing_subscriber::util::SubscriberInitExt;
 
+mod bespoke_event_handling;
 mod codex_message_processor;
+mod config_api;
 mod error_code;
 mod fuzzy_file_search;
 mod message_processor;
@@ -46,7 +49,7 @@ 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({
@@ -79,11 +82,12 @@ pub async fn run_main(
             format!("error parsing -c overrides: {e}"),
         )
     })?;
-    let config = Config::load_with_cli_overrides(cli_kv_overrides, ConfigOverrides::default())
-        .await
-        .map_err(|e| {
-            std::io::Error::new(ErrorKind::InvalidData, format!("error loading config: {e}"))
-        })?;
+    let config =
+        Config::load_with_cli_overrides(cli_kv_overrides.clone(), ConfigOverrides::default())
+            .await
+            .map_err(|e| {
+                std::io::Error::new(ErrorKind::InvalidData, format!("error loading config: {e}"))
+            })?;
 
     let feedback = CodexFeedback::new();
 
@@ -120,10 +124,12 @@ pub async fn run_main(
     // Task: process incoming messages.
     let processor_handle = tokio::spawn({
         let outgoing_message_sender = OutgoingMessageSender::new(outgoing_tx);
+        let cli_overrides: Vec<(String, TomlValue)> = cli_kv_overrides.clone();
         let mut processor = MessageProcessor::new(
             outgoing_message_sender,
             codex_linux_sandbox_exe,
             std::sync::Arc::new(config),
+            cli_overrides,
             feedback.clone(),
         );
         async move {

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

@@ -1,17 +1,22 @@
 use std::path::PathBuf;
+use std::sync::Arc;
 
 use crate::codex_message_processor::CodexMessageProcessor;
+use crate::config_api::ConfigApi;
 use crate::error_code::INVALID_REQUEST_ERROR_CODE;
 use crate::outgoing_message::OutgoingMessageSender;
 use codex_app_server_protocol::ClientInfo;
 use codex_app_server_protocol::ClientRequest;
+use codex_app_server_protocol::ConfigBatchWriteParams;
+use codex_app_server_protocol::ConfigReadParams;
+use codex_app_server_protocol::ConfigValueWriteParams;
 use codex_app_server_protocol::InitializeResponse;
-
 use codex_app_server_protocol::JSONRPCError;
 use codex_app_server_protocol::JSONRPCErrorError;
 use codex_app_server_protocol::JSONRPCNotification;
 use codex_app_server_protocol::JSONRPCRequest;
 use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::RequestId;
 use codex_core::AuthManager;
 use codex_core::ConversationManager;
 use codex_core::config::Config;
@@ -19,11 +24,12 @@ use codex_core::default_client::USER_AGENT_SUFFIX;
 use codex_core::default_client::get_codex_user_agent;
 use codex_feedback::CodexFeedback;
 use codex_protocol::protocol::SessionSource;
-use std::sync::Arc;
+use toml::Value as TomlValue;
 
 pub(crate) struct MessageProcessor {
     outgoing: Arc<OutgoingMessageSender>,
     codex_message_processor: CodexMessageProcessor,
+    config_api: ConfigApi,
     initialized: bool,
 }
 
@@ -34,6 +40,7 @@ impl MessageProcessor {
         outgoing: OutgoingMessageSender,
         codex_linux_sandbox_exe: Option<PathBuf>,
         config: Arc<Config>,
+        cli_overrides: Vec<(String, TomlValue)>,
         feedback: CodexFeedback,
     ) -> Self {
         let outgoing = Arc::new(outgoing);
@@ -51,13 +58,15 @@ impl MessageProcessor {
             conversation_manager,
             outgoing.clone(),
             codex_linux_sandbox_exe,
-            config,
+            Arc::clone(&config),
             feedback,
         );
+        let config_api = ConfigApi::new(config.codex_home.clone(), cli_overrides);
 
         Self {
             outgoing,
             codex_message_processor,
+            config_api,
             initialized: false,
         }
     }
@@ -118,6 +127,7 @@ impl MessageProcessor {
                     self.outgoing.send_response(request_id, response).await;
 
                     self.initialized = true;
+
                     return;
                 }
             }
@@ -134,9 +144,20 @@ impl MessageProcessor {
             }
         }
 
-        self.codex_message_processor
-            .process_request(codex_request)
-            .await;
+        match codex_request {
+            ClientRequest::ConfigRead { request_id, params } => {
+                self.handle_config_read(request_id, params).await;
+            }
+            ClientRequest::ConfigValueWrite { request_id, params } => {
+                self.handle_config_value_write(request_id, params).await;
+            }
+            ClientRequest::ConfigBatchWrite { request_id, params } => {
+                self.handle_config_batch_write(request_id, params).await;
+            }
+            other => {
+                self.codex_message_processor.process_request(other).await;
+            }
+        }
     }
 
     pub(crate) async fn process_notification(&self, notification: JSONRPCNotification) {
@@ -156,4 +177,33 @@ impl MessageProcessor {
     pub(crate) fn process_error(&mut self, err: JSONRPCError) {
         tracing::error!("<- error: {:?}", err);
     }
+
+    async fn handle_config_read(&self, request_id: RequestId, params: ConfigReadParams) {
+        match self.config_api.read(params).await {
+            Ok(response) => self.outgoing.send_response(request_id, response).await,
+            Err(error) => self.outgoing.send_error(request_id, error).await,
+        }
+    }
+
+    async fn handle_config_value_write(
+        &self,
+        request_id: RequestId,
+        params: ConfigValueWriteParams,
+    ) {
+        match self.config_api.write_value(params).await {
+            Ok(response) => self.outgoing.send_response(request_id, response).await,
+            Err(error) => self.outgoing.send_error(request_id, error).await,
+        }
+    }
+
+    async fn handle_config_batch_write(
+        &self,
+        request_id: RequestId,
+        params: ConfigBatchWriteParams,
+    ) {
+        match self.config_api.batch_write(params).await {
+            Ok(response) => self.outgoing.send_response(request_id, response).await,
+            Err(error) => self.outgoing.send_error(request_id, error).await,
+        }
+    }
 }

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

@@ -1,9 +1,9 @@
 use codex_app_server_protocol::AuthMode;
 use codex_app_server_protocol::Model;
 use codex_app_server_protocol::ReasoningEffortOption;
-use codex_common::model_presets::ModelPreset;
-use codex_common::model_presets::ReasoningEffortPreset;
-use codex_common::model_presets::builtin_model_presets;
+use codex_core::openai_models::model_presets::builtin_model_presets;
+use codex_protocol::openai_models::ModelPreset;
+use codex_protocol::openai_models::ReasoningEffortPreset;
 
 pub fn supported_models(auth_mode: Option<AuthMode>) -> Vec<Model> {
     builtin_model_presets(auth_mode)
@@ -27,7 +27,7 @@ fn model_from_preset(preset: ModelPreset) -> Model {
 }
 
 fn reasoning_efforts_from_preset(
-    efforts: &'static [ReasoningEffortPreset],
+    efforts: Vec<ReasoningEffortPreset>,
 ) -> Vec<ReasoningEffortOption> {
     efforts
         .iter()

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:?}");
+        }
     }
 }
 
@@ -215,6 +229,7 @@ mod tests {
                         resets_at: Some(123),
                     }),
                     secondary: None,
+                    credits: None,
                 },
             });
 
@@ -229,7 +244,8 @@ mod tests {
                             "windowDurationMins": 15,
                             "resetsAt": 123
                         },
-                        "secondary": null
+                        "secondary": null,
+                        "credits": null
                     }
                 },
             }),

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

@@ -1,7 +1,8 @@
 [package]
-edition = "2024"
 name = "app_test_support"
-version = { workspace = true }
+version.workspace = true
+edition.workspace = true
+license.workspace = true
 
 [lib]
 path = "lib.rs"
@@ -24,3 +25,5 @@ tokio = { workspace = true, features = [
 ] }
 uuid = { workspace = true }
 wiremock = { workspace = true }
+core_test_support = { path = "../../../core/tests/common" }
+shlex = { workspace = true }

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

@@ -9,12 +9,15 @@ pub use auth_fixtures::ChatGptIdTokenClaims;
 pub use auth_fixtures::encode_id_token;
 pub use auth_fixtures::write_chatgpt_auth;
 use codex_app_server_protocol::JSONRPCResponse;
+pub use core_test_support::format_with_current_shell;
+pub use core_test_support::format_with_current_shell_display;
 pub use mcp_process::McpProcess;
 pub use mock_model_server::create_mock_chat_completions_server;
 pub use mock_model_server::create_mock_chat_completions_server_unchecked;
 pub use responses::create_apply_patch_sse_response;
+pub use responses::create_exec_command_sse_response;
 pub use responses::create_final_assistant_message_sse_response;
-pub use responses::create_shell_sse_response;
+pub use responses::create_shell_command_sse_response;
 pub use rollout::create_fake_rollout;
 use serde::de::DeserializeOwned;
 

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

@@ -18,6 +18,9 @@ use codex_app_server_protocol::CancelLoginAccountParams;
 use codex_app_server_protocol::CancelLoginChatGptParams;
 use codex_app_server_protocol::ClientInfo;
 use codex_app_server_protocol::ClientNotification;
+use codex_app_server_protocol::ConfigBatchWriteParams;
+use codex_app_server_protocol::ConfigReadParams;
+use codex_app_server_protocol::ConfigValueWriteParams;
 use codex_app_server_protocol::FeedbackUploadParams;
 use codex_app_server_protocol::GetAccountParams;
 use codex_app_server_protocol::GetAuthStatusParams;
@@ -35,6 +38,7 @@ use codex_app_server_protocol::NewConversationParams;
 use codex_app_server_protocol::RemoveConversationListenerParams;
 use codex_app_server_protocol::RequestId;
 use codex_app_server_protocol::ResumeConversationParams;
+use codex_app_server_protocol::ReviewStartParams;
 use codex_app_server_protocol::SendUserMessageParams;
 use codex_app_server_protocol::SendUserTurnParams;
 use codex_app_server_protocol::ServerRequest;
@@ -377,6 +381,15 @@ impl McpProcess {
         self.send_request("turn/interrupt", params).await
     }
 
+    /// Send a `review/start` JSON-RPC request (v2).
+    pub async fn send_review_start_request(
+        &mut self,
+        params: ReviewStartParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("review/start", params).await
+    }
+
     /// Send a `cancelLoginChatGpt` JSON-RPC request.
     pub async fn send_cancel_login_chat_gpt_request(
         &mut self,
@@ -391,6 +404,30 @@ impl McpProcess {
         self.send_request("logoutChatGpt", None).await
     }
 
+    pub async fn send_config_read_request(
+        &mut self,
+        params: ConfigReadParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("config/read", params).await
+    }
+
+    pub async fn send_config_value_write_request(
+        &mut self,
+        params: ConfigValueWriteParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("config/value/write", params).await
+    }
+
+    pub async fn send_config_batch_write_request(
+        &mut self,
+        params: ConfigBatchWriteParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("config/batchWrite", params).await
+    }
+
     /// Send an `account/logout` JSON-RPC request.
     pub async fn send_logout_account_request(&mut self) -> anyhow::Result<i64> {
         self.send_request("account/logout", None).await

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

@@ -1,17 +1,18 @@
 use serde_json::json;
 use std::path::Path;
 
-pub fn create_shell_sse_response(
+pub fn create_shell_command_sse_response(
     command: Vec<String>,
     workdir: Option<&Path>,
     timeout_ms: Option<u64>,
     call_id: &str,
 ) -> anyhow::Result<String> {
-    // The `arguments`` for the `shell` tool is a serialized JSON object.
+    // The `arguments` for the `shell_command` tool is a serialized JSON object.
+    let command_str = shlex::try_join(command.iter().map(String::as_str))?;
     let tool_call_arguments = serde_json::to_string(&json!({
-        "command": command,
+        "command": command_str,
         "workdir": workdir.map(|w| w.to_string_lossy()),
-        "timeout": timeout_ms
+        "timeout_ms": timeout_ms
     }))?;
     let tool_call = json!({
         "choices": [
@@ -21,7 +22,7 @@ pub fn create_shell_sse_response(
                         {
                             "id": call_id,
                             "function": {
-                                "name": "shell",
+                                "name": "shell_command",
                                 "arguments": tool_call_arguments
                             }
                         }
@@ -62,10 +63,10 @@ pub fn create_apply_patch_sse_response(
     patch_content: &str,
     call_id: &str,
 ) -> anyhow::Result<String> {
-    // Use shell command to call apply_patch with heredoc format
-    let shell_command = format!("apply_patch <<'EOF'\n{patch_content}\nEOF");
+    // Use shell_command to call apply_patch with heredoc format
+    let command = format!("apply_patch <<'EOF'\n{patch_content}\nEOF");
     let tool_call_arguments = serde_json::to_string(&json!({
-        "command": ["bash", "-lc", shell_command]
+        "command": command
     }))?;
 
     let tool_call = json!({
@@ -76,7 +77,46 @@ pub fn create_apply_patch_sse_response(
                         {
                             "id": call_id,
                             "function": {
-                                "name": "shell",
+                                "name": "shell_command",
+                                "arguments": tool_call_arguments
+                            }
+                        }
+                    ]
+                },
+                "finish_reason": "tool_calls"
+            }
+        ]
+    });
+
+    let sse = format!(
+        "data: {}\n\ndata: DONE\n\n",
+        serde_json::to_string(&tool_call)?
+    );
+    Ok(sse)
+}
+
+pub fn create_exec_command_sse_response(call_id: &str) -> anyhow::Result<String> {
+    let (cmd, args) = if cfg!(windows) {
+        ("cmd.exe", vec!["/d", "/c", "echo hi"])
+    } else {
+        ("/bin/sh", vec!["-c", "echo hi"])
+    };
+    let command = std::iter::once(cmd.to_string())
+        .chain(args.into_iter().map(str::to_string))
+        .collect::<Vec<_>>();
+    let tool_call_arguments = serde_json::to_string(&json!({
+        "cmd": command.join(" "),
+        "yield_time_ms": 500
+    }))?;
+    let tool_call = json!({
+        "choices": [
+            {
+                "delta": {
+                    "tool_calls": [
+                        {
+                            "id": call_id,
+                            "function": {
+                                "name": "exec_command",
                                 "arguments": tool_call_arguments
                             }
                         }

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

@@ -1,6 +1,8 @@
 use anyhow::Result;
 use codex_protocol::ConversationId;
+use codex_protocol::protocol::GitInfo;
 use codex_protocol::protocol::SessionMeta;
+use codex_protocol::protocol::SessionMetaLine;
 use codex_protocol::protocol::SessionSource;
 use serde_json::json;
 use std::fs;
@@ -22,6 +24,7 @@ pub fn create_fake_rollout(
     meta_rfc3339: &str,
     preview: &str,
     model_provider: Option<&str>,
+    git_info: Option<GitInfo>,
 ) -> Result<String> {
     let uuid = Uuid::new_v4();
     let uuid_str = uuid.to_string();
@@ -37,7 +40,7 @@ pub fn create_fake_rollout(
     let file_path = dir.join(format!("rollout-{filename_ts}-{uuid}.jsonl"));
 
     // Build JSONL lines
-    let payload = serde_json::to_value(SessionMeta {
+    let meta = SessionMeta {
         id: conversation_id,
         timestamp: meta_rfc3339.to_string(),
         cwd: PathBuf::from("/"),
@@ -46,6 +49,10 @@ pub fn create_fake_rollout(
         instructions: None,
         source: SessionSource::Cli,
         model_provider: model_provider.map(str::to_string),
+    };
+    let payload = serde_json::to_value(SessionMetaLine {
+        meta,
+        git: git_info,
     })?;
 
     let lines = [

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

@@ -2,7 +2,8 @@ use anyhow::Result;
 use app_test_support::McpProcess;
 use app_test_support::create_final_assistant_message_sse_response;
 use app_test_support::create_mock_chat_completions_server;
-use app_test_support::create_shell_sse_response;
+use app_test_support::create_shell_command_sse_response;
+use app_test_support::format_with_current_shell;
 use app_test_support::to_response;
 use codex_app_server_protocol::AddConversationListenerParams;
 use codex_app_server_protocol::AddConversationSubscriptionResponse;
@@ -22,10 +23,10 @@ use codex_app_server_protocol::SendUserTurnResponse;
 use codex_app_server_protocol::ServerRequest;
 use codex_core::protocol::AskForApproval;
 use codex_core::protocol::SandboxPolicy;
-use codex_core::protocol_config_types::ReasoningEffort;
 use codex_core::protocol_config_types::ReasoningSummary;
 use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
 use codex_protocol::config_types::SandboxMode;
+use codex_protocol::openai_models::ReasoningEffort;
 use codex_protocol::parse_command::ParsedCommand;
 use codex_protocol::protocol::Event;
 use codex_protocol::protocol::EventMsg;
@@ -56,7 +57,7 @@ async fn test_codex_jsonrpc_conversation_flow() -> Result<()> {
     // Create a mock model server that immediately ends each turn.
     // Two turns are expected: initial session configure + one user message.
     let responses = vec![
-        create_shell_sse_response(
+        create_shell_command_sse_response(
             vec!["ls".to_string()],
             Some(&working_directory),
             Some(5000),
@@ -175,7 +176,7 @@ async fn test_send_user_turn_changes_approval_policy_behavior() -> Result<()> {
 
     // Mock server will request a python shell call for the first and second turn, then finish.
     let responses = vec![
-        create_shell_sse_response(
+        create_shell_command_sse_response(
             vec![
                 "python3".to_string(),
                 "-c".to_string(),
@@ -186,7 +187,7 @@ async fn test_send_user_turn_changes_approval_policy_behavior() -> Result<()> {
             "call1",
         )?,
         create_final_assistant_message_sse_response("done 1")?,
-        create_shell_sse_response(
+        create_shell_command_sse_response(
             vec![
                 "python3".to_string(),
                 "-c".to_string(),
@@ -267,11 +268,7 @@ async fn test_send_user_turn_changes_approval_policy_behavior() -> Result<()> {
         ExecCommandApprovalParams {
             conversation_id,
             call_id: "call1".to_string(),
-            command: vec![
-                "python3".to_string(),
-                "-c".to_string(),
-                "print(42)".to_string(),
-            ],
+            command: format_with_current_shell("python3 -c 'print(42)'"),
             cwd: working_directory.clone(),
             reason: None,
             risk: None,
@@ -353,23 +350,15 @@ async fn test_send_user_turn_updates_sandbox_and_cwd_between_turns() -> Result<(
     std::fs::create_dir(&second_cwd)?;
 
     let responses = vec![
-        create_shell_sse_response(
-            vec![
-                "bash".to_string(),
-                "-lc".to_string(),
-                "echo first turn".to_string(),
-            ],
+        create_shell_command_sse_response(
+            vec!["echo".to_string(), "first".to_string(), "turn".to_string()],
             None,
             Some(5000),
             "call-first",
         )?,
         create_final_assistant_message_sse_response("done first")?,
-        create_shell_sse_response(
-            vec![
-                "bash".to_string(),
-                "-lc".to_string(),
-                "echo second turn".to_string(),
-            ],
+        create_shell_command_sse_response(
+            vec!["echo".to_string(), "second".to_string(), "turn".to_string()],
             None,
             Some(5000),
             "call-second",
@@ -481,13 +470,9 @@ async fn test_send_user_turn_updates_sandbox_and_cwd_between_turns() -> Result<(
         exec_begin.cwd, second_cwd,
         "exec turn should run from updated cwd"
     );
+    let expected_command = format_with_current_shell("echo second turn");
     assert_eq!(
-        exec_begin.command,
-        vec![
-            "bash".to_string(),
-            "-lc".to_string(),
-            "echo second turn".to_string()
-        ],
+        exec_begin.command, expected_command,
         "exec turn should run expected command"
     );
 

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

@@ -10,10 +10,10 @@ use codex_app_server_protocol::Tools;
 use codex_app_server_protocol::UserSavedConfig;
 use codex_core::protocol::AskForApproval;
 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::Verbosity;
+use codex_protocol::openai_models::ReasoningEffort;
 use pretty_assertions::assert_eq;
 use std::collections::HashMap;
 use std::path::Path;
@@ -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-max"
 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-max".into()),
             model_reasoning_effort: Some(ReasoningEffort::High),
             model_reasoning_summary: Some(ReasoningSummary::Detailed),
             model_verbosity: Some(Verbosity::Medium),

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

@@ -19,7 +19,7 @@ use tokio::time::timeout;
 
 use app_test_support::McpProcess;
 use app_test_support::create_mock_chat_completions_server;
-use app_test_support::create_shell_sse_response;
+use app_test_support::create_shell_command_sse_response;
 use app_test_support::to_response;
 
 const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
@@ -56,7 +56,7 @@ async fn shell_command_interruption() -> anyhow::Result<()> {
     std::fs::create_dir(&working_directory)?;
 
     // Create mock server with a single SSE response: the long sleep command
-    let server = create_mock_chat_completions_server(vec![create_shell_sse_response(
+    let server = create_mock_chat_completions_server(vec![create_shell_command_sse_response(
         shell_command.clone(),
         Some(&working_directory),
         Some(10_000), // 10 seconds timeout in ms

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

@@ -31,6 +31,7 @@ async fn test_list_and_resume_conversations() -> Result<()> {
         "2025-01-02T12:00:00Z",
         "Hello A",
         Some("openai"),
+        None,
     )?;
     create_fake_rollout(
         codex_home.path(),
@@ -38,6 +39,7 @@ async fn test_list_and_resume_conversations() -> Result<()> {
         "2025-01-01T13:00:00Z",
         "Hello B",
         Some("openai"),
+        None,
     )?;
     create_fake_rollout(
         codex_home.path(),
@@ -45,6 +47,7 @@ async fn test_list_and_resume_conversations() -> Result<()> {
         "2025-01-01T12:00:00Z",
         "Hello C",
         None,
+        None,
     )?;
 
     let mut mcp = McpProcess::new(codex_home.path()).await?;
@@ -105,6 +108,7 @@ async fn test_list_and_resume_conversations() -> Result<()> {
         "2025-01-01T11:30:00Z",
         "Hello TP",
         Some("test-provider"),
+        None,
     )?;
 
     // Filtering by model provider should return only matching sessions.

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

@@ -272,40 +272,45 @@ async fn read_raw_response_item(
     mcp: &mut McpProcess,
     conversation_id: ConversationId,
 ) -> ResponseItem {
-    let raw_notification: JSONRPCNotification = timeout(
-        DEFAULT_READ_TIMEOUT,
-        mcp.read_stream_until_notification_message("codex/event/raw_response_item"),
-    )
-    .await
-    .expect("codex/event/raw_response_item notification timeout")
-    .expect("codex/event/raw_response_item notification resp");
+    loop {
+        let raw_notification: JSONRPCNotification = timeout(
+            DEFAULT_READ_TIMEOUT,
+            mcp.read_stream_until_notification_message("codex/event/raw_response_item"),
+        )
+        .await
+        .expect("codex/event/raw_response_item notification timeout")
+        .expect("codex/event/raw_response_item notification resp");
 
-    let serde_json::Value::Object(params) = raw_notification
-        .params
-        .expect("codex/event/raw_response_item should have params")
-    else {
-        panic!("codex/event/raw_response_item should have params");
-    };
+        let serde_json::Value::Object(params) = raw_notification
+            .params
+            .expect("codex/event/raw_response_item should have params")
+        else {
+            panic!("codex/event/raw_response_item should have params");
+        };
 
-    let conversation_id_value = params
-        .get("conversationId")
-        .and_then(|value| value.as_str())
-        .expect("raw response item should include conversationId");
+        let conversation_id_value = params
+            .get("conversationId")
+            .and_then(|value| value.as_str())
+            .expect("raw response item should include conversationId");
 
-    assert_eq!(
-        conversation_id_value,
-        conversation_id.to_string(),
-        "raw response item conversation mismatch"
-    );
+        assert_eq!(
+            conversation_id_value,
+            conversation_id.to_string(),
+            "raw response item conversation mismatch"
+        );
 
-    let msg_value = params
-        .get("msg")
-        .cloned()
-        .expect("raw response item should include msg payload");
+        let msg_value = params
+            .get("msg")
+            .cloned()
+            .expect("raw response item should include msg payload");
 
-    let event: RawResponseItemEvent =
-        serde_json::from_value(msg_value).expect("deserialize raw response item");
-    event.item
+        // Ghost snapshots are produced concurrently and may arrive before the model reply.
+        let event: RawResponseItemEvent =
+            serde_json::from_value(msg_value).expect("deserialize raw response item");
+        if !matches!(event.item, ResponseItem::GhostSnapshot { .. }) {
+            return event.item;
+        }
+    }
 }
 
 fn assert_instructions_message(item: &ResponseItem) {

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-max"
 model_reasoning_effort = "medium"
 "#,
     )

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

@@ -0,0 +1,347 @@
+use anyhow::Result;
+use app_test_support::McpProcess;
+use app_test_support::to_response;
+use codex_app_server_protocol::ConfigBatchWriteParams;
+use codex_app_server_protocol::ConfigEdit;
+use codex_app_server_protocol::ConfigLayerName;
+use codex_app_server_protocol::ConfigReadParams;
+use codex_app_server_protocol::ConfigReadResponse;
+use codex_app_server_protocol::ConfigValueWriteParams;
+use codex_app_server_protocol::ConfigWriteResponse;
+use codex_app_server_protocol::JSONRPCError;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::MergeStrategy;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::WriteStatus;
+use pretty_assertions::assert_eq;
+use serde_json::json;
+use tempfile::TempDir;
+use tokio::time::timeout;
+
+const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
+
+fn write_config(codex_home: &TempDir, contents: &str) -> Result<()> {
+    Ok(std::fs::write(
+        codex_home.path().join("config.toml"),
+        contents,
+    )?)
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn config_read_returns_effective_and_layers() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    write_config(
+        &codex_home,
+        r#"
+model = "gpt-user"
+sandbox_mode = "workspace-write"
+"#,
+    )?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let request_id = mcp
+        .send_config_read_request(ConfigReadParams {
+            include_layers: true,
+        })
+        .await?;
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
+    )
+    .await??;
+    let ConfigReadResponse {
+        config,
+        origins,
+        layers,
+    } = to_response(resp)?;
+
+    assert_eq!(config.get("model"), Some(&json!("gpt-user")));
+    assert_eq!(
+        origins.get("model").expect("origin").name,
+        ConfigLayerName::User
+    );
+    let layers = layers.expect("layers present");
+    assert_eq!(layers.len(), 2);
+    assert_eq!(layers[0].name, ConfigLayerName::SessionFlags);
+    assert_eq!(layers[1].name, ConfigLayerName::User);
+
+    Ok(())
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn config_read_includes_system_layer_and_overrides() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    write_config(
+        &codex_home,
+        r#"
+model = "gpt-user"
+approval_policy = "on-request"
+sandbox_mode = "workspace-write"
+
+[sandbox_workspace_write]
+writable_roots = ["/user"]
+network_access = true
+"#,
+    )?;
+
+    let managed_path = codex_home.path().join("managed_config.toml");
+    std::fs::write(
+        &managed_path,
+        r#"
+model = "gpt-system"
+approval_policy = "never"
+
+[sandbox_workspace_write]
+writable_roots = ["/system"]
+"#,
+    )?;
+
+    let managed_path_str = managed_path.display().to_string();
+
+    let mut mcp = McpProcess::new_with_env(
+        codex_home.path(),
+        &[("CODEX_MANAGED_CONFIG_PATH", Some(&managed_path_str))],
+    )
+    .await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let request_id = mcp
+        .send_config_read_request(ConfigReadParams {
+            include_layers: true,
+        })
+        .await?;
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
+    )
+    .await??;
+    let ConfigReadResponse {
+        config,
+        origins,
+        layers,
+    } = to_response(resp)?;
+
+    assert_eq!(config.get("model"), Some(&json!("gpt-system")));
+    assert_eq!(
+        origins.get("model").expect("origin").name,
+        ConfigLayerName::System
+    );
+
+    assert_eq!(config.get("approval_policy"), Some(&json!("never")));
+    assert_eq!(
+        origins.get("approval_policy").expect("origin").name,
+        ConfigLayerName::System
+    );
+
+    assert_eq!(config.get("sandbox_mode"), Some(&json!("workspace-write")));
+    assert_eq!(
+        origins.get("sandbox_mode").expect("origin").name,
+        ConfigLayerName::User
+    );
+
+    assert_eq!(
+        config
+            .get("sandbox_workspace_write")
+            .and_then(|v| v.get("writable_roots")),
+        Some(&json!(["/system"]))
+    );
+    assert_eq!(
+        origins
+            .get("sandbox_workspace_write.writable_roots.0")
+            .expect("origin")
+            .name,
+        ConfigLayerName::System
+    );
+
+    assert_eq!(
+        config
+            .get("sandbox_workspace_write")
+            .and_then(|v| v.get("network_access")),
+        Some(&json!(true))
+    );
+    assert_eq!(
+        origins
+            .get("sandbox_workspace_write.network_access")
+            .expect("origin")
+            .name,
+        ConfigLayerName::User
+    );
+
+    let layers = layers.expect("layers present");
+    assert_eq!(layers.len(), 3);
+    assert_eq!(layers[0].name, ConfigLayerName::System);
+    assert_eq!(layers[1].name, ConfigLayerName::SessionFlags);
+    assert_eq!(layers[2].name, ConfigLayerName::User);
+
+    Ok(())
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn config_value_write_replaces_value() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    write_config(
+        &codex_home,
+        r#"
+model = "gpt-old"
+"#,
+    )?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let read_id = mcp
+        .send_config_read_request(ConfigReadParams {
+            include_layers: false,
+        })
+        .await?;
+    let read_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(read_id)),
+    )
+    .await??;
+    let read: ConfigReadResponse = to_response(read_resp)?;
+    let expected_version = read.origins.get("model").map(|m| m.version.clone());
+
+    let write_id = mcp
+        .send_config_value_write_request(ConfigValueWriteParams {
+            file_path: codex_home.path().join("config.toml").display().to_string(),
+            key_path: "model".to_string(),
+            value: json!("gpt-new"),
+            merge_strategy: MergeStrategy::Replace,
+            expected_version,
+        })
+        .await?;
+    let write_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(write_id)),
+    )
+    .await??;
+    let write: ConfigWriteResponse = to_response(write_resp)?;
+
+    assert_eq!(write.status, WriteStatus::Ok);
+    assert!(write.overridden_metadata.is_none());
+
+    let verify_id = mcp
+        .send_config_read_request(ConfigReadParams {
+            include_layers: false,
+        })
+        .await?;
+    let verify_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(verify_id)),
+    )
+    .await??;
+    let verify: ConfigReadResponse = to_response(verify_resp)?;
+    assert_eq!(verify.config.get("model"), Some(&json!("gpt-new")));
+
+    Ok(())
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn config_value_write_rejects_version_conflict() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    write_config(
+        &codex_home,
+        r#"
+model = "gpt-old"
+"#,
+    )?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let write_id = mcp
+        .send_config_value_write_request(ConfigValueWriteParams {
+            file_path: codex_home.path().join("config.toml").display().to_string(),
+            key_path: "model".to_string(),
+            value: json!("gpt-new"),
+            merge_strategy: MergeStrategy::Replace,
+            expected_version: Some("sha256:stale".to_string()),
+        })
+        .await?;
+
+    let err: JSONRPCError = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_error_message(RequestId::Integer(write_id)),
+    )
+    .await??;
+    let code = err
+        .error
+        .data
+        .as_ref()
+        .and_then(|d| d.get("config_write_error_code"))
+        .and_then(|v| v.as_str());
+    assert_eq!(code, Some("configVersionConflict"));
+
+    Ok(())
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn config_batch_write_applies_multiple_edits() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    write_config(&codex_home, "")?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let batch_id = mcp
+        .send_config_batch_write_request(ConfigBatchWriteParams {
+            file_path: codex_home.path().join("config.toml").display().to_string(),
+            edits: vec![
+                ConfigEdit {
+                    key_path: "sandbox_mode".to_string(),
+                    value: json!("workspace-write"),
+                    merge_strategy: MergeStrategy::Replace,
+                },
+                ConfigEdit {
+                    key_path: "sandbox_workspace_write".to_string(),
+                    value: json!({
+                        "writable_roots": ["/tmp"],
+                        "network_access": false
+                    }),
+                    merge_strategy: MergeStrategy::Replace,
+                },
+            ],
+            expected_version: None,
+        })
+        .await?;
+    let batch_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(batch_id)),
+    )
+    .await??;
+    let batch_write: ConfigWriteResponse = to_response(batch_resp)?;
+    assert_eq!(batch_write.status, WriteStatus::Ok);
+
+    let read_id = mcp
+        .send_config_read_request(ConfigReadParams {
+            include_layers: false,
+        })
+        .await?;
+    let read_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(read_id)),
+    )
+    .await??;
+    let read: ConfigReadResponse = to_response(read_resp)?;
+    assert_eq!(
+        read.config.get("sandbox_mode"),
+        Some(&json!("workspace-write"))
+    );
+    assert_eq!(
+        read.config
+            .get("sandbox_workspace_write")
+            .and_then(|v| v.get("writable_roots")),
+        Some(&json!(["/tmp"]))
+    );
+    assert_eq!(
+        read.config
+            .get("sandbox_workspace_write")
+            .and_then(|v| v.get("network_access")),
+        Some(&json!(false))
+    );
+
+    Ok(())
+}

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

@@ -1,6 +1,8 @@
 mod account;
+mod config_rpc;
 mod model_list;
 mod rate_limits;
+mod review;
 mod thread_archive;
 mod thread_list;
 mod thread_resume;

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

@@ -11,7 +11,7 @@ use codex_app_server_protocol::ModelListParams;
 use codex_app_server_protocol::ModelListResponse;
 use codex_app_server_protocol::ReasoningEffortOption;
 use codex_app_server_protocol::RequestId;
-use codex_protocol::config_types::ReasoningEffort;
+use codex_protocol::openai_models::ReasoningEffort;
 use pretty_assertions::assert_eq;
 use tempfile::TempDir;
 use tokio::time::timeout;
@@ -46,9 +46,36 @@ 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-max".to_string(),
+            model: "gpt-5.1-codex-max".to_string(),
+            display_name: "gpt-5.1-codex-max".to_string(),
+            description: "Latest Codex-optimized flagship for deep and fast reasoning.".to_string(),
+            supported_reasoning_efforts: vec![
+                ReasoningEffortOption {
+                    reasoning_effort: ReasoningEffort::Low,
+                    description: "Fast responses with lighter reasoning".to_string(),
+                },
+                ReasoningEffortOption {
+                    reasoning_effort: ReasoningEffort::Medium,
+                    description: "Balances speed and reasoning depth for everyday tasks"
+                        .to_string(),
+                },
+                ReasoningEffortOption {
+                    reasoning_effort: ReasoningEffort::High,
+                    description: "Maximizes reasoning depth for complex problems".to_string(),
+                },
+                ReasoningEffortOption {
+                    reasoning_effort: ReasoningEffort::XHigh,
+                    description: "Extra high reasoning depth for complex problems".to_string(),
+                },
+            ],
+            default_reasoning_effort: ReasoningEffort::Medium,
+            is_default: true,
+        },
+        Model {
+            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 {
@@ -66,18 +93,33 @@ async fn list_models_returns_all_models_with_large_limit() -> Result<()> {
                 },
             ],
             default_reasoning_effort: ReasoningEffort::Medium,
-            is_default: true,
+            is_default: false,
         },
         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 +174,7 @@ 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-max");
     let next_cursor = first_cursor.ok_or_else(|| anyhow!("cursor for second page"))?;
 
     let second_request = mcp
@@ -154,8 +196,52 @@ 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");
+    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-codex-mini");
+    let fourth_cursor = third_cursor.ok_or_else(|| anyhow!("cursor for fourth page"))?;
+
+    let fourth_request = mcp
+        .send_list_models_request(ModelListParams {
+            limit: Some(1),
+            cursor: Some(fourth_cursor.clone()),
+        })
+        .await?;
+
+    let fourth_response: JSONRPCResponse = timeout(
+        DEFAULT_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(fourth_request)),
+    )
+    .await??;
+
+    let ModelListResponse {
+        data: fourth_items,
+        next_cursor: fourth_cursor,
+    } = to_response::<ModelListResponse>(fourth_response)?;
+
+    assert_eq!(fourth_items.len(), 1);
+    assert_eq!(fourth_items[0].id, "gpt-5.1");
+    assert!(fourth_cursor.is_none());
     Ok(())
 }
 

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

@@ -152,6 +152,7 @@ async fn get_account_rate_limits_returns_snapshot() -> Result<()> {
                 window_duration_mins: Some(1440),
                 resets_at: Some(secondary_reset_timestamp),
             }),
+            credits: None,
         },
     };
     assert_eq!(received, expected);

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

@@ -0,0 +1,329 @@
+use anyhow::Result;
+use app_test_support::McpProcess;
+use app_test_support::create_final_assistant_message_sse_response;
+use app_test_support::create_mock_chat_completions_server_unchecked;
+use app_test_support::to_response;
+use codex_app_server_protocol::ItemCompletedNotification;
+use codex_app_server_protocol::ItemStartedNotification;
+use codex_app_server_protocol::JSONRPCError;
+use codex_app_server_protocol::JSONRPCNotification;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::ReviewDelivery;
+use codex_app_server_protocol::ReviewStartParams;
+use codex_app_server_protocol::ReviewStartResponse;
+use codex_app_server_protocol::ReviewTarget;
+use codex_app_server_protocol::ThreadItem;
+use codex_app_server_protocol::ThreadStartParams;
+use codex_app_server_protocol::ThreadStartResponse;
+use codex_app_server_protocol::TurnStatus;
+use serde_json::json;
+use tempfile::TempDir;
+use tokio::time::timeout;
+
+const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
+const INVALID_REQUEST_ERROR_CODE: i64 = -32600;
+
+#[tokio::test]
+async fn review_start_runs_review_turn_and_emits_code_review_item() -> Result<()> {
+    let review_payload = json!({
+        "findings": [
+            {
+                "title": "Prefer Stylize helpers",
+                "body": "Use .dim()/.bold() chaining instead of manual Style.",
+                "confidence_score": 0.9,
+                "priority": 1,
+                "code_location": {
+                    "absolute_file_path": "/tmp/file.rs",
+                    "line_range": {"start": 10, "end": 20}
+                }
+            }
+        ],
+        "overall_correctness": "good",
+        "overall_explanation": "Looks solid overall with minor polish suggested.",
+        "overall_confidence_score": 0.75
+    })
+    .to_string();
+    let responses = vec![create_final_assistant_message_sse_response(
+        &review_payload,
+    )?];
+    let server = create_mock_chat_completions_server_unchecked(responses).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 thread_id = start_default_thread(&mut mcp).await?;
+
+    let review_req = mcp
+        .send_review_start_request(ReviewStartParams {
+            thread_id: thread_id.clone(),
+            delivery: Some(ReviewDelivery::Inline),
+            target: ReviewTarget::Commit {
+                sha: "1234567deadbeef".to_string(),
+                title: Some("Tidy UI colors".to_string()),
+            },
+        })
+        .await?;
+    let review_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(review_req)),
+    )
+    .await??;
+    let ReviewStartResponse {
+        turn,
+        review_thread_id,
+    } = to_response::<ReviewStartResponse>(review_resp)?;
+    assert_eq!(review_thread_id, thread_id.clone());
+    let turn_id = turn.id.clone();
+    assert_eq!(turn.status, TurnStatus::InProgress);
+
+    // Confirm we see the EnteredReviewMode marker on the main thread.
+    let mut saw_entered_review_mode = false;
+    for _ in 0..10 {
+        let item_started: JSONRPCNotification = timeout(
+            DEFAULT_READ_TIMEOUT,
+            mcp.read_stream_until_notification_message("item/started"),
+        )
+        .await??;
+        let started: ItemStartedNotification =
+            serde_json::from_value(item_started.params.expect("params must be present"))?;
+        match started.item {
+            ThreadItem::EnteredReviewMode { id, review } => {
+                assert_eq!(id, turn_id);
+                assert_eq!(review, "commit 1234567: Tidy UI colors");
+                saw_entered_review_mode = true;
+                break;
+            }
+            _ => continue,
+        }
+    }
+    assert!(
+        saw_entered_review_mode,
+        "did not observe enteredReviewMode item"
+    );
+
+    // Confirm we see the ExitedReviewMode marker (with review text)
+    // on the same turn. Ignore any other items the stream surfaces.
+    let mut review_body: Option<String> = None;
+    for _ in 0..10 {
+        let review_notif: JSONRPCNotification = timeout(
+            DEFAULT_READ_TIMEOUT,
+            mcp.read_stream_until_notification_message("item/completed"),
+        )
+        .await??;
+        let completed: ItemCompletedNotification =
+            serde_json::from_value(review_notif.params.expect("params must be present"))?;
+        match completed.item {
+            ThreadItem::ExitedReviewMode { id, review } => {
+                assert_eq!(id, turn_id);
+                review_body = Some(review);
+                break;
+            }
+            _ => continue,
+        }
+    }
+
+    let review = review_body.expect("did not observe a code review item");
+    assert!(review.contains("Prefer Stylize helpers"));
+    assert!(review.contains("/tmp/file.rs:10-20"));
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn review_start_rejects_empty_base_branch() -> Result<()> {
+    let server = create_mock_chat_completions_server_unchecked(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 thread_id = start_default_thread(&mut mcp).await?;
+
+    let request_id = mcp
+        .send_review_start_request(ReviewStartParams {
+            thread_id,
+            delivery: Some(ReviewDelivery::Inline),
+            target: ReviewTarget::BaseBranch {
+                branch: "   ".to_string(),
+            },
+        })
+        .await?;
+    let error: JSONRPCError = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_error_message(RequestId::Integer(request_id)),
+    )
+    .await??;
+    assert_eq!(error.error.code, INVALID_REQUEST_ERROR_CODE);
+    assert!(
+        error.error.message.contains("branch must not be empty"),
+        "unexpected message: {}",
+        error.error.message
+    );
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn review_start_with_detached_delivery_returns_new_thread_id() -> Result<()> {
+    let review_payload = json!({
+        "findings": [],
+        "overall_correctness": "ok",
+        "overall_explanation": "detached review",
+        "overall_confidence_score": 0.5
+    })
+    .to_string();
+    let responses = vec![create_final_assistant_message_sse_response(
+        &review_payload,
+    )?];
+    let server = create_mock_chat_completions_server_unchecked(responses).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 thread_id = start_default_thread(&mut mcp).await?;
+
+    let review_req = mcp
+        .send_review_start_request(ReviewStartParams {
+            thread_id: thread_id.clone(),
+            delivery: Some(ReviewDelivery::Detached),
+            target: ReviewTarget::Custom {
+                instructions: "detached review".to_string(),
+            },
+        })
+        .await?;
+    let review_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(review_req)),
+    )
+    .await??;
+    let ReviewStartResponse {
+        turn,
+        review_thread_id,
+    } = to_response::<ReviewStartResponse>(review_resp)?;
+
+    assert_eq!(turn.status, TurnStatus::InProgress);
+    assert_ne!(
+        review_thread_id, thread_id,
+        "detached review should run on a different thread"
+    );
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn review_start_rejects_empty_commit_sha() -> Result<()> {
+    let server = create_mock_chat_completions_server_unchecked(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 thread_id = start_default_thread(&mut mcp).await?;
+
+    let request_id = mcp
+        .send_review_start_request(ReviewStartParams {
+            thread_id,
+            delivery: Some(ReviewDelivery::Inline),
+            target: ReviewTarget::Commit {
+                sha: "\t".to_string(),
+                title: None,
+            },
+        })
+        .await?;
+    let error: JSONRPCError = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_error_message(RequestId::Integer(request_id)),
+    )
+    .await??;
+    assert_eq!(error.error.code, INVALID_REQUEST_ERROR_CODE);
+    assert!(
+        error.error.message.contains("sha must not be empty"),
+        "unexpected message: {}",
+        error.error.message
+    );
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn review_start_rejects_empty_custom_instructions() -> Result<()> {
+    let server = create_mock_chat_completions_server_unchecked(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 thread_id = start_default_thread(&mut mcp).await?;
+
+    let request_id = mcp
+        .send_review_start_request(ReviewStartParams {
+            thread_id,
+            delivery: Some(ReviewDelivery::Inline),
+            target: ReviewTarget::Custom {
+                instructions: "\n\n".to_string(),
+            },
+        })
+        .await?;
+    let error: JSONRPCError = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_error_message(RequestId::Integer(request_id)),
+    )
+    .await??;
+    assert_eq!(error.error.code, INVALID_REQUEST_ERROR_CODE);
+    assert!(
+        error
+            .error
+            .message
+            .contains("instructions must not be empty"),
+        "unexpected message: {}",
+        error.error.message
+    );
+
+    Ok(())
+}
+
+async fn start_default_thread(mcp: &mut McpProcess) -> Result<String> {
+    let thread_req = mcp
+        .send_thread_start_request(ThreadStartParams {
+            model: Some("mock-model".to_string()),
+            ..Default::default()
+        })
+        .await?;
+    let thread_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(thread_req)),
+    )
+    .await??;
+    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(thread_resp)?;
+    Ok(thread.id)
+}
+
+fn create_config_toml(codex_home: &std::path::Path, server_uri: &str) -> std::io::Result<()> {
+    let config_toml = codex_home.join("config.toml");
+    std::fs::write(
+        config_toml,
+        format!(
+            r#"
+model = "mock-model"
+approval_policy = "never"
+sandbox_mode = "read-only"
+
+model_provider = "mock_provider"
+
+[model_providers.mock_provider]
+name = "Mock provider"
+base_url = "{server_uri}/v1"
+wire_api = "chat"
+request_max_retries = 0
+stream_max_retries = 0
+"#
+        ),
+    )
+}

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

@@ -35,7 +35,7 @@ async fn thread_archive_moves_rollout_into_archived_directory() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
     )
     .await??;
-    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(start_resp)?;
+    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(start_resp)?;
     assert!(!thread.id.is_empty());
 
     // Locate the rollout path recorded for this thread id.

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

@@ -2,14 +2,16 @@ use anyhow::Result;
 use app_test_support::McpProcess;
 use app_test_support::create_fake_rollout;
 use app_test_support::to_response;
+use codex_app_server_protocol::GitInfo as ApiGitInfo;
 use codex_app_server_protocol::JSONRPCResponse;
 use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::SessionSource;
 use codex_app_server_protocol::ThreadListParams;
 use codex_app_server_protocol::ThreadListResponse;
-use serde_json::json;
+use codex_protocol::protocol::GitInfo as CoreGitInfo;
+use std::path::PathBuf;
 use tempfile::TempDir;
 use tokio::time::timeout;
-use uuid::Uuid;
 
 const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
 
@@ -26,7 +28,7 @@ async fn thread_list_basic_empty() -> Result<()> {
         .send_thread_list_request(ThreadListParams {
             cursor: None,
             limit: Some(10),
-            model_providers: None,
+            model_providers: Some(vec!["mock_provider".to_string()]),
         })
         .await?;
     let list_resp: JSONRPCResponse = timeout(
@@ -65,6 +67,7 @@ async fn thread_list_pagination_next_cursor_none_on_last_page() -> Result<()> {
         "2025-01-02T12:00:00Z",
         "Hello",
         Some("mock_provider"),
+        None,
     )?;
     let _b = create_fake_rollout(
         codex_home.path(),
@@ -72,6 +75,7 @@ async fn thread_list_pagination_next_cursor_none_on_last_page() -> Result<()> {
         "2025-01-01T13:00:00Z",
         "Hello",
         Some("mock_provider"),
+        None,
     )?;
     let _c = create_fake_rollout(
         codex_home.path(),
@@ -79,6 +83,7 @@ async fn thread_list_pagination_next_cursor_none_on_last_page() -> Result<()> {
         "2025-01-01T12:00:00Z",
         "Hello",
         Some("mock_provider"),
+        None,
     )?;
 
     let mut mcp = McpProcess::new(codex_home.path()).await?;
@@ -106,6 +111,10 @@ async fn thread_list_pagination_next_cursor_none_on_last_page() -> Result<()> {
         assert_eq!(thread.preview, "Hello");
         assert_eq!(thread.model_provider, "mock_provider");
         assert!(thread.created_at > 0);
+        assert_eq!(thread.cwd, PathBuf::from("/"));
+        assert_eq!(thread.cli_version, "0.0.0");
+        assert_eq!(thread.source, SessionSource::Cli);
+        assert_eq!(thread.git_info, None);
     }
     let cursor1 = cursor1.expect("expected nextCursor on first page");
 
@@ -131,6 +140,10 @@ async fn thread_list_pagination_next_cursor_none_on_last_page() -> Result<()> {
         assert_eq!(thread.preview, "Hello");
         assert_eq!(thread.model_provider, "mock_provider");
         assert!(thread.created_at > 0);
+        assert_eq!(thread.cwd, PathBuf::from("/"));
+        assert_eq!(thread.cli_version, "0.0.0");
+        assert_eq!(thread.source, SessionSource::Cli);
+        assert_eq!(thread.git_info, None);
     }
     assert_eq!(cursor2, None, "expected nextCursor to be null on last page");
 
@@ -149,47 +162,16 @@ async fn thread_list_respects_provider_filter() -> Result<()> {
         "2025-01-02T10:00:00Z",
         "X",
         Some("mock_provider"),
+        None,
     )?; // 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"),
+        None,
+    )?;
 
     let mut mcp = McpProcess::new(codex_home.path()).await?;
     timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
@@ -215,6 +197,63 @@ async fn thread_list_respects_provider_filter() -> Result<()> {
     assert_eq!(thread.model_provider, "other_provider");
     let expected_ts = chrono::DateTime::parse_from_rfc3339("2025-01-02T11:00:00Z")?.timestamp();
     assert_eq!(thread.created_at, expected_ts);
+    assert_eq!(thread.cwd, PathBuf::from("/"));
+    assert_eq!(thread.cli_version, "0.0.0");
+    assert_eq!(thread.source, SessionSource::Cli);
+    assert_eq!(thread.git_info, None);
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn thread_list_includes_git_info() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_minimal_config(codex_home.path())?;
+
+    let git_info = CoreGitInfo {
+        commit_hash: Some("abc123".to_string()),
+        branch: Some("main".to_string()),
+        repository_url: Some("https://example.com/repo.git".to_string()),
+    };
+    let conversation_id = create_fake_rollout(
+        codex_home.path(),
+        "2025-02-01T09-00-00",
+        "2025-02-01T09:00:00Z",
+        "Git info preview",
+        Some("mock_provider"),
+        Some(git_info),
+    )?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let list_id = mcp
+        .send_thread_list_request(ThreadListParams {
+            cursor: None,
+            limit: Some(10),
+            model_providers: Some(vec!["mock_provider".to_string()]),
+        })
+        .await?;
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(list_id)),
+    )
+    .await??;
+    let ThreadListResponse { data, .. } = to_response::<ThreadListResponse>(resp)?;
+    let thread = data
+        .iter()
+        .find(|t| t.id == conversation_id)
+        .expect("expected thread for created rollout");
+
+    let expected_git = ApiGitInfo {
+        sha: Some("abc123".to_string()),
+        branch: Some("main".to_string()),
+        origin_url: Some("https://example.com/repo.git".to_string()),
+    };
+    assert_eq!(thread.git_info, Some(expected_git));
+    assert_eq!(thread.source, SessionSource::Cli);
+    assert_eq!(thread.cwd, PathBuf::from("/"));
+    assert_eq!(thread.cli_version, "0.0.0");
 
     Ok(())
 }

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

@@ -1,20 +1,28 @@
 use anyhow::Result;
 use app_test_support::McpProcess;
+use app_test_support::create_fake_rollout;
 use app_test_support::create_mock_chat_completions_server;
 use app_test_support::to_response;
 use codex_app_server_protocol::JSONRPCResponse;
 use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::SessionSource;
+use codex_app_server_protocol::ThreadItem;
 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_app_server_protocol::TurnStatus;
+use codex_app_server_protocol::UserInput;
+use codex_protocol::models::ContentItem;
+use codex_protocol::models::ResponseItem;
+use std::path::PathBuf;
 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 +33,7 @@ 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-max".to_string()),
             ..Default::default()
         })
         .await?;
@@ -34,12 +42,13 @@ async fn thread_resume_returns_existing_thread() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
     )
     .await??;
-    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(start_resp)?;
+    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(start_resp)?;
 
     // Resume it via v2 API.
     let resume_id = mcp
         .send_thread_resume_request(ThreadResumeParams {
             thread_id: thread.id.clone(),
+            ..Default::default()
         })
         .await?;
     let resume_resp: JSONRPCResponse = timeout(
@@ -47,13 +56,181 @@ async fn thread_resume_returns_existing_thread() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(resume_id)),
     )
     .await??;
-    let ThreadResumeResponse { thread: resumed } =
-        to_response::<ThreadResumeResponse>(resume_resp)?;
+    let ThreadResumeResponse {
+        thread: resumed, ..
+    } = to_response::<ThreadResumeResponse>(resume_resp)?;
     assert_eq!(resumed, thread);
 
     Ok(())
 }
 
+#[tokio::test]
+async fn thread_resume_returns_rollout_history() -> Result<()> {
+    let server = create_mock_chat_completions_server(vec![]).await;
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), &server.uri())?;
+
+    let preview = "Saved user message";
+    let conversation_id = create_fake_rollout(
+        codex_home.path(),
+        "2025-01-05T12-00-00",
+        "2025-01-05T12:00:00Z",
+        preview,
+        Some("mock_provider"),
+        None,
+    )?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let resume_id = mcp
+        .send_thread_resume_request(ThreadResumeParams {
+            thread_id: conversation_id.clone(),
+            ..Default::default()
+        })
+        .await?;
+    let resume_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(resume_id)),
+    )
+    .await??;
+    let ThreadResumeResponse { thread, .. } = to_response::<ThreadResumeResponse>(resume_resp)?;
+
+    assert_eq!(thread.id, conversation_id);
+    assert_eq!(thread.preview, preview);
+    assert_eq!(thread.model_provider, "mock_provider");
+    assert!(thread.path.is_absolute());
+    assert_eq!(thread.cwd, PathBuf::from("/"));
+    assert_eq!(thread.cli_version, "0.0.0");
+    assert_eq!(thread.source, SessionSource::Cli);
+    assert_eq!(thread.git_info, None);
+
+    assert_eq!(
+        thread.turns.len(),
+        1,
+        "expected rollouts to include one turn"
+    );
+    let turn = &thread.turns[0];
+    assert_eq!(turn.status, TurnStatus::Completed);
+    assert_eq!(turn.items.len(), 1, "expected user message item");
+    match &turn.items[0] {
+        ThreadItem::UserMessage { content, .. } => {
+            assert_eq!(
+                content,
+                &vec![UserInput::Text {
+                    text: preview.to_string()
+                }]
+            );
+        }
+        other => panic!("expected user message item, got {other:?}"),
+    }
+
+    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-max".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-max".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,
+        model_provider,
+        ..
+    } = to_response::<ThreadResumeResponse>(resume_resp)?;
+    assert!(!resumed.id.is_empty());
+    assert_eq!(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?;
@@ -40,13 +40,17 @@ async fn thread_start_creates_thread_and_emits_started() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(req_id)),
     )
     .await??;
-    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(resp)?;
+    let ThreadStartResponse {
+        thread,
+        model_provider,
+        ..
+    } = to_response::<ThreadStartResponse>(resp)?;
     assert!(!thread.id.is_empty(), "thread id should not be empty");
     assert!(
         thread.preview.is_empty(),
         "new threads should start with an empty preview"
     );
-    assert_eq!(thread.model_provider, "mock_provider");
+    assert_eq!(model_provider, "mock_provider");
     assert!(
         thread.created_at > 0,
         "created_at should be a positive UNIX timestamp"

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

@@ -3,16 +3,19 @@
 use anyhow::Result;
 use app_test_support::McpProcess;
 use app_test_support::create_mock_chat_completions_server;
-use app_test_support::create_shell_sse_response;
+use app_test_support::create_shell_command_sse_response;
 use app_test_support::to_response;
+use codex_app_server_protocol::JSONRPCNotification;
 use codex_app_server_protocol::JSONRPCResponse;
 use codex_app_server_protocol::RequestId;
 use codex_app_server_protocol::ThreadStartParams;
 use codex_app_server_protocol::ThreadStartResponse;
+use codex_app_server_protocol::TurnCompletedNotification;
 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::TurnStatus;
 use codex_app_server_protocol::UserInput as V2UserInput;
 use tempfile::TempDir;
 use tokio::time::timeout;
@@ -38,7 +41,7 @@ async fn turn_interrupt_aborts_running_turn() -> Result<()> {
     std::fs::create_dir(&working_directory)?;
 
     // Mock server: long-running shell command then (after abort) nothing else needed.
-    let server = create_mock_chat_completions_server(vec![create_shell_sse_response(
+    let server = create_mock_chat_completions_server(vec![create_shell_command_sse_response(
         shell_command.clone(),
         Some(&working_directory),
         Some(10_000),
@@ -62,7 +65,7 @@ async fn turn_interrupt_aborts_running_turn() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(thread_req)),
     )
     .await??;
-    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(thread_resp)?;
+    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(thread_resp)?;
 
     // Start a turn that triggers a long-running command.
     let turn_req = mcp
@@ -85,10 +88,11 @@ async fn turn_interrupt_aborts_running_turn() -> Result<()> {
     // Give the command a brief moment to start.
     tokio::time::sleep(std::time::Duration::from_secs(1)).await;
 
+    let thread_id = thread.id.clone();
     // Interrupt the in-progress turn by id (v2 API).
     let interrupt_id = mcp
         .send_turn_interrupt_request(TurnInterruptParams {
-            thread_id: thread.id,
+            thread_id: thread_id.clone(),
             turn_id: turn.id,
         })
         .await?;
@@ -99,7 +103,19 @@ async fn turn_interrupt_aborts_running_turn() -> Result<()> {
     .await??;
     let _resp: TurnInterruptResponse = to_response::<TurnInterruptResponse>(interrupt_resp)?;
 
-    // No fields to assert on; successful deserialization confirms proper response shape.
+    let completed_notif: JSONRPCNotification = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("turn/completed"),
+    )
+    .await??;
+    let completed: TurnCompletedNotification = serde_json::from_value(
+        completed_notif
+            .params
+            .expect("turn/completed params must be present"),
+    )?;
+    assert_eq!(completed.thread_id, thread_id);
+    assert_eq!(completed.turn.status, TurnStatus::Interrupted);
+
     Ok(())
 }
 

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

@@ -1,25 +1,37 @@
 use anyhow::Result;
 use app_test_support::McpProcess;
+use app_test_support::create_apply_patch_sse_response;
+use app_test_support::create_exec_command_sse_response;
 use app_test_support::create_final_assistant_message_sse_response;
 use app_test_support::create_mock_chat_completions_server;
 use app_test_support::create_mock_chat_completions_server_unchecked;
-use app_test_support::create_shell_sse_response;
+use app_test_support::create_shell_command_sse_response;
+use app_test_support::format_with_current_shell_display;
 use app_test_support::to_response;
+use codex_app_server_protocol::ApprovalDecision;
+use codex_app_server_protocol::CommandExecutionRequestApprovalResponse;
+use codex_app_server_protocol::CommandExecutionStatus;
+use codex_app_server_protocol::FileChangeOutputDeltaNotification;
+use codex_app_server_protocol::FileChangeRequestApprovalResponse;
+use codex_app_server_protocol::ItemCompletedNotification;
+use codex_app_server_protocol::ItemStartedNotification;
 use codex_app_server_protocol::JSONRPCNotification;
 use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::PatchApplyStatus;
+use codex_app_server_protocol::PatchChangeKind;
 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::TurnCompletedNotification;
 use codex_app_server_protocol::TurnStartParams;
 use codex_app_server_protocol::TurnStartResponse;
 use codex_app_server_protocol::TurnStartedNotification;
+use codex_app_server_protocol::TurnStatus;
 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 codex_protocol::openai_models::ReasoningEffort;
 use core_test_support::skip_if_no_network;
 use pretty_assertions::assert_eq;
 use std::path::Path;
@@ -57,7 +69,7 @@ async fn turn_start_emits_notifications_and_accepts_model_override() -> Result<(
         mcp.read_stream_until_response_message(RequestId::Integer(thread_req)),
     )
     .await??;
-    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(thread_resp)?;
+    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(thread_resp)?;
 
     // Start a turn with only input and thread_id set (no overrides).
     let turn_req = mcp
@@ -85,6 +97,7 @@ async fn turn_start_emits_notifications_and_accepts_model_override() -> Result<(
     .await??;
     let started: TurnStartedNotification =
         serde_json::from_value(notif.params.expect("params must be present"))?;
+    assert_eq!(started.thread_id, thread.id);
     assert_eq!(
         started.turn.status,
         codex_app_server_protocol::TurnStatus::InProgress
@@ -118,13 +131,18 @@ async fn turn_start_emits_notifications_and_accepts_model_override() -> Result<(
     )
     .await??;
 
-    // And we should ultimately get a task_complete without having to add a
-    // legacy conversation listener explicitly (auto-attached by thread/start).
-    let _task_complete: JSONRPCNotification = timeout(
+    let completed_notif: JSONRPCNotification = timeout(
         DEFAULT_READ_TIMEOUT,
-        mcp.read_stream_until_notification_message("codex/event/task_complete"),
+        mcp.read_stream_until_notification_message("turn/completed"),
     )
     .await??;
+    let completed: TurnCompletedNotification = serde_json::from_value(
+        completed_notif
+            .params
+            .expect("turn/completed params must be present"),
+    )?;
+    assert_eq!(completed.thread_id, thread.id);
+    assert_eq!(completed.turn.status, TurnStatus::Completed);
 
     Ok(())
 }
@@ -157,7 +175,7 @@ async fn turn_start_accepts_local_image_input() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(thread_req)),
     )
     .await??;
-    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(thread_resp)?;
+    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(thread_resp)?;
 
     let image_path = codex_home.path().join("image.png");
     // No need to actually write the file; we just exercise the input path.
@@ -191,7 +209,7 @@ async fn turn_start_exec_approval_toggle_v2() -> Result<()> {
     // Mock server: first turn requests a shell call (elicitation), then completes.
     // Second turn same, but we'll set approval_policy=never to avoid elicitation.
     let responses = vec![
-        create_shell_sse_response(
+        create_shell_command_sse_response(
             vec![
                 "python3".to_string(),
                 "-c".to_string(),
@@ -202,7 +220,7 @@ async fn turn_start_exec_approval_toggle_v2() -> Result<()> {
             "call1",
         )?,
         create_final_assistant_message_sse_response("done 1")?,
-        create_shell_sse_response(
+        create_shell_command_sse_response(
             vec![
                 "python3".to_string(),
                 "-c".to_string(),
@@ -233,9 +251,9 @@ async fn turn_start_exec_approval_toggle_v2() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
     )
     .await??;
-    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(start_resp)?;
+    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 +276,10 @@ 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(
@@ -280,6 +292,11 @@ async fn turn_start_exec_approval_toggle_v2() -> Result<()> {
         mcp.read_stream_until_notification_message("codex/event/task_complete"),
     )
     .await??;
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("turn/completed"),
+    )
+    .await??;
 
     // Second turn with approval_policy=never should not elicit approval
     let second_turn_id = mcp
@@ -302,7 +319,151 @@ 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"),
+    )
+    .await??;
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("turn/completed"),
+    )
+    .await??;
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn turn_start_exec_approval_decline_v2() -> Result<()> {
+    skip_if_no_network!(Ok(()));
+
+    let tmp = TempDir::new()?;
+    let codex_home = tmp.path().to_path_buf();
+    let workspace = tmp.path().join("workspace");
+    std::fs::create_dir(&workspace)?;
+
+    let responses = vec![
+        create_shell_command_sse_response(
+            vec![
+                "python3".to_string(),
+                "-c".to_string(),
+                "print(42)".to_string(),
+            ],
+            None,
+            Some(5000),
+            "call-decline",
+        )?,
+        create_final_assistant_message_sse_response("done")?,
+    ];
+    let server = create_mock_chat_completions_server(responses).await;
+    create_config_toml(codex_home.as_path(), &server.uri(), "untrusted")?;
+
+    let mut mcp = McpProcess::new(codex_home.as_path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let start_id = mcp
+        .send_thread_start_request(ThreadStartParams {
+            model: Some("mock-model".to_string()),
+            ..Default::default()
+        })
+        .await?;
+    let start_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
+    )
+    .await??;
+    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(start_resp)?;
+
+    let turn_id = mcp
+        .send_turn_start_request(TurnStartParams {
+            thread_id: thread.id.clone(),
+            input: vec![V2UserInput::Text {
+                text: "run python".to_string(),
+            }],
+            cwd: Some(workspace.clone()),
+            ..Default::default()
+        })
+        .await?;
+    let turn_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(turn_id)),
+    )
+    .await??;
+    let TurnStartResponse { turn } = to_response::<TurnStartResponse>(turn_resp)?;
+
+    let started_command_execution = timeout(DEFAULT_READ_TIMEOUT, async {
+        loop {
+            let started_notif = mcp
+                .read_stream_until_notification_message("item/started")
+                .await?;
+            let started: ItemStartedNotification =
+                serde_json::from_value(started_notif.params.clone().expect("item/started params"))?;
+            if let ThreadItem::CommandExecution { .. } = started.item {
+                return Ok::<ThreadItem, anyhow::Error>(started.item);
+            }
+        }
+    })
+    .await??;
+    let ThreadItem::CommandExecution { id, status, .. } = started_command_execution else {
+        unreachable!("loop ensures we break on command execution items");
+    };
+    assert_eq!(id, "call-decline");
+    assert_eq!(status, CommandExecutionStatus::InProgress);
+
+    let server_req = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_request_message(),
+    )
+    .await??;
+    let ServerRequest::CommandExecutionRequestApproval { request_id, params } = server_req else {
+        panic!("expected CommandExecutionRequestApproval request")
+    };
+    assert_eq!(params.item_id, "call-decline");
+    assert_eq!(params.thread_id, thread.id);
+    assert_eq!(params.turn_id, turn.id);
+
+    mcp.send_response(
+        request_id,
+        serde_json::to_value(CommandExecutionRequestApprovalResponse {
+            decision: ApprovalDecision::Decline,
+            accept_settings: None,
+        })?,
+    )
+    .await?;
+
+    let completed_command_execution = timeout(DEFAULT_READ_TIMEOUT, async {
+        loop {
+            let completed_notif = mcp
+                .read_stream_until_notification_message("item/completed")
+                .await?;
+            let completed: ItemCompletedNotification = serde_json::from_value(
+                completed_notif
+                    .params
+                    .clone()
+                    .expect("item/completed params"),
+            )?;
+            if let ThreadItem::CommandExecution { .. } = completed.item {
+                return Ok::<ThreadItem, anyhow::Error>(completed.item);
+            }
+        }
+    })
+    .await??;
+    let ThreadItem::CommandExecution {
+        id,
+        status,
+        exit_code,
+        aggregated_output,
+        ..
+    } = completed_command_execution
+    else {
+        unreachable!("loop ensures we break on command execution items");
+    };
+    assert_eq!(id, "call-decline");
+    assert_eq!(status, CommandExecutionStatus::Declined);
+    assert!(exit_code.is_none());
+    assert!(aggregated_output.is_none());
+
     timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_notification_message("codex/event/task_complete"),
@@ -314,8 +475,6 @@ 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()?;
@@ -329,23 +488,15 @@ async fn turn_start_updates_sandbox_and_cwd_between_turns_v2() -> Result<()> {
     std::fs::create_dir(&second_cwd)?;
 
     let responses = vec![
-        create_shell_sse_response(
-            vec![
-                "bash".to_string(),
-                "-lc".to_string(),
-                "echo first turn".to_string(),
-            ],
+        create_shell_command_sse_response(
+            vec!["echo".to_string(), "first".to_string(), "turn".to_string()],
             None,
             Some(5000),
             "call-first",
         )?,
         create_final_assistant_message_sse_response("done first")?,
-        create_shell_sse_response(
-            vec![
-                "bash".to_string(),
-                "-lc".to_string(),
-                "echo second turn".to_string(),
-            ],
+        create_shell_command_sse_response(
+            vec!["echo".to_string(), "second".to_string(), "turn".to_string()],
             None,
             Some(5000),
             "call-second",
@@ -370,7 +521,7 @@ async fn turn_start_updates_sandbox_and_cwd_between_turns_v2() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
     )
     .await??;
-    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(start_resp)?;
+    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(start_resp)?;
 
     // first turn with workspace-write sandbox and first_cwd
     let first_turn = mcp
@@ -424,29 +575,36 @@ 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);
+    let expected_command = format_with_current_shell_display("echo second turn");
+    assert_eq!(command, expected_command);
+    assert_eq!(status, CommandExecutionStatus::InProgress);
 
     timeout(
         DEFAULT_READ_TIMEOUT,
@@ -457,6 +615,448 @@ async fn turn_start_updates_sandbox_and_cwd_between_turns_v2() -> Result<()> {
     Ok(())
 }
 
+#[tokio::test]
+async fn turn_start_file_change_approval_v2() -> Result<()> {
+    skip_if_no_network!(Ok(()));
+
+    let tmp = TempDir::new()?;
+    let codex_home = tmp.path().join("codex_home");
+    std::fs::create_dir(&codex_home)?;
+    let workspace = tmp.path().join("workspace");
+    std::fs::create_dir(&workspace)?;
+
+    let patch = r#"*** Begin Patch
+*** Add File: README.md
++new line
+*** End Patch
+"#;
+    let responses = vec![
+        create_apply_patch_sse_response(patch, "patch-call")?,
+        create_final_assistant_message_sse_response("patch applied")?,
+    ];
+    let server = create_mock_chat_completions_server(responses).await;
+    create_config_toml(&codex_home, &server.uri(), "untrusted")?;
+
+    let mut mcp = McpProcess::new(&codex_home).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let start_req = mcp
+        .send_thread_start_request(ThreadStartParams {
+            model: Some("mock-model".to_string()),
+            cwd: Some(workspace.to_string_lossy().into_owned()),
+            ..Default::default()
+        })
+        .await?;
+    let start_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(start_req)),
+    )
+    .await??;
+    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(start_resp)?;
+
+    let turn_req = mcp
+        .send_turn_start_request(TurnStartParams {
+            thread_id: thread.id.clone(),
+            input: vec![V2UserInput::Text {
+                text: "apply patch".into(),
+            }],
+            cwd: Some(workspace.clone()),
+            ..Default::default()
+        })
+        .await?;
+    let turn_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(turn_req)),
+    )
+    .await??;
+    let TurnStartResponse { turn } = to_response::<TurnStartResponse>(turn_resp)?;
+
+    let started_file_change = timeout(DEFAULT_READ_TIMEOUT, async {
+        loop {
+            let started_notif = mcp
+                .read_stream_until_notification_message("item/started")
+                .await?;
+            let started: ItemStartedNotification =
+                serde_json::from_value(started_notif.params.clone().expect("item/started params"))?;
+            if let ThreadItem::FileChange { .. } = started.item {
+                return Ok::<ThreadItem, anyhow::Error>(started.item);
+            }
+        }
+    })
+    .await??;
+    let ThreadItem::FileChange {
+        ref id,
+        status,
+        ref changes,
+    } = started_file_change
+    else {
+        unreachable!("loop ensures we break on file change items");
+    };
+    assert_eq!(id, "patch-call");
+    assert_eq!(status, PatchApplyStatus::InProgress);
+    let started_changes = changes.clone();
+
+    let server_req = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_request_message(),
+    )
+    .await??;
+    let ServerRequest::FileChangeRequestApproval { request_id, params } = server_req else {
+        panic!("expected FileChangeRequestApproval request")
+    };
+    assert_eq!(params.item_id, "patch-call");
+    assert_eq!(params.thread_id, thread.id);
+    assert_eq!(params.turn_id, turn.id);
+    let expected_readme_path = workspace.join("README.md");
+    let expected_readme_path = expected_readme_path.to_string_lossy().into_owned();
+    pretty_assertions::assert_eq!(
+        started_changes,
+        vec![codex_app_server_protocol::FileUpdateChange {
+            path: expected_readme_path.clone(),
+            kind: PatchChangeKind::Add,
+            diff: "new line\n".to_string(),
+        }]
+    );
+
+    mcp.send_response(
+        request_id,
+        serde_json::to_value(FileChangeRequestApprovalResponse {
+            decision: ApprovalDecision::Accept,
+        })?,
+    )
+    .await?;
+
+    let output_delta_notif = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("item/fileChange/outputDelta"),
+    )
+    .await??;
+    let output_delta: FileChangeOutputDeltaNotification = serde_json::from_value(
+        output_delta_notif
+            .params
+            .clone()
+            .expect("item/fileChange/outputDelta params"),
+    )?;
+    assert_eq!(output_delta.thread_id, thread.id);
+    assert_eq!(output_delta.turn_id, turn.id);
+    assert_eq!(output_delta.item_id, "patch-call");
+    assert!(
+        !output_delta.delta.is_empty(),
+        "expected delta to be non-empty, got: {}",
+        output_delta.delta
+    );
+
+    let completed_file_change = timeout(DEFAULT_READ_TIMEOUT, async {
+        loop {
+            let completed_notif = mcp
+                .read_stream_until_notification_message("item/completed")
+                .await?;
+            let completed: ItemCompletedNotification = serde_json::from_value(
+                completed_notif
+                    .params
+                    .clone()
+                    .expect("item/completed params"),
+            )?;
+            if let ThreadItem::FileChange { .. } = completed.item {
+                return Ok::<ThreadItem, anyhow::Error>(completed.item);
+            }
+        }
+    })
+    .await??;
+    let ThreadItem::FileChange { ref id, status, .. } = completed_file_change else {
+        unreachable!("loop ensures we break on file change items");
+    };
+    assert_eq!(id, "patch-call");
+    assert_eq!(status, PatchApplyStatus::Completed);
+
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("codex/event/task_complete"),
+    )
+    .await??;
+
+    let readme_contents = std::fs::read_to_string(expected_readme_path)?;
+    assert_eq!(readme_contents, "new line\n");
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn turn_start_file_change_approval_decline_v2() -> Result<()> {
+    skip_if_no_network!(Ok(()));
+
+    let tmp = TempDir::new()?;
+    let codex_home = tmp.path().join("codex_home");
+    std::fs::create_dir(&codex_home)?;
+    let workspace = tmp.path().join("workspace");
+    std::fs::create_dir(&workspace)?;
+
+    let patch = r#"*** Begin Patch
+*** Add File: README.md
++new line
+*** End Patch
+"#;
+    let responses = vec![
+        create_apply_patch_sse_response(patch, "patch-call")?,
+        create_final_assistant_message_sse_response("patch declined")?,
+    ];
+    let server = create_mock_chat_completions_server(responses).await;
+    create_config_toml(&codex_home, &server.uri(), "untrusted")?;
+
+    let mut mcp = McpProcess::new(&codex_home).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let start_req = mcp
+        .send_thread_start_request(ThreadStartParams {
+            model: Some("mock-model".to_string()),
+            cwd: Some(workspace.to_string_lossy().into_owned()),
+            ..Default::default()
+        })
+        .await?;
+    let start_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(start_req)),
+    )
+    .await??;
+    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(start_resp)?;
+
+    let turn_req = mcp
+        .send_turn_start_request(TurnStartParams {
+            thread_id: thread.id.clone(),
+            input: vec![V2UserInput::Text {
+                text: "apply patch".into(),
+            }],
+            cwd: Some(workspace.clone()),
+            ..Default::default()
+        })
+        .await?;
+    let turn_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(turn_req)),
+    )
+    .await??;
+    let TurnStartResponse { turn } = to_response::<TurnStartResponse>(turn_resp)?;
+
+    let started_file_change = timeout(DEFAULT_READ_TIMEOUT, async {
+        loop {
+            let started_notif = mcp
+                .read_stream_until_notification_message("item/started")
+                .await?;
+            let started: ItemStartedNotification =
+                serde_json::from_value(started_notif.params.clone().expect("item/started params"))?;
+            if let ThreadItem::FileChange { .. } = started.item {
+                return Ok::<ThreadItem, anyhow::Error>(started.item);
+            }
+        }
+    })
+    .await??;
+    let ThreadItem::FileChange {
+        ref id,
+        status,
+        ref changes,
+    } = started_file_change
+    else {
+        unreachable!("loop ensures we break on file change items");
+    };
+    assert_eq!(id, "patch-call");
+    assert_eq!(status, PatchApplyStatus::InProgress);
+    let started_changes = changes.clone();
+
+    let server_req = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_request_message(),
+    )
+    .await??;
+    let ServerRequest::FileChangeRequestApproval { request_id, params } = server_req else {
+        panic!("expected FileChangeRequestApproval request")
+    };
+    assert_eq!(params.item_id, "patch-call");
+    assert_eq!(params.thread_id, thread.id);
+    assert_eq!(params.turn_id, turn.id);
+    let expected_readme_path = workspace.join("README.md");
+    let expected_readme_path_str = expected_readme_path.to_string_lossy().into_owned();
+    pretty_assertions::assert_eq!(
+        started_changes,
+        vec![codex_app_server_protocol::FileUpdateChange {
+            path: expected_readme_path_str.clone(),
+            kind: PatchChangeKind::Add,
+            diff: "new line\n".to_string(),
+        }]
+    );
+
+    mcp.send_response(
+        request_id,
+        serde_json::to_value(FileChangeRequestApprovalResponse {
+            decision: ApprovalDecision::Decline,
+        })?,
+    )
+    .await?;
+
+    let completed_file_change = timeout(DEFAULT_READ_TIMEOUT, async {
+        loop {
+            let completed_notif = mcp
+                .read_stream_until_notification_message("item/completed")
+                .await?;
+            let completed: ItemCompletedNotification = serde_json::from_value(
+                completed_notif
+                    .params
+                    .clone()
+                    .expect("item/completed params"),
+            )?;
+            if let ThreadItem::FileChange { .. } = completed.item {
+                return Ok::<ThreadItem, anyhow::Error>(completed.item);
+            }
+        }
+    })
+    .await??;
+    let ThreadItem::FileChange { ref id, status, .. } = completed_file_change else {
+        unreachable!("loop ensures we break on file change items");
+    };
+    assert_eq!(id, "patch-call");
+    assert_eq!(status, PatchApplyStatus::Declined);
+
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("codex/event/task_complete"),
+    )
+    .await??;
+
+    assert!(
+        !expected_readme_path.exists(),
+        "declined patch should not be applied"
+    );
+
+    Ok(())
+}
+
+#[tokio::test]
+#[cfg_attr(windows, ignore = "process id reporting differs on Windows")]
+async fn command_execution_notifications_include_process_id() -> Result<()> {
+    skip_if_no_network!(Ok(()));
+
+    let responses = vec![
+        create_exec_command_sse_response("uexec-1")?,
+        create_final_assistant_message_sse_response("done")?,
+    ];
+    let server = create_mock_chat_completions_server(responses).await;
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), &server.uri(), "never")?;
+    let config_toml = codex_home.path().join("config.toml");
+    let mut config_contents = std::fs::read_to_string(&config_toml)?;
+    config_contents.push_str(
+        r#"
+[features]
+unified_exec = true
+"#,
+    );
+    std::fs::write(&config_toml, config_contents)?;
+
+    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("mock-model".to_string()),
+            ..Default::default()
+        })
+        .await?;
+    let start_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
+    )
+    .await??;
+    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(start_resp)?;
+
+    let turn_id = mcp
+        .send_turn_start_request(TurnStartParams {
+            thread_id: thread.id.clone(),
+            input: vec![V2UserInput::Text {
+                text: "run a command".to_string(),
+            }],
+            ..Default::default()
+        })
+        .await?;
+    let turn_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(turn_id)),
+    )
+    .await??;
+    let TurnStartResponse { turn: _turn } = to_response::<TurnStartResponse>(turn_resp)?;
+
+    let started_command = timeout(DEFAULT_READ_TIMEOUT, async {
+        loop {
+            let notif = mcp
+                .read_stream_until_notification_message("item/started")
+                .await?;
+            let started: ItemStartedNotification = serde_json::from_value(
+                notif
+                    .params
+                    .clone()
+                    .expect("item/started should include params"),
+            )?;
+            if let ThreadItem::CommandExecution { .. } = started.item {
+                return Ok::<ThreadItem, anyhow::Error>(started.item);
+            }
+        }
+    })
+    .await??;
+    let ThreadItem::CommandExecution {
+        id,
+        process_id: started_process_id,
+        status,
+        ..
+    } = started_command
+    else {
+        unreachable!("loop ensures we break on command execution items");
+    };
+    assert_eq!(id, "uexec-1");
+    assert_eq!(status, CommandExecutionStatus::InProgress);
+    let started_process_id = started_process_id.expect("process id should be present");
+
+    let completed_command = timeout(DEFAULT_READ_TIMEOUT, async {
+        loop {
+            let notif = mcp
+                .read_stream_until_notification_message("item/completed")
+                .await?;
+            let completed: ItemCompletedNotification = serde_json::from_value(
+                notif
+                    .params
+                    .clone()
+                    .expect("item/completed should include params"),
+            )?;
+            if let ThreadItem::CommandExecution { .. } = completed.item {
+                return Ok::<ThreadItem, anyhow::Error>(completed.item);
+            }
+        }
+    })
+    .await??;
+    let ThreadItem::CommandExecution {
+        id: completed_id,
+        process_id: completed_process_id,
+        status: completed_status,
+        exit_code,
+        ..
+    } = completed_command
+    else {
+        unreachable!("loop ensures we break on command execution items");
+    };
+    assert_eq!(completed_id, "uexec-1");
+    assert_eq!(completed_status, CommandExecutionStatus::Completed);
+    assert_eq!(exit_code, Some(0));
+    assert_eq!(
+        completed_process_id.as_deref(),
+        Some(started_process_id.as_str())
+    );
+
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("turn/completed"),
+    )
+    .await??;
+
+    Ok(())
+}
+
 // Helper to create a config.toml pointing at the mock model server.
 fn create_config_toml(
     codex_home: &Path,

codex-rs/apply-patch/Cargo.toml

@@ -1,7 +1,8 @@
 [package]
-edition = "2024"
 name = "codex-apply-patch"
-version = { workspace = true }
+version.workspace = true
+edition.workspace = true
+license.workspace = true
 
 [lib]
 name = "codex_apply_patch"

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

@@ -31,6 +31,13 @@ pub const APPLY_PATCH_TOOL_INSTRUCTIONS: &str = include_str!("../apply_patch_too
 
 const APPLY_PATCH_COMMANDS: [&str; 2] = ["apply_patch", "applypatch"];
 
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+enum ApplyPatchShell {
+    Unix,
+    PowerShell,
+    Cmd,
+}
+
 #[derive(Debug, Error, PartialEq)]
 pub enum ApplyPatchError {
     #[error(transparent)]
@@ -96,6 +103,57 @@ pub struct ApplyPatchArgs {
     pub workdir: Option<String>,
 }
 
+fn classify_shell_name(shell: &str) -> Option<String> {
+    std::path::Path::new(shell)
+        .file_stem()
+        .and_then(|name| name.to_str())
+        .map(str::to_ascii_lowercase)
+}
+
+fn classify_shell(shell: &str, flag: &str) -> Option<ApplyPatchShell> {
+    classify_shell_name(shell).and_then(|name| match name.as_str() {
+        "bash" | "zsh" | "sh" if flag == "-lc" => Some(ApplyPatchShell::Unix),
+        "pwsh" | "powershell" if flag.eq_ignore_ascii_case("-command") => {
+            Some(ApplyPatchShell::PowerShell)
+        }
+        "cmd" if flag.eq_ignore_ascii_case("/c") => Some(ApplyPatchShell::Cmd),
+        _ => None,
+    })
+}
+
+fn can_skip_flag(shell: &str, flag: &str) -> bool {
+    classify_shell_name(shell).is_some_and(|name| {
+        matches!(name.as_str(), "pwsh" | "powershell") && flag.eq_ignore_ascii_case("-noprofile")
+    })
+}
+
+fn parse_shell_script(argv: &[String]) -> Option<(ApplyPatchShell, &str)> {
+    match argv {
+        [shell, flag, script] => classify_shell(shell, flag).map(|shell_type| {
+            let script = script.as_str();
+            (shell_type, script)
+        }),
+        [shell, skip_flag, flag, script] if can_skip_flag(shell, skip_flag) => {
+            classify_shell(shell, flag).map(|shell_type| {
+                let script = script.as_str();
+                (shell_type, script)
+            })
+        }
+        _ => None,
+    }
+}
+
+fn extract_apply_patch_from_shell(
+    shell: ApplyPatchShell,
+    script: &str,
+) -> std::result::Result<(String, Option<String>), ExtractHeredocError> {
+    match shell {
+        ApplyPatchShell::Unix | ApplyPatchShell::PowerShell | ApplyPatchShell::Cmd => {
+            extract_apply_patch_from_bash(script)
+        }
+    }
+}
+
 pub fn maybe_parse_apply_patch(argv: &[String]) -> MaybeApplyPatch {
     match argv {
         // Direct invocation: apply_patch <patch>
@@ -103,9 +161,9 @@ pub fn maybe_parse_apply_patch(argv: &[String]) -> MaybeApplyPatch {
             Ok(source) => MaybeApplyPatch::Body(source),
             Err(e) => MaybeApplyPatch::PatchParseError(e),
         },
-        // Bash heredoc form: (optional `cd <path> &&`) apply_patch <<'EOF' ...
-        [bash, flag, script] if bash == "bash" && flag == "-lc" => {
-            match extract_apply_patch_from_bash(script) {
+        // Shell heredoc form: (optional `cd <path> &&`) apply_patch <<'EOF' ...
+        _ => match parse_shell_script(argv) {
+            Some((shell, script)) => match extract_apply_patch_from_shell(shell, script) {
                 Ok((body, workdir)) => match parse_patch(&body) {
                     Ok(mut source) => {
                         source.workdir = workdir;
@@ -117,9 +175,9 @@ pub fn maybe_parse_apply_patch(argv: &[String]) -> MaybeApplyPatch {
                     MaybeApplyPatch::NotApplyPatch
                 }
                 Err(e) => MaybeApplyPatch::ShellParseError(e),
-            }
-        }
-        _ => MaybeApplyPatch::NotApplyPatch,
+            },
+            None => MaybeApplyPatch::NotApplyPatch,
+        },
     }
 }
 
@@ -214,24 +272,17 @@ impl ApplyPatchAction {
 /// cwd must be an absolute path so that we can resolve relative paths in the
 /// patch.
 pub fn maybe_parse_apply_patch_verified(argv: &[String], cwd: &Path) -> MaybeApplyPatchVerified {
-    // Detect a raw patch body passed directly as the command or as the body of a bash -lc
+    // Detect a raw patch body passed directly as the command or as the body of a shell
     // script. In these cases, report an explicit error rather than applying the patch.
-    match argv {
-        [body] => {
-            if parse_patch(body).is_ok() {
-                return MaybeApplyPatchVerified::CorrectnessError(
-                    ApplyPatchError::ImplicitInvocation,
-                );
-            }
-        }
-        [bash, flag, script] if bash == "bash" && flag == "-lc" => {
-            if parse_patch(script).is_ok() {
-                return MaybeApplyPatchVerified::CorrectnessError(
-                    ApplyPatchError::ImplicitInvocation,
-                );
-            }
-        }
-        _ => {}
+    if let [body] = argv
+        && parse_patch(body).is_ok()
+    {
+        return MaybeApplyPatchVerified::CorrectnessError(ApplyPatchError::ImplicitInvocation);
+    }
+    if let Some((_, script)) = parse_shell_script(argv)
+        && parse_patch(script).is_ok()
+    {
+        return MaybeApplyPatchVerified::CorrectnessError(ApplyPatchError::ImplicitInvocation);
     }
 
     match maybe_parse_apply_patch(argv) {
@@ -863,6 +914,22 @@ mod tests {
         strs_to_strings(&["bash", "-lc", script])
     }
 
+    fn args_powershell(script: &str) -> Vec<String> {
+        strs_to_strings(&["powershell.exe", "-Command", script])
+    }
+
+    fn args_powershell_no_profile(script: &str) -> Vec<String> {
+        strs_to_strings(&["powershell.exe", "-NoProfile", "-Command", script])
+    }
+
+    fn args_pwsh(script: &str) -> Vec<String> {
+        strs_to_strings(&["pwsh", "-NoProfile", "-Command", script])
+    }
+
+    fn args_cmd(script: &str) -> Vec<String> {
+        strs_to_strings(&["cmd.exe", "/c", script])
+    }
+
     fn heredoc_script(prefix: &str) -> String {
         format!(
             "{prefix}apply_patch <<'PATCH'\n*** Begin Patch\n*** Add File: foo\n+hi\n*** End Patch\nPATCH"
@@ -882,8 +949,7 @@ mod tests {
         }]
     }
 
-    fn assert_match(script: &str, expected_workdir: Option<&str>) {
-        let args = args_bash(script);
+    fn assert_match_args(args: Vec<String>, expected_workdir: Option<&str>) {
         match maybe_parse_apply_patch(&args) {
             MaybeApplyPatch::Body(ApplyPatchArgs { hunks, workdir, .. }) => {
                 assert_eq!(workdir.as_deref(), expected_workdir);
@@ -893,6 +959,11 @@ mod tests {
         }
     }
 
+    fn assert_match(script: &str, expected_workdir: Option<&str>) {
+        let args = args_bash(script);
+        assert_match_args(args, expected_workdir);
+    }
+
     fn assert_not_match(script: &str) {
         let args = args_bash(script);
         assert_matches!(
@@ -1006,6 +1077,28 @@ PATCH"#,
         }
     }
 
+    #[test]
+    fn test_powershell_heredoc() {
+        let script = heredoc_script("");
+        assert_match_args(args_powershell(&script), None);
+    }
+    #[test]
+    fn test_powershell_heredoc_no_profile() {
+        let script = heredoc_script("");
+        assert_match_args(args_powershell_no_profile(&script), None);
+    }
+    #[test]
+    fn test_pwsh_heredoc() {
+        let script = heredoc_script("");
+        assert_match_args(args_pwsh(&script), None);
+    }
+
+    #[test]
+    fn test_cmd_heredoc_with_cd() {
+        let script = heredoc_script("cd foo && ");
+        assert_match_args(args_cmd(&script), Some("foo"));
+    }
+
     #[test]
     fn test_heredoc_with_leading_cd() {
         assert_match(&heredoc_script("cd foo && "), Some("foo"));

codex-rs/arg0/Cargo.toml

@@ -1,7 +1,8 @@
 [package]
-edition = "2024"
 name = "codex-arg0"
-version = { workspace = true }
+version.workspace = true
+edition.workspace = true
+license.workspace = true
 
 [lib]
 name = "codex_arg0"

codex-rs/arg0/src/lib.rs

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

codex-rs/async-utils/Cargo.toml

@@ -1,7 +1,8 @@
 [package]
-edition.workspace = true
 name = "codex-async-utils"
 version.workspace = true
+edition.workspace = true
+license.workspace = true
 
 [lints]
 workspace = true

codex-rs/backend-client/Cargo.toml

@@ -1,7 +1,8 @@
 [package]
 name = "codex-backend-client"
-version = "0.0.0"
-edition = "2024"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
 publish = false
 
 [lib]

codex-rs/backend-client/src/client.rs

@@ -1,4 +1,5 @@
 use crate::types::CodeTaskDetailsResponse;
+use crate::types::CreditStatusDetails;
 use crate::types::PaginatedListTaskListItem;
 use crate::types::RateLimitStatusPayload;
 use crate::types::RateLimitWindowSnapshot;
@@ -6,6 +7,7 @@ use crate::types::TurnAttemptsSiblingTurnsResponse;
 use anyhow::Result;
 use codex_core::auth::CodexAuth;
 use codex_core::default_client::get_codex_user_agent;
+use codex_protocol::protocol::CreditsSnapshot;
 use codex_protocol::protocol::RateLimitSnapshot;
 use codex_protocol::protocol::RateLimitWindow;
 use reqwest::header::AUTHORIZATION;
@@ -272,19 +274,23 @@ impl Client {
 
     // rate limit helpers
     fn rate_limit_snapshot_from_payload(payload: RateLimitStatusPayload) -> RateLimitSnapshot {
-        let Some(details) = payload
+        let rate_limit_details = payload
             .rate_limit
-            .and_then(|inner| inner.map(|boxed| *boxed))
-        else {
-            return RateLimitSnapshot {
-                primary: None,
-                secondary: None,
-            };
+            .and_then(|inner| inner.map(|boxed| *boxed));
+
+        let (primary, secondary) = if let Some(details) = rate_limit_details {
+            (
+                Self::map_rate_limit_window(details.primary_window),
+                Self::map_rate_limit_window(details.secondary_window),
+            )
+        } else {
+            (None, None)
         };
 
         RateLimitSnapshot {
-            primary: Self::map_rate_limit_window(details.primary_window),
-            secondary: Self::map_rate_limit_window(details.secondary_window),
+            primary,
+            secondary,
+            credits: Self::map_credits(payload.credits),
         }
     }
 
@@ -306,6 +312,19 @@ impl Client {
         })
     }
 
+    fn map_credits(credits: Option<Option<Box<CreditStatusDetails>>>) -> Option<CreditsSnapshot> {
+        let details = match credits {
+            Some(Some(details)) => *details,
+            _ => return None,
+        };
+
+        Some(CreditsSnapshot {
+            has_credits: details.has_credits,
+            unlimited: details.unlimited,
+            balance: details.balance.and_then(|inner| inner),
+        })
+    }
+
     fn window_minutes_from_seconds(seconds: i32) -> Option<i64> {
         if seconds <= 0 {
             return None;

codex-rs/backend-client/src/types.rs

@@ -1,3 +1,4 @@
+pub use codex_backend_openapi_models::models::CreditStatusDetails;
 pub use codex_backend_openapi_models::models::PaginatedListTaskListItem;
 pub use codex_backend_openapi_models::models::PlanType;
 pub use codex_backend_openapi_models::models::RateLimitStatusDetails;

codex-rs/chatgpt/Cargo.toml

@@ -1,7 +1,8 @@
 [package]
-edition = "2024"
 name = "codex-chatgpt"
-version = { workspace = true }
+version.workspace = true
+edition.workspace = true
+license.workspace = true
 
 [lints]
 workspace = true

codex-rs/cli/Cargo.toml

@@ -1,7 +1,8 @@
 [package]
-edition = "2024"
 name = "codex-cli"
-version = { workspace = true }
+version.workspace = true
+edition.workspace = true
+license.workspace = true
 
 [[bin]]
 name = "codex"
@@ -26,6 +27,7 @@ codex-cloud-tasks = { path = "../cloud-tasks" }
 codex-common = { workspace = true, features = ["cli"] }
 codex-core = { workspace = true }
 codex-exec = { workspace = true }
+codex-execpolicy = { workspace = true }
 codex-login = { workspace = true }
 codex-mcp-server = { workspace = true }
 codex-process-hardening = { workspace = true }
@@ -35,7 +37,9 @@ codex-rmcp-client = { workspace = true }
 codex-stdio-to-uds = { workspace = true }
 codex-tui = { workspace = true }
 ctor = { workspace = true }
+libc = { workspace = true }
 owo-colors = { workspace = true }
+regex-lite = { workspace = true}
 serde_json = { workspace = true }
 supports-color = { workspace = true }
 toml = { workspace = true }
@@ -46,6 +50,7 @@ tokio = { workspace = true, features = [
     "rt-multi-thread",
     "signal",
 ] }
+tracing = { workspace = true }
 
 [target.'cfg(target_os = "windows")'.dependencies]
 codex_windows_sandbox = { package = "codex-windows-sandbox", path = "../windows-sandbox-rs" }

codex-rs/cli/src/debug_sandbox.rs

@@ -1,3 +1,8 @@
+#[cfg(target_os = "macos")]
+mod pid_tracker;
+#[cfg(target_os = "macos")]
+mod seatbelt;
+
 use std::path::PathBuf;
 
 use codex_common::CliConfigOverrides;
@@ -15,6 +20,9 @@ use crate::SeatbeltCommand;
 use crate::WindowsCommand;
 use crate::exit_status::handle_exit_status;
 
+#[cfg(target_os = "macos")]
+use seatbelt::DenialLogger;
+
 #[cfg(target_os = "macos")]
 pub async fn run_command_under_seatbelt(
     command: SeatbeltCommand,
@@ -22,6 +30,7 @@ pub async fn run_command_under_seatbelt(
 ) -> anyhow::Result<()> {
     let SeatbeltCommand {
         full_auto,
+        log_denials,
         config_overrides,
         command,
     } = command;
@@ -31,6 +40,7 @@ pub async fn run_command_under_seatbelt(
         config_overrides,
         codex_linux_sandbox_exe,
         SandboxType::Seatbelt,
+        log_denials,
     )
     .await
 }
@@ -58,6 +68,7 @@ pub async fn run_command_under_landlock(
         config_overrides,
         codex_linux_sandbox_exe,
         SandboxType::Landlock,
+        false,
     )
     .await
 }
@@ -77,6 +88,7 @@ pub async fn run_command_under_windows(
         config_overrides,
         codex_linux_sandbox_exe,
         SandboxType::Windows,
+        false,
     )
     .await
 }
@@ -94,6 +106,7 @@ async fn run_command_under_sandbox(
     config_overrides: CliConfigOverrides,
     codex_linux_sandbox_exe: Option<PathBuf>,
     sandbox_type: SandboxType,
+    log_denials: bool,
 ) -> anyhow::Result<()> {
     let sandbox_mode = create_sandbox_mode(full_auto);
     let config = Config::load_with_cli_overrides(
@@ -125,11 +138,7 @@ async fn run_command_under_sandbox(
         {
             use codex_windows_sandbox::run_windows_sandbox_capture;
 
-            let policy_str = match &config.sandbox_policy {
-                codex_core::protocol::SandboxPolicy::DangerFullAccess => "workspace-write",
-                codex_core::protocol::SandboxPolicy::ReadOnly => "read-only",
-                codex_core::protocol::SandboxPolicy::WorkspaceWrite { .. } => "workspace-write",
-            };
+            let policy_str = serde_json::to_string(&config.sandbox_policy)?;
 
             let sandbox_cwd = sandbox_policy_cwd.clone();
             let cwd_clone = cwd.clone();
@@ -140,13 +149,13 @@ async fn run_command_under_sandbox(
             // Preflight audit is invoked elsewhere at the appropriate times.
             let res = tokio::task::spawn_blocking(move || {
                 run_windows_sandbox_capture(
-                    policy_str,
+                    policy_str.as_str(),
                     &sandbox_cwd,
+                    base_dir.as_path(),
                     command_vec,
                     &cwd_clone,
                     env_map,
                     None,
-                    Some(base_dir.as_path()),
                 )
             })
             .await;
@@ -180,6 +189,11 @@ async fn run_command_under_sandbox(
         }
     }
 
+    #[cfg(target_os = "macos")]
+    let mut denial_logger = log_denials.then(DenialLogger::new).flatten();
+    #[cfg(not(target_os = "macos"))]
+    let _ = log_denials;
+
     let mut child = match sandbox_type {
         #[cfg(target_os = "macos")]
         SandboxType::Seatbelt => {
@@ -213,8 +227,27 @@ async fn run_command_under_sandbox(
             unreachable!("Windows sandbox should have been handled above");
         }
     };
+
+    #[cfg(target_os = "macos")]
+    if let Some(denial_logger) = &mut denial_logger {
+        denial_logger.on_child_spawn(&child);
+    }
+
     let status = child.wait().await?;
 
+    #[cfg(target_os = "macos")]
+    if let Some(denial_logger) = denial_logger {
+        let denials = denial_logger.finish().await;
+        eprintln!("\n=== Sandbox denials ===");
+        if denials.is_empty() {
+            eprintln!("None found.");
+        } else {
+            for seatbelt::SandboxDenial { name, capability } in denials {
+                eprintln!("({name}) {capability}");
+            }
+        }
+    }
+
     handle_exit_status(status);
 }
 

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

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

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

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

codex-rs/cli/src/lib.rs

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

codex-rs/cli/src/main.rs

@@ -18,6 +18,9 @@ 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_exec::Command as ExecCommand;
+use codex_exec::ReviewArgs;
+use codex_execpolicy::ExecPolicyCheckCommand;
 use codex_responses_api_proxy::Args as ResponsesApiProxyArgs;
 use codex_tui::AppExitInfo;
 use codex_tui::Cli as TuiCli;
@@ -71,6 +74,9 @@ enum Subcommand {
     #[clap(visible_alias = "e")]
     Exec(ExecCli),
 
+    /// Run a code review non-interactively.
+    Review(ReviewArgs),
+
     /// Manage login.
     Login(LoginCommand),
 
@@ -93,6 +99,10 @@ enum Subcommand {
     #[clap(visible_alias = "debug")]
     Sandbox(SandboxArgs),
 
+    /// Execpolicy tooling.
+    #[clap(hide = true)]
+    Execpolicy(ExecpolicyCommand),
+
     /// Apply the latest diff produced by Codex agent as a `git apply` to your local working tree.
     #[clap(visible_alias = "a")]
     Apply(ApplyCommand),
@@ -134,6 +144,10 @@ struct ResumeCommand {
     #[arg(long = "last", default_value_t = false, conflicts_with = "session_id")]
     last: bool,
 
+    /// Show all sessions (disables cwd filtering and shows CWD column).
+    #[arg(long = "all", default_value_t = false)]
+    all: bool,
+
     #[clap(flatten)]
     config_overrides: TuiCli,
 }
@@ -158,6 +172,19 @@ enum SandboxCommand {
     Windows(WindowsCommand),
 }
 
+#[derive(Debug, Parser)]
+struct ExecpolicyCommand {
+    #[command(subcommand)]
+    sub: ExecpolicySubcommand,
+}
+
+#[derive(Debug, clap::Subcommand)]
+enum ExecpolicySubcommand {
+    /// Check execpolicy files against a command.
+    #[clap(name = "check")]
+    Check(ExecPolicyCheckCommand),
+}
+
 #[derive(Debug, Parser)]
 struct LoginCommand {
     #[clap(skip)]
@@ -323,6 +350,10 @@ fn run_update_action(action: UpdateAction) -> anyhow::Result<()> {
     Ok(())
 }
 
+fn run_execpolicycheck(cmd: ExecPolicyCheckCommand) -> anyhow::Result<()> {
+    cmd.run()
+}
+
 #[derive(Debug, Default, Parser, Clone)]
 struct FeatureToggles {
     /// Enable a feature (repeatable). Equivalent to `-c features.<name>=true`.
@@ -423,6 +454,15 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
             );
             codex_exec::run_main(exec_cli, codex_linux_sandbox_exe).await?;
         }
+        Some(Subcommand::Review(review_args)) => {
+            let mut exec_cli = ExecCli::try_parse_from(["codex", "exec"])?;
+            exec_cli.command = Some(ExecCommand::Review(review_args));
+            prepend_config_flags(
+                &mut exec_cli.config_overrides,
+                root_config_overrides.clone(),
+            );
+            codex_exec::run_main(exec_cli, codex_linux_sandbox_exe).await?;
+        }
         Some(Subcommand::McpServer) => {
             codex_mcp_server::run_main(codex_linux_sandbox_exe, root_config_overrides).await?;
         }
@@ -448,6 +488,7 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
         Some(Subcommand::Resume(ResumeCommand {
             session_id,
             last,
+            all,
             config_overrides,
         })) => {
             interactive = finalize_resume_interactive(
@@ -455,6 +496,7 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
                 root_config_overrides.clone(),
                 session_id,
                 last,
+                all,
                 config_overrides,
             );
             let exit_info = codex_tui::run_main(interactive, codex_linux_sandbox_exe).await?;
@@ -543,6 +585,9 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
                 .await?;
             }
         },
+        Some(Subcommand::Execpolicy(ExecpolicyCommand { sub })) => match sub {
+            ExecpolicySubcommand::Check(cmd) => run_execpolicycheck(cmd)?,
+        },
         Some(Subcommand::Apply(mut apply_cli)) => {
             prepend_config_flags(
                 &mut apply_cli.config_overrides,
@@ -611,6 +656,7 @@ fn finalize_resume_interactive(
     root_config_overrides: CliConfigOverrides,
     session_id: Option<String>,
     last: bool,
+    show_all: bool,
     resume_cli: TuiCli,
 ) -> TuiCli {
     // Start with the parsed interactive CLI so resume shares the same
@@ -619,6 +665,7 @@ fn finalize_resume_interactive(
     interactive.resume_picker = resume_session_id.is_none() && !last;
     interactive.resume_last = last;
     interactive.resume_session_id = resume_session_id;
+    interactive.resume_show_all = show_all;
 
     // Merge resume-scoped flags and overrides with highest precedence.
     merge_resume_cli_flags(&mut interactive, resume_cli);
@@ -702,13 +749,21 @@ mod tests {
         let Subcommand::Resume(ResumeCommand {
             session_id,
             last,
+            all,
             config_overrides: resume_cli,
         }) = subcommand.expect("resume present")
         else {
             unreachable!()
         };
 
-        finalize_resume_interactive(interactive, root_overrides, session_id, last, resume_cli)
+        finalize_resume_interactive(
+            interactive,
+            root_overrides,
+            session_id,
+            last,
+            all,
+            resume_cli,
+        )
     }
 
     fn sample_exit_info(conversation: Option<&str>) -> AppExitInfo {
@@ -761,9 +816,9 @@ 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);
@@ -775,6 +830,7 @@ mod tests {
         assert!(interactive.resume_picker);
         assert!(!interactive.resume_last);
         assert_eq!(interactive.resume_session_id, None);
+        assert!(!interactive.resume_show_all);
     }
 
     #[test]
@@ -783,6 +839,7 @@ mod tests {
         assert!(!interactive.resume_picker);
         assert!(interactive.resume_last);
         assert_eq!(interactive.resume_session_id, None);
+        assert!(!interactive.resume_show_all);
     }
 
     #[test]
@@ -791,6 +848,14 @@ mod tests {
         assert!(!interactive.resume_picker);
         assert!(!interactive.resume_last);
         assert_eq!(interactive.resume_session_id.as_deref(), Some("1234"));
+        assert!(!interactive.resume_show_all);
+    }
+
+    #[test]
+    fn resume_all_flag_sets_show_all() {
+        let interactive = finalize_from_args(["codex", "resume", "--all"].as_ref());
+        assert!(interactive.resume_picker);
+        assert!(interactive.resume_show_all);
     }
 
     #[test]
@@ -808,7 +873,7 @@ mod tests {
                 "--ask-for-approval",
                 "on-request",
                 "-m",
-                "gpt-5-test",
+                "gpt-5.1-test",
                 "-p",
                 "my-profile",
                 "-C",
@@ -819,7 +884,7 @@ 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/cli/src/mcp_cmd.rs

@@ -14,7 +14,6 @@ use codex_core::config::find_codex_home;
 use codex_core::config::load_global_mcp_servers;
 use codex_core::config::types::McpServerConfig;
 use codex_core::config::types::McpServerTransportConfig;
-use codex_core::features::Feature;
 use codex_core::mcp::auth::compute_auth_statuses;
 use codex_core::protocol::McpAuthStatus;
 use codex_rmcp_client::delete_oauth_tokens;
@@ -53,11 +52,9 @@ pub enum McpSubcommand {
     Remove(RemoveArgs),
 
     /// [experimental] Authenticate with a configured MCP server via OAuth.
-    /// Requires experimental_use_rmcp_client = true in config.toml.
     Login(LoginArgs),
 
     /// [experimental] Remove stored OAuth credentials for a server.
-    /// Requires experimental_use_rmcp_client = true in config.toml.
     Logout(LogoutArgs),
 }
 
@@ -79,6 +76,7 @@ pub struct GetArgs {
 }
 
 #[derive(Debug, clap::Parser)]
+#[command(override_usage = "codex mcp add [OPTIONS] <NAME> (--url <URL> | -- <COMMAND>...)")]
 pub struct AddArgs {
     /// Name for the MCP server configuration.
     pub name: String,
@@ -282,24 +280,17 @@ async fn run_add(config_overrides: &CliConfigOverrides, add_args: AddArgs) -> Re
     {
         match supports_oauth_login(&url).await {
             Ok(true) => {
-                if !config.features.enabled(Feature::RmcpClient) {
-                    println!(
-                        "MCP server supports login. Add `experimental_use_rmcp_client = true` \
-                         to your config.toml and run `codex mcp login {name}` to login."
-                    );
-                } else {
-                    println!("Detected OAuth support. Starting OAuth flow…");
-                    perform_oauth_login(
-                        &name,
-                        &url,
-                        config.mcp_oauth_credentials_store_mode,
-                        http_headers.clone(),
-                        env_http_headers.clone(),
-                        &Vec::new(),
-                    )
-                    .await?;
-                    println!("Successfully logged in.");
-                }
+                println!("Detected OAuth support. Starting OAuth flow…");
+                perform_oauth_login(
+                    &name,
+                    &url,
+                    config.mcp_oauth_credentials_store_mode,
+                    http_headers.clone(),
+                    env_http_headers.clone(),
+                    &Vec::new(),
+                )
+                .await?;
+                println!("Successfully logged in.");
             }
             Ok(false) => {}
             Err(_) => println!(
@@ -352,12 +343,6 @@ async fn run_login(config_overrides: &CliConfigOverrides, login_args: LoginArgs)
         .await
         .context("failed to load configuration")?;
 
-    if !config.features.enabled(Feature::RmcpClient) {
-        bail!(
-            "OAuth login is only supported when [features].rmcp_client is true in config.toml. See https://github.com/openai/codex/blob/main/docs/config.md#feature-flags for details."
-        );
-    }
-
     let LoginArgs { name, scopes } = login_args;
 
     let Some(server) = config.mcp_servers.get(&name) else {

codex-rs/cli/tests/execpolicy.rs

@@ -0,0 +1,58 @@
+use std::fs;
+
+use assert_cmd::Command;
+use pretty_assertions::assert_eq;
+use serde_json::json;
+use tempfile::TempDir;
+
+#[test]
+fn execpolicy_check_matches_expected_json() -> Result<(), Box<dyn std::error::Error>> {
+    let codex_home = TempDir::new()?;
+    let policy_path = codex_home.path().join("policy.codexpolicy");
+    fs::write(
+        &policy_path,
+        r#"
+prefix_rule(
+    pattern = ["git", "push"],
+    decision = "forbidden",
+)
+"#,
+    )?;
+
+    let output = Command::cargo_bin("codex")?
+        .env("CODEX_HOME", codex_home.path())
+        .args([
+            "execpolicy",
+            "check",
+            "--policy",
+            policy_path
+                .to_str()
+                .expect("policy path should be valid UTF-8"),
+            "git",
+            "push",
+            "origin",
+            "main",
+        ])
+        .output()?;
+
+    assert!(output.status.success());
+    let result: serde_json::Value = serde_json::from_slice(&output.stdout)?;
+    assert_eq!(
+        result,
+        json!({
+            "match": {
+                "decision": "forbidden",
+                "matchedRules": [
+                    {
+                        "prefixRuleMatch": {
+                            "matchedPrefix": ["git", "push"],
+                            "decision": "forbidden"
+                        }
+                    }
+                ]
+            }
+        })
+    );
+
+    Ok(())
+}

codex-rs/cloud-tasks-client/Cargo.toml

@@ -1,7 +1,8 @@
 [package]
 name = "codex-cloud-tasks-client"
-version = { workspace = true }
-edition = "2024"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
 
 [lib]
 name = "codex_cloud_tasks_client"

codex-rs/cloud-tasks/Cargo.toml

@@ -1,7 +1,8 @@
 [package]
-edition = "2024"
 name = "codex-cloud-tasks"
-version = { workspace = true }
+version.workspace = true
+edition.workspace = true
+license.workspace = true
 
 [lib]
 name = "codex_cloud_tasks"

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/codex-api/Cargo.toml

@@ -0,0 +1,30 @@
+[package]
+name = "codex-api"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+async-trait = { workspace = true }
+bytes = { workspace = true }
+codex-client = { workspace = true }
+codex-protocol = { workspace = true }
+futures = { workspace = true }
+http = { workspace = true }
+serde = { workspace = true, features = ["derive"] }
+serde_json = { workspace = true }
+thiserror = { workspace = true }
+tokio = { workspace = true, features = ["macros", "rt", "sync", "time"] }
+tracing = { workspace = true }
+eventsource-stream = { workspace = true }
+regex-lite = { workspace = true }
+tokio-util = { workspace = true, features = ["codec"] }
+
+[dev-dependencies]
+anyhow = { workspace = true }
+assert_matches = { workspace = true }
+pretty_assertions = { workspace = true }
+tokio-test = { workspace = true }
+
+[lints]
+workspace = true

codex-rs/codex-api/README.md

@@ -0,0 +1,32 @@
+# codex-api
+
+Typed clients for Codex/OpenAI APIs built on top of the generic transport in `codex-client`.
+
+- Hosts the request/response models and prompt helpers for Responses, Chat Completions, and Compact APIs.
+- Owns provider configuration (base URLs, headers, query params), auth header injection, retry tuning, and stream idle settings.
+- Parses SSE streams into `ResponseEvent`/`ResponseStream`, including rate-limit snapshots and API-specific error mapping.
+- Serves as the wire-level layer consumed by `codex-core`; higher layers handle auth refresh and business logic.
+
+## Core interface
+
+The public interface of this crate is intentionally small and uniform:
+
+- **Prompted endpoints (Chat + Responses)**
+  - Input: a single `Prompt` plus endpoint-specific options.
+    - `Prompt` (re-exported as `codex_api::Prompt`) carries:
+      - `instructions: String` – the fully-resolved system prompt for this turn.
+      - `input: Vec<ResponseItem>` – conversation history and user/tool messages.
+      - `tools: Vec<serde_json::Value>` – JSON tools compatible with the target API.
+      - `parallel_tool_calls: bool`.
+      - `output_schema: Option<Value>` – used to build `text.format` when present.
+  - Output: a `ResponseStream` of `ResponseEvent` (both re-exported from `common`).
+
+- **Compaction endpoint**
+  - Input: `CompactionInput<'a>` (re-exported as `codex_api::CompactionInput`):
+    - `model: &str`.
+    - `input: &[ResponseItem]` – history to compact.
+    - `instructions: &str` – fully-resolved compaction instructions.
+  - Output: `Vec<ResponseItem>`.
+  - `CompactClient::compact_input(&CompactionInput, extra_headers)` wraps the JSON encoding and retry/telemetry wiring.
+
+All HTTP details (URLs, headers, retry/backoff policies, SSE framing) are encapsulated in `codex-api` and `codex-client`. Callers construct prompts/inputs using protocol types and work with typed streams of `ResponseEvent` or compacted `ResponseItem` values.

codex-rs/codex-api/src/auth.rs

@@ -0,0 +1,27 @@
+use codex_client::Request;
+
+/// Provides bearer and account identity information for API requests.
+///
+/// Implementations should be cheap and non-blocking; any asynchronous
+/// refresh or I/O should be handled by higher layers before requests
+/// reach this interface.
+pub trait AuthProvider: Send + Sync {
+    fn bearer_token(&self) -> Option<String>;
+    fn account_id(&self) -> Option<String> {
+        None
+    }
+}
+
+pub(crate) fn add_auth_headers<A: AuthProvider>(auth: &A, mut req: Request) -> Request {
+    if let Some(token) = auth.bearer_token()
+        && let Ok(header) = format!("Bearer {token}").parse()
+    {
+        let _ = req.headers.insert(http::header::AUTHORIZATION, header);
+    }
+    if let Some(account_id) = auth.account_id()
+        && let Ok(header) = account_id.parse()
+    {
+        let _ = req.headers.insert("ChatGPT-Account-ID", header);
+    }
+    req
+}

codex-rs/codex-api/src/common.rs

@@ -0,0 +1,167 @@
+use crate::error::ApiError;
+use codex_protocol::config_types::ReasoningSummary as ReasoningSummaryConfig;
+use codex_protocol::config_types::Verbosity as VerbosityConfig;
+use codex_protocol::models::ResponseItem;
+use codex_protocol::openai_models::ReasoningEffort as ReasoningEffortConfig;
+use codex_protocol::protocol::RateLimitSnapshot;
+use codex_protocol::protocol::TokenUsage;
+use futures::Stream;
+use serde::Serialize;
+use serde_json::Value;
+use std::pin::Pin;
+use std::task::Context;
+use std::task::Poll;
+use tokio::sync::mpsc;
+
+/// Canonical prompt input for Chat and Responses endpoints.
+#[derive(Debug, Clone)]
+pub struct Prompt {
+    /// Fully-resolved system instructions for this turn.
+    pub instructions: String,
+    /// Conversation history and user/tool messages.
+    pub input: Vec<ResponseItem>,
+    /// JSON-encoded tool definitions compatible with the target API.
+    // TODO(jif) have a proper type here
+    pub tools: Vec<Value>,
+    /// Whether parallel tool calls are permitted.
+    pub parallel_tool_calls: bool,
+    /// Optional output schema used to build the `text.format` controls.
+    pub output_schema: Option<Value>,
+}
+
+/// Canonical input payload for the compaction endpoint.
+#[derive(Debug, Clone, Serialize)]
+pub struct CompactionInput<'a> {
+    pub model: &'a str,
+    pub input: &'a [ResponseItem],
+    pub instructions: &'a str,
+}
+
+#[derive(Debug)]
+pub enum ResponseEvent {
+    Created,
+    OutputItemDone(ResponseItem),
+    OutputItemAdded(ResponseItem),
+    Completed {
+        response_id: String,
+        token_usage: Option<TokenUsage>,
+    },
+    OutputTextDelta(String),
+    ReasoningSummaryDelta {
+        delta: String,
+        summary_index: i64,
+    },
+    ReasoningContentDelta {
+        delta: String,
+        content_index: i64,
+    },
+    ReasoningSummaryPartAdded {
+        summary_index: i64,
+    },
+    RateLimits(RateLimitSnapshot),
+}
+
+#[derive(Debug, Serialize, Clone)]
+pub struct Reasoning {
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub effort: Option<ReasoningEffortConfig>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub summary: Option<ReasoningSummaryConfig>,
+}
+
+#[derive(Debug, Serialize, Default, Clone)]
+#[serde(rename_all = "snake_case")]
+pub enum TextFormatType {
+    #[default]
+    JsonSchema,
+}
+
+#[derive(Debug, Serialize, Default, Clone)]
+pub struct TextFormat {
+    /// Format type used by the OpenAI text controls.
+    pub r#type: TextFormatType,
+    /// When true, the server is expected to strictly validate responses.
+    pub strict: bool,
+    /// JSON schema for the desired output.
+    pub schema: Value,
+    /// Friendly name for the format, used in telemetry/debugging.
+    pub name: String,
+}
+
+/// Controls the `text` field for the Responses API, combining verbosity and
+/// optional JSON schema output formatting.
+#[derive(Debug, Serialize, Default, Clone)]
+pub struct TextControls {
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub verbosity: Option<OpenAiVerbosity>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub format: Option<TextFormat>,
+}
+
+#[derive(Debug, Serialize, Default, Clone)]
+#[serde(rename_all = "lowercase")]
+pub enum OpenAiVerbosity {
+    Low,
+    #[default]
+    Medium,
+    High,
+}
+
+impl From<VerbosityConfig> for OpenAiVerbosity {
+    fn from(v: VerbosityConfig) -> Self {
+        match v {
+            VerbosityConfig::Low => OpenAiVerbosity::Low,
+            VerbosityConfig::Medium => OpenAiVerbosity::Medium,
+            VerbosityConfig::High => OpenAiVerbosity::High,
+        }
+    }
+}
+
+#[derive(Debug, Serialize)]
+pub struct ResponsesApiRequest<'a> {
+    pub model: &'a str,
+    pub instructions: &'a str,
+    pub input: &'a [ResponseItem],
+    pub tools: &'a [serde_json::Value],
+    pub tool_choice: &'static str,
+    pub parallel_tool_calls: bool,
+    pub reasoning: Option<Reasoning>,
+    pub store: bool,
+    pub stream: bool,
+    pub include: Vec<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub prompt_cache_key: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub text: Option<TextControls>,
+}
+
+pub fn create_text_param_for_request(
+    verbosity: Option<VerbosityConfig>,
+    output_schema: &Option<Value>,
+) -> Option<TextControls> {
+    if verbosity.is_none() && output_schema.is_none() {
+        return None;
+    }
+
+    Some(TextControls {
+        verbosity: verbosity.map(std::convert::Into::into),
+        format: output_schema.as_ref().map(|schema| TextFormat {
+            r#type: TextFormatType::JsonSchema,
+            strict: true,
+            schema: schema.clone(),
+            name: "codex_output_schema".to_string(),
+        }),
+    })
+}
+
+pub struct ResponseStream {
+    pub rx_event: mpsc::Receiver<Result<ResponseEvent, ApiError>>,
+}
+
+impl Stream for ResponseStream {
+    type Item = Result<ResponseEvent, ApiError>;
+
+    fn poll_next(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Option<Self::Item>> {
+        self.rx_event.poll_recv(cx)
+    }
+}

codex-rs/codex-api/src/endpoint/chat.rs

@@ -0,0 +1,266 @@
+use crate::ChatRequest;
+use crate::auth::AuthProvider;
+use crate::common::Prompt as ApiPrompt;
+use crate::common::ResponseEvent;
+use crate::common::ResponseStream;
+use crate::endpoint::streaming::StreamingClient;
+use crate::error::ApiError;
+use crate::provider::Provider;
+use crate::provider::WireApi;
+use crate::sse::chat::spawn_chat_stream;
+use crate::telemetry::SseTelemetry;
+use codex_client::HttpTransport;
+use codex_client::RequestTelemetry;
+use codex_protocol::models::ContentItem;
+use codex_protocol::models::ReasoningItemContent;
+use codex_protocol::models::ResponseItem;
+use codex_protocol::protocol::SessionSource;
+use futures::Stream;
+use http::HeaderMap;
+use serde_json::Value;
+use std::collections::VecDeque;
+use std::pin::Pin;
+use std::sync::Arc;
+use std::task::Context;
+use std::task::Poll;
+
+pub struct ChatClient<T: HttpTransport, A: AuthProvider> {
+    streaming: StreamingClient<T, A>,
+}
+
+impl<T: HttpTransport, A: AuthProvider> ChatClient<T, A> {
+    pub fn new(transport: T, provider: Provider, auth: A) -> Self {
+        Self {
+            streaming: StreamingClient::new(transport, provider, auth),
+        }
+    }
+
+    pub fn with_telemetry(
+        self,
+        request: Option<Arc<dyn RequestTelemetry>>,
+        sse: Option<Arc<dyn SseTelemetry>>,
+    ) -> Self {
+        Self {
+            streaming: self.streaming.with_telemetry(request, sse),
+        }
+    }
+
+    pub async fn stream_request(&self, request: ChatRequest) -> Result<ResponseStream, ApiError> {
+        self.stream(request.body, request.headers).await
+    }
+
+    pub async fn stream_prompt(
+        &self,
+        model: &str,
+        prompt: &ApiPrompt,
+        conversation_id: Option<String>,
+        session_source: Option<SessionSource>,
+    ) -> Result<ResponseStream, ApiError> {
+        use crate::requests::ChatRequestBuilder;
+
+        let request =
+            ChatRequestBuilder::new(model, &prompt.instructions, &prompt.input, &prompt.tools)
+                .conversation_id(conversation_id)
+                .session_source(session_source)
+                .build(self.streaming.provider())?;
+
+        self.stream_request(request).await
+    }
+
+    fn path(&self) -> &'static str {
+        match self.streaming.provider().wire {
+            WireApi::Chat => "chat/completions",
+            _ => "responses",
+        }
+    }
+
+    pub async fn stream(
+        &self,
+        body: Value,
+        extra_headers: HeaderMap,
+    ) -> Result<ResponseStream, ApiError> {
+        self.streaming
+            .stream(self.path(), body, extra_headers, spawn_chat_stream)
+            .await
+    }
+}
+
+#[derive(Copy, Clone, Eq, PartialEq)]
+pub enum AggregateMode {
+    AggregatedOnly,
+    Streaming,
+}
+
+/// Stream adapter that merges token deltas into a single assistant message per turn.
+pub struct AggregatedStream {
+    inner: ResponseStream,
+    cumulative: String,
+    cumulative_reasoning: String,
+    pending: VecDeque<ResponseEvent>,
+    mode: AggregateMode,
+}
+
+impl Stream for AggregatedStream {
+    type Item = Result<ResponseEvent, ApiError>;
+
+    fn poll_next(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Option<Self::Item>> {
+        let this = self.get_mut();
+
+        if let Some(ev) = this.pending.pop_front() {
+            return Poll::Ready(Some(Ok(ev)));
+        }
+
+        loop {
+            match Pin::new(&mut this.inner).poll_next(cx) {
+                Poll::Pending => return Poll::Pending,
+                Poll::Ready(None) => return Poll::Ready(None),
+                Poll::Ready(Some(Err(e))) => return Poll::Ready(Some(Err(e))),
+                Poll::Ready(Some(Ok(ResponseEvent::OutputItemDone(item)))) => {
+                    let is_assistant_message = matches!(
+                        &item,
+                        ResponseItem::Message { role, .. } if role == "assistant"
+                    );
+
+                    if is_assistant_message {
+                        match this.mode {
+                            AggregateMode::AggregatedOnly => {
+                                if this.cumulative.is_empty()
+                                    && let ResponseItem::Message { content, .. } = &item
+                                    && let Some(text) = content.iter().find_map(|c| match c {
+                                        ContentItem::OutputText { text } => Some(text),
+                                        _ => None,
+                                    })
+                                {
+                                    this.cumulative.push_str(text);
+                                }
+                                continue;
+                            }
+                            AggregateMode::Streaming => {
+                                if this.cumulative.is_empty() {
+                                    return Poll::Ready(Some(Ok(ResponseEvent::OutputItemDone(
+                                        item,
+                                    ))));
+                                } else {
+                                    continue;
+                                }
+                            }
+                        }
+                    }
+
+                    return Poll::Ready(Some(Ok(ResponseEvent::OutputItemDone(item))));
+                }
+                Poll::Ready(Some(Ok(ResponseEvent::RateLimits(snapshot)))) => {
+                    return Poll::Ready(Some(Ok(ResponseEvent::RateLimits(snapshot))));
+                }
+                Poll::Ready(Some(Ok(ResponseEvent::Completed {
+                    response_id,
+                    token_usage,
+                }))) => {
+                    let mut emitted_any = false;
+
+                    if !this.cumulative_reasoning.is_empty() {
+                        let aggregated_reasoning = ResponseItem::Reasoning {
+                            id: String::new(),
+                            summary: Vec::new(),
+                            content: Some(vec![ReasoningItemContent::ReasoningText {
+                                text: std::mem::take(&mut this.cumulative_reasoning),
+                            }]),
+                            encrypted_content: None,
+                        };
+                        this.pending
+                            .push_back(ResponseEvent::OutputItemDone(aggregated_reasoning));
+                        emitted_any = true;
+                    }
+
+                    if !this.cumulative.is_empty() {
+                        let aggregated_message = ResponseItem::Message {
+                            id: None,
+                            role: "assistant".to_string(),
+                            content: vec![ContentItem::OutputText {
+                                text: std::mem::take(&mut this.cumulative),
+                            }],
+                        };
+                        this.pending
+                            .push_back(ResponseEvent::OutputItemDone(aggregated_message));
+                        emitted_any = true;
+                    }
+
+                    if emitted_any {
+                        this.pending.push_back(ResponseEvent::Completed {
+                            response_id: response_id.clone(),
+                            token_usage: token_usage.clone(),
+                        });
+                        if let Some(ev) = this.pending.pop_front() {
+                            return Poll::Ready(Some(Ok(ev)));
+                        }
+                    }
+
+                    return Poll::Ready(Some(Ok(ResponseEvent::Completed {
+                        response_id,
+                        token_usage,
+                    })));
+                }
+                Poll::Ready(Some(Ok(ResponseEvent::Created))) => {
+                    continue;
+                }
+                Poll::Ready(Some(Ok(ResponseEvent::OutputTextDelta(delta)))) => {
+                    this.cumulative.push_str(&delta);
+                    if matches!(this.mode, AggregateMode::Streaming) {
+                        return Poll::Ready(Some(Ok(ResponseEvent::OutputTextDelta(delta))));
+                    } else {
+                        continue;
+                    }
+                }
+                Poll::Ready(Some(Ok(ResponseEvent::ReasoningContentDelta {
+                    delta,
+                    content_index,
+                }))) => {
+                    this.cumulative_reasoning.push_str(&delta);
+                    if matches!(this.mode, AggregateMode::Streaming) {
+                        return Poll::Ready(Some(Ok(ResponseEvent::ReasoningContentDelta {
+                            delta,
+                            content_index,
+                        })));
+                    } else {
+                        continue;
+                    }
+                }
+                Poll::Ready(Some(Ok(ResponseEvent::ReasoningSummaryDelta { .. }))) => continue,
+                Poll::Ready(Some(Ok(ResponseEvent::ReasoningSummaryPartAdded { .. }))) => {
+                    continue;
+                }
+                Poll::Ready(Some(Ok(ResponseEvent::OutputItemAdded(item)))) => {
+                    return Poll::Ready(Some(Ok(ResponseEvent::OutputItemAdded(item))));
+                }
+            }
+        }
+    }
+}
+
+pub trait AggregateStreamExt {
+    fn aggregate(self) -> AggregatedStream;
+
+    fn streaming_mode(self) -> ResponseStream;
+}
+
+impl AggregateStreamExt for ResponseStream {
+    fn aggregate(self) -> AggregatedStream {
+        AggregatedStream::new(self, AggregateMode::AggregatedOnly)
+    }
+
+    fn streaming_mode(self) -> ResponseStream {
+        self
+    }
+}
+
+impl AggregatedStream {
+    fn new(inner: ResponseStream, mode: AggregateMode) -> Self {
+        AggregatedStream {
+            inner,
+            cumulative: String::new(),
+            cumulative_reasoning: String::new(),
+            pending: VecDeque::new(),
+            mode,
+        }
+    }
+}

codex-rs/codex-api/src/endpoint/compact.rs

@@ -0,0 +1,162 @@
+use crate::auth::AuthProvider;
+use crate::auth::add_auth_headers;
+use crate::common::CompactionInput;
+use crate::error::ApiError;
+use crate::provider::Provider;
+use crate::provider::WireApi;
+use crate::telemetry::run_with_request_telemetry;
+use codex_client::HttpTransport;
+use codex_client::RequestTelemetry;
+use codex_protocol::models::ResponseItem;
+use http::HeaderMap;
+use http::Method;
+use serde::Deserialize;
+use serde_json::to_value;
+use std::sync::Arc;
+
+pub struct CompactClient<T: HttpTransport, A: AuthProvider> {
+    transport: T,
+    provider: Provider,
+    auth: A,
+    request_telemetry: Option<Arc<dyn RequestTelemetry>>,
+}
+
+impl<T: HttpTransport, A: AuthProvider> CompactClient<T, A> {
+    pub fn new(transport: T, provider: Provider, auth: A) -> Self {
+        Self {
+            transport,
+            provider,
+            auth,
+            request_telemetry: None,
+        }
+    }
+
+    pub fn with_telemetry(mut self, request: Option<Arc<dyn RequestTelemetry>>) -> Self {
+        self.request_telemetry = request;
+        self
+    }
+
+    fn path(&self) -> Result<&'static str, ApiError> {
+        match self.provider.wire {
+            WireApi::Compact | WireApi::Responses => Ok("responses/compact"),
+            WireApi::Chat => Err(ApiError::Stream(
+                "compact endpoint requires responses wire api".to_string(),
+            )),
+        }
+    }
+
+    pub async fn compact(
+        &self,
+        body: serde_json::Value,
+        extra_headers: HeaderMap,
+    ) -> Result<Vec<ResponseItem>, ApiError> {
+        let path = self.path()?;
+        let builder = || {
+            let mut req = self.provider.build_request(Method::POST, path);
+            req.headers.extend(extra_headers.clone());
+            req.body = Some(body.clone());
+            add_auth_headers(&self.auth, req)
+        };
+
+        let resp = run_with_request_telemetry(
+            self.provider.retry.to_policy(),
+            self.request_telemetry.clone(),
+            builder,
+            |req| self.transport.execute(req),
+        )
+        .await?;
+        let parsed: CompactHistoryResponse =
+            serde_json::from_slice(&resp.body).map_err(|e| ApiError::Stream(e.to_string()))?;
+        Ok(parsed.output)
+    }
+
+    pub async fn compact_input(
+        &self,
+        input: &CompactionInput<'_>,
+        extra_headers: HeaderMap,
+    ) -> Result<Vec<ResponseItem>, ApiError> {
+        let body = to_value(input)
+            .map_err(|e| ApiError::Stream(format!("failed to encode compaction input: {e}")))?;
+        self.compact(body, extra_headers).await
+    }
+}
+
+#[derive(Debug, Deserialize)]
+struct CompactHistoryResponse {
+    output: Vec<ResponseItem>,
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::provider::RetryConfig;
+    use async_trait::async_trait;
+    use codex_client::Request;
+    use codex_client::Response;
+    use codex_client::StreamResponse;
+    use codex_client::TransportError;
+    use http::HeaderMap;
+    use std::time::Duration;
+
+    #[derive(Clone, Default)]
+    struct DummyTransport;
+
+    #[async_trait]
+    impl HttpTransport for DummyTransport {
+        async fn execute(&self, _req: Request) -> Result<Response, TransportError> {
+            Err(TransportError::Build("execute should not run".to_string()))
+        }
+
+        async fn stream(&self, _req: Request) -> Result<StreamResponse, TransportError> {
+            Err(TransportError::Build("stream should not run".to_string()))
+        }
+    }
+
+    #[derive(Clone, Default)]
+    struct DummyAuth;
+
+    impl AuthProvider for DummyAuth {
+        fn bearer_token(&self) -> Option<String> {
+            None
+        }
+    }
+
+    fn provider(wire: WireApi) -> Provider {
+        Provider {
+            name: "test".to_string(),
+            base_url: "https://example.com/v1".to_string(),
+            query_params: None,
+            wire,
+            headers: HeaderMap::new(),
+            retry: RetryConfig {
+                max_attempts: 1,
+                base_delay: Duration::from_millis(1),
+                retry_429: false,
+                retry_5xx: true,
+                retry_transport: true,
+            },
+            stream_idle_timeout: Duration::from_secs(1),
+        }
+    }
+
+    #[tokio::test]
+    async fn errors_when_wire_is_chat() {
+        let client = CompactClient::new(DummyTransport, provider(WireApi::Chat), DummyAuth);
+        let input = CompactionInput {
+            model: "gpt-test",
+            input: &[],
+            instructions: "inst",
+        };
+        let err = client
+            .compact_input(&input, HeaderMap::new())
+            .await
+            .expect_err("expected wire mismatch to fail");
+
+        match err {
+            ApiError::Stream(msg) => {
+                assert_eq!(msg, "compact endpoint requires responses wire api");
+            }
+            other => panic!("unexpected error: {other:?}"),
+        }
+    }
+}

codex-rs/codex-api/src/endpoint/mod.rs

@@ -0,0 +1,4 @@
+pub mod chat;
+pub mod compact;
+pub mod responses;
+mod streaming;

codex-rs/codex-api/src/endpoint/responses.rs

@@ -0,0 +1,107 @@
+use crate::auth::AuthProvider;
+use crate::common::Prompt as ApiPrompt;
+use crate::common::Reasoning;
+use crate::common::ResponseStream;
+use crate::common::TextControls;
+use crate::endpoint::streaming::StreamingClient;
+use crate::error::ApiError;
+use crate::provider::Provider;
+use crate::provider::WireApi;
+use crate::requests::ResponsesRequest;
+use crate::requests::ResponsesRequestBuilder;
+use crate::sse::spawn_response_stream;
+use crate::telemetry::SseTelemetry;
+use codex_client::HttpTransport;
+use codex_client::RequestTelemetry;
+use codex_protocol::protocol::SessionSource;
+use http::HeaderMap;
+use serde_json::Value;
+use std::sync::Arc;
+
+pub struct ResponsesClient<T: HttpTransport, A: AuthProvider> {
+    streaming: StreamingClient<T, A>,
+}
+
+#[derive(Default)]
+pub struct ResponsesOptions {
+    pub reasoning: Option<Reasoning>,
+    pub include: Vec<String>,
+    pub prompt_cache_key: Option<String>,
+    pub text: Option<TextControls>,
+    pub store_override: Option<bool>,
+    pub conversation_id: Option<String>,
+    pub session_source: Option<SessionSource>,
+}
+
+impl<T: HttpTransport, A: AuthProvider> ResponsesClient<T, A> {
+    pub fn new(transport: T, provider: Provider, auth: A) -> Self {
+        Self {
+            streaming: StreamingClient::new(transport, provider, auth),
+        }
+    }
+
+    pub fn with_telemetry(
+        self,
+        request: Option<Arc<dyn RequestTelemetry>>,
+        sse: Option<Arc<dyn SseTelemetry>>,
+    ) -> Self {
+        Self {
+            streaming: self.streaming.with_telemetry(request, sse),
+        }
+    }
+
+    pub async fn stream_request(
+        &self,
+        request: ResponsesRequest,
+    ) -> Result<ResponseStream, ApiError> {
+        self.stream(request.body, request.headers).await
+    }
+
+    pub async fn stream_prompt(
+        &self,
+        model: &str,
+        prompt: &ApiPrompt,
+        options: ResponsesOptions,
+    ) -> Result<ResponseStream, ApiError> {
+        let ResponsesOptions {
+            reasoning,
+            include,
+            prompt_cache_key,
+            text,
+            store_override,
+            conversation_id,
+            session_source,
+        } = options;
+
+        let request = ResponsesRequestBuilder::new(model, &prompt.instructions, &prompt.input)
+            .tools(&prompt.tools)
+            .parallel_tool_calls(prompt.parallel_tool_calls)
+            .reasoning(reasoning)
+            .include(include)
+            .prompt_cache_key(prompt_cache_key)
+            .text(text)
+            .conversation(conversation_id)
+            .session_source(session_source)
+            .store_override(store_override)
+            .build(self.streaming.provider())?;
+
+        self.stream_request(request).await
+    }
+
+    fn path(&self) -> &'static str {
+        match self.streaming.provider().wire {
+            WireApi::Responses | WireApi::Compact => "responses",
+            WireApi::Chat => "chat/completions",
+        }
+    }
+
+    pub async fn stream(
+        &self,
+        body: Value,
+        extra_headers: HeaderMap,
+    ) -> Result<ResponseStream, ApiError> {
+        self.streaming
+            .stream(self.path(), body, extra_headers, spawn_response_stream)
+            .await
+    }
+}

codex-rs/codex-api/src/endpoint/streaming.rs

@@ -0,0 +1,82 @@
+use crate::auth::AuthProvider;
+use crate::auth::add_auth_headers;
+use crate::common::ResponseStream;
+use crate::error::ApiError;
+use crate::provider::Provider;
+use crate::telemetry::SseTelemetry;
+use crate::telemetry::run_with_request_telemetry;
+use codex_client::HttpTransport;
+use codex_client::RequestTelemetry;
+use codex_client::StreamResponse;
+use http::HeaderMap;
+use http::Method;
+use serde_json::Value;
+use std::sync::Arc;
+use std::time::Duration;
+
+pub(crate) struct StreamingClient<T: HttpTransport, A: AuthProvider> {
+    transport: T,
+    provider: Provider,
+    auth: A,
+    request_telemetry: Option<Arc<dyn RequestTelemetry>>,
+    sse_telemetry: Option<Arc<dyn SseTelemetry>>,
+}
+
+impl<T: HttpTransport, A: AuthProvider> StreamingClient<T, A> {
+    pub(crate) fn new(transport: T, provider: Provider, auth: A) -> Self {
+        Self {
+            transport,
+            provider,
+            auth,
+            request_telemetry: None,
+            sse_telemetry: None,
+        }
+    }
+
+    pub(crate) fn with_telemetry(
+        mut self,
+        request: Option<Arc<dyn RequestTelemetry>>,
+        sse: Option<Arc<dyn SseTelemetry>>,
+    ) -> Self {
+        self.request_telemetry = request;
+        self.sse_telemetry = sse;
+        self
+    }
+
+    pub(crate) fn provider(&self) -> &Provider {
+        &self.provider
+    }
+
+    pub(crate) async fn stream(
+        &self,
+        path: &str,
+        body: Value,
+        extra_headers: HeaderMap,
+        spawner: fn(StreamResponse, Duration, Option<Arc<dyn SseTelemetry>>) -> ResponseStream,
+    ) -> Result<ResponseStream, ApiError> {
+        let builder = || {
+            let mut req = self.provider.build_request(Method::POST, path);
+            req.headers.extend(extra_headers.clone());
+            req.headers.insert(
+                http::header::ACCEPT,
+                http::HeaderValue::from_static("text/event-stream"),
+            );
+            req.body = Some(body.clone());
+            add_auth_headers(&self.auth, req)
+        };
+
+        let stream_response = run_with_request_telemetry(
+            self.provider.retry.to_policy(),
+            self.request_telemetry.clone(),
+            builder,
+            |req| self.transport.stream(req),
+        )
+        .await?;
+
+        Ok(spawner(
+            stream_response,
+            self.provider.stream_idle_timeout,
+            self.sse_telemetry.clone(),
+        ))
+    }
+}

codex-rs/codex-api/src/error.rs

@@ -0,0 +1,34 @@
+use crate::rate_limits::RateLimitError;
+use codex_client::TransportError;
+use http::StatusCode;
+use std::time::Duration;
+use thiserror::Error;
+
+#[derive(Debug, Error)]
+pub enum ApiError {
+    #[error(transparent)]
+    Transport(#[from] TransportError),
+    #[error("api error {status}: {message}")]
+    Api { status: StatusCode, message: String },
+    #[error("stream error: {0}")]
+    Stream(String),
+    #[error("context window exceeded")]
+    ContextWindowExceeded,
+    #[error("quota exceeded")]
+    QuotaExceeded,
+    #[error("usage not included")]
+    UsageNotIncluded,
+    #[error("retryable error: {message}")]
+    Retryable {
+        message: String,
+        delay: Option<Duration>,
+    },
+    #[error("rate limit: {0}")]
+    RateLimit(String),
+}
+
+impl From<RateLimitError> for ApiError {
+    fn from(err: RateLimitError) -> Self {
+        Self::RateLimit(err.to_string())
+    }
+}

codex-rs/codex-api/src/lib.rs

@@ -0,0 +1,35 @@
+pub mod auth;
+pub mod common;
+pub mod endpoint;
+pub mod error;
+pub mod provider;
+pub mod rate_limits;
+pub mod requests;
+pub mod sse;
+pub mod telemetry;
+
+pub use codex_client::RequestTelemetry;
+pub use codex_client::ReqwestTransport;
+pub use codex_client::TransportError;
+
+pub use crate::auth::AuthProvider;
+pub use crate::common::CompactionInput;
+pub use crate::common::Prompt;
+pub use crate::common::ResponseEvent;
+pub use crate::common::ResponseStream;
+pub use crate::common::ResponsesApiRequest;
+pub use crate::common::create_text_param_for_request;
+pub use crate::endpoint::chat::AggregateStreamExt;
+pub use crate::endpoint::chat::ChatClient;
+pub use crate::endpoint::compact::CompactClient;
+pub use crate::endpoint::responses::ResponsesClient;
+pub use crate::endpoint::responses::ResponsesOptions;
+pub use crate::error::ApiError;
+pub use crate::provider::Provider;
+pub use crate::provider::WireApi;
+pub use crate::requests::ChatRequest;
+pub use crate::requests::ChatRequestBuilder;
+pub use crate::requests::ResponsesRequest;
+pub use crate::requests::ResponsesRequestBuilder;
+pub use crate::sse::stream_from_fixture;
+pub use crate::telemetry::SseTelemetry;