Clarify subagent flag docs

Prevent certain Powershell/cmd invocations from reaching the sandbox
when they are trying to launch a browser, or run a command with a URL,
etc.

.github/codex/home/config.toml

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

.github/workflows/cla.yml

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

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

@@ -0,0 +1,105 @@
+name: Close stale contributor PRs
+
+on:
+  workflow_dispatch:
+  schedule:
+    - cron: "0 6 * * *"
+
+permissions:
+  contents: read
+  issues: write
+  pull-requests: write
+
+jobs:
+  close-stale-contributor-prs:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Close inactive PRs from contributors
+        uses: actions/github-script@v8
+        with:
+          github-token: ${{ secrets.GITHUB_TOKEN }}
+          script: |
+            const DAYS_INACTIVE = 14;
+            const cutoff = new Date(Date.now() - DAYS_INACTIVE * 24 * 60 * 60 * 1000);
+            const { owner, repo } = context.repo;
+            const dryRun = false;
+            const stalePrs = [];
+
+            core.info(`Dry run mode: ${dryRun}`);
+
+            const prs = await github.paginate(github.rest.pulls.list, {
+              owner,
+              repo,
+              state: "open",
+              per_page: 100,
+              sort: "updated",
+              direction: "asc",
+            });
+
+            for (const pr of prs) {
+              const lastUpdated = new Date(pr.updated_at);
+              if (lastUpdated > cutoff) {
+                core.info(`PR ${pr.number} is fresh`);
+                continue;
+              }
+
+              if (!pr.user || pr.user.type !== "User") {
+                core.info(`PR ${pr.number} wasn't created by a user`);
+                continue;
+              }
+
+              let permission;
+              try {
+                const permissionResponse = await github.rest.repos.getCollaboratorPermissionLevel({
+                  owner,
+                  repo,
+                  username: pr.user.login,
+                });
+                permission = permissionResponse.data.permission;
+              } catch (error) {
+                if (error.status === 404) {
+                  core.info(`Author ${pr.user.login} is not a collaborator; skipping #${pr.number}`);
+                  continue;
+                }
+                throw error;
+              }
+
+              const hasContributorAccess = ["admin", "maintain", "write"].includes(permission);
+              if (!hasContributorAccess) {
+                core.info(`Author ${pr.user.login} has ${permission} access; skipping #${pr.number}`);
+                continue;
+              }
+
+              stalePrs.push(pr);
+            }
+
+            if (!stalePrs.length) {
+              core.info("No stale contributor pull requests found.");
+              return;
+            }
+
+            for (const pr of stalePrs) {
+              const issue_number = pr.number;
+              const closeComment = `Closing this pull request because it has had no updates for more than ${DAYS_INACTIVE} days. If you plan to continue working on it, feel free to reopen or open a new PR.`;
+
+              if (dryRun) {
+                core.info(`[dry-run] Would close contributor PR #${issue_number} from ${pr.user.login}`);
+                continue;
+              }
+
+              await github.rest.issues.createComment({
+                owner,
+                repo,
+                issue_number,
+                body: closeComment,
+              });
+
+              await github.rest.pulls.update({
+                owner,
+                repo,
+                pull_number: issue_number,
+                state: "closed",
+              });
+
+              core.info(`Closed contributor PR #${issue_number} from ${pr.user.login}`);
+            }

.github/workflows/issue-deduplicator.yml

@@ -46,7 +46,6 @@ 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/rust-ci.yml

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

.github/workflows/rust-release.yml

@@ -371,8 +371,20 @@ 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 }}
+      # We are not ready to publish yet.
+      publish: false
+    secrets: inherit
+
   release:
-    needs: build
+    needs:
+      - build
+      - shell-tool-mcp
     name: release
     runs-on: ubuntu-latest
     permissions:
@@ -395,6 +407,14 @@ 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/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@v5
+
+      - 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 +1,412 @@
+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@v5
+
+      - 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@v4
+        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: ubuntu-20.04
+            image: ubuntu:20.04
+          - 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@v5
+
+      - 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 "$dest/bash"
+
+      - uses: actions/upload-artifact@v4
+        with:
+          name: shell-tool-mcp-bash-${{ 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: ${{ matrix.runner }}
+    timeout-minutes: 30
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - runner: macos-15-xlarge
+            target: aarch64-apple-darwin
+            variant: macos-15
+          - runner: macos-14
+            target: aarch64-apple-darwin
+            variant: macos-14
+          - runner: macos-13
+            target: x86_64-apple-darwin
+            variant: macos-13
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v5
+
+      - name: Build patched Bash
+        shell: bash
+        run: |
+          set -euo pipefail
+          git clone --depth 1 https://github.com/bminor/bash /tmp/bash
+          cd /tmp/bash
+          git fetch --depth 1 origin a8a1c2fac029404d3f42cd39f5a20f24b6e4fe4b
+          git checkout a8a1c2fac029404d3f42cd39f5a20f24b6e4fe4b
+          git apply "${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 "$dest/bash"
+
+      - uses: actions/upload-artifact@v4
+        with:
+          name: shell-tool-mcp-bash-${{ 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@v5
+
+      - 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-${PACKAGE_VERSION}.tgz"
+
+      - uses: actions/upload-artifact@v4
+        with:
+          name: codex-shell-tool-mcp-npm
+          path: dist/npm/codex-shell-tool-mcp-npm-${{ 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
@@ -82,3 +85,5 @@ CHANGELOG.ignore.md
 # nix related
 .direnv
 .envrc
+
+plans/

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 +69,38 @@ Codex can access MCP servers. To configure them, refer to the [config docs](./do
 
 Codex CLI supports a rich set of configuration options, with preferences stored in `~/.codex/config.toml`. For full configuration options, see [Configuration](./docs/config.md).
 
----
+### Execpolicy Quickstart
+
+Codex can enforce your own rules-based execution policy before it runs shell commands.
+
+1. Create a policy directory: `mkdir -p ~/.codex/policy`.
+2. Create one or more `.codexpolicy` files in that folder. Codex automatically loads every `.codexpolicy` file in there on startup.
+3. Write `prefix_rule` entries to describe the commands you want to allow, prompt, or block:
+
+```starlark
+prefix_rule(
+    pattern = ["git", ["push", "fetch"]],
+    decision = "prompt",  # allow | prompt | forbidden
+    match = [["git", "push", "origin", "main"]],  # examples that must match
+    not_match = [["git", "status"]],              # examples that must not match
+)
+```
+
+- `pattern` is a list of shell tokens, evaluated from left to right; wrap tokens in a nested list to express alternatives (e.g., match both `push` and `fetch`).
+- `decision` sets the severity; Codex picks the strictest decision when multiple rules match (forbidden > prompt > allow).
+- `match` and `not_match` act as (optional) unit tests. Codex validates them when it loads your policy, so you get feedback if an example has unexpected behavior.
+
+In this example rule, if Codex wants to run commands with the prefix `git push` or `git fetch`, it will first ask for user approval.
+
+Use the `codex execpolicy check` subcommand to preview decisions before you save a rule (see the [`codex-execpolicy` README](./codex-rs/execpolicy/README.md) for syntax details):
+
+```shell
+codex execpolicy check --policy ~/.codex/policy/default.codexpolicy git push origin main
+```
+
+Pass multiple `--policy` flags to test how several files combine, and use `--pretty` for formatted JSON output. See the [`codex-rs/execpolicy` README](./codex-rs/execpolicy/README.md) for a more detailed walkthrough of the available syntax.
+
+## Note: `execpolicy` commands are still in preview. The API may have breaking changes in the future.
 
 ### Docs & FAQ
 

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/Cargo.lock

@@ -187,8 +187,10 @@ dependencies = [
  "codex-app-server-protocol",
  "codex-core",
  "codex-protocol",
+ "core_test_support",
  "serde",
  "serde_json",
+ "shlex",
  "tokio",
  "uuid",
  "wiremock",
@@ -260,7 +262,7 @@ dependencies = [
  "memchr",
  "proc-macro2",
  "quote",
- "rustc-hash 2.1.1",
+ "rustc-hash",
  "serde",
  "serde_derive",
  "syn 2.0.104",
@@ -726,6 +728,17 @@ version = "0.2.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "613afe47fcd5fac7ccf1db93babcb082c5994d996f20b8b159f2ad1658eb5724"
 
+[[package]]
+name = "chardetng"
+version = "0.1.17"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "14b8f0b65b7b08ae3c8187e8d77174de20cb6777864c6b832d8ad365999cf1ea"
+dependencies = [
+ "cfg-if",
+ "encoding_rs",
+ "memchr",
+]
+
 [[package]]
 name = "chrono"
 version = "0.4.42"
@@ -850,12 +863,14 @@ dependencies = [
  "codex-protocol",
  "codex-utils-json-to-toml",
  "core_test_support",
+ "mcp-types",
  "opentelemetry-appender-tracing",
  "os_info",
  "pretty_assertions",
  "serde",
  "serde_json",
  "serial_test",
+ "shlex",
  "tempfile",
  "tokio",
  "toml",
@@ -873,16 +888,29 @@ dependencies = [
  "clap",
  "codex-protocol",
  "mcp-types",
- "paste",
  "pretty_assertions",
  "schemars 0.8.22",
  "serde",
  "serde_json",
  "strum_macros 0.27.2",
+ "thiserror 2.0.17",
  "ts-rs",
  "uuid",
 ]
 
+[[package]]
+name = "codex-app-server-test-client"
+version = "0.0.0"
+dependencies = [
+ "anyhow",
+ "clap",
+ "codex-app-server-protocol",
+ "codex-protocol",
+ "serde",
+ "serde_json",
+ "uuid",
+]
+
 [[package]]
 name = "codex-apply-patch"
 version = "0.0.0"
@@ -976,6 +1004,7 @@ dependencies = [
  "codex-common",
  "codex-core",
  "codex-exec",
+ "codex-execpolicy",
  "codex-login",
  "codex-mcp-server",
  "codex-process-hardening",
@@ -1047,7 +1076,10 @@ dependencies = [
  "clap",
  "codex-app-server-protocol",
  "codex-core",
+ "codex-lmstudio",
+ "codex-ollama",
  "codex-protocol",
+ "once_cell",
  "serde",
  "toml",
 ]
@@ -1064,11 +1096,13 @@ dependencies = [
  "async-trait",
  "base64",
  "bytes",
+ "chardetng",
  "chrono",
  "codex-app-server-protocol",
  "codex-apply-patch",
  "codex-arg0",
  "codex-async-utils",
+ "codex-execpolicy",
  "codex-file-search",
  "codex-git",
  "codex-keyring-store",
@@ -1078,13 +1112,13 @@ dependencies = [
  "codex-utils-pty",
  "codex-utils-readiness",
  "codex-utils-string",
- "codex-utils-tokenizer",
  "codex-windows-sandbox",
  "core-foundation 0.9.4",
  "core_test_support",
  "ctor 0.5.0",
  "dirs",
  "dunce",
+ "encoding_rs",
  "env-flags",
  "escargot",
  "eventsource-stream",
@@ -1092,16 +1126,19 @@ dependencies = [
  "http",
  "image",
  "indexmap 2.12.0",
+ "insta",
  "keyring",
  "landlock",
  "libc",
  "maplit",
  "mcp-types",
+ "once_cell",
  "openssl-sys",
  "os_info",
  "predicates",
  "pretty_assertions",
  "rand 0.9.2",
+ "regex",
  "regex-lite",
  "reqwest",
  "seccompiler",
@@ -1114,6 +1151,7 @@ dependencies = [
  "similar",
  "strum_macros 0.27.2",
  "tempfile",
+ "test-case",
  "test-log",
  "thiserror 2.0.17",
  "time",
@@ -1126,6 +1164,7 @@ dependencies = [
  "tracing-test",
  "tree-sitter",
  "tree-sitter-bash",
+ "url",
  "uuid",
  "walkdir",
  "which",
@@ -1143,7 +1182,6 @@ dependencies = [
  "codex-arg0",
  "codex-common",
  "codex-core",
- "codex-ollama",
  "codex-protocol",
  "core_test_support",
  "libc",
@@ -1166,9 +1204,47 @@ dependencies = [
  "wiremock",
 ]
 
+[[package]]
+name = "codex-exec-server"
+version = "0.0.0"
+dependencies = [
+ "anyhow",
+ "async-trait",
+ "clap",
+ "codex-core",
+ "libc",
+ "path-absolutize",
+ "pretty_assertions",
+ "rmcp",
+ "serde",
+ "serde_json",
+ "shlex",
+ "socket2 0.6.0",
+ "tempfile",
+ "tokio",
+ "tokio-util",
+ "tracing",
+ "tracing-subscriber",
+]
+
 [[package]]
 name = "codex-execpolicy"
 version = "0.0.0"
+dependencies = [
+ "anyhow",
+ "clap",
+ "multimap",
+ "pretty_assertions",
+ "serde",
+ "serde_json",
+ "shlex",
+ "starlark",
+ "thiserror 2.0.17",
+]
+
+[[package]]
+name = "codex-execpolicy-legacy"
+version = "0.0.0"
 dependencies = [
  "allocative",
  "anyhow",
@@ -1247,6 +1323,19 @@ dependencies = [
  "tokio",
 ]
 
+[[package]]
+name = "codex-lmstudio"
+version = "0.0.0"
+dependencies = [
+ "codex-core",
+ "reqwest",
+ "serde_json",
+ "tokio",
+ "tracing",
+ "which",
+ "wiremock",
+]
+
 [[package]]
 name = "codex-login"
 version = "0.0.0"
@@ -1322,6 +1411,7 @@ dependencies = [
  "codex-app-server-protocol",
  "codex-protocol",
  "eventsource-stream",
+ "http",
  "opentelemetry",
  "opentelemetry-otlp",
  "opentelemetry-semantic-conventions",
@@ -1355,6 +1445,7 @@ dependencies = [
  "icu_provider",
  "mcp-types",
  "mime_guess",
+ "pretty_assertions",
  "schemars 0.8.22",
  "serde",
  "serde_json",
@@ -1411,6 +1502,7 @@ dependencies = [
  "tracing",
  "urlencoding",
  "webbrowser",
+ "which",
 ]
 
 [[package]]
@@ -1438,12 +1530,12 @@ dependencies = [
  "codex-ansi-escape",
  "codex-app-server-protocol",
  "codex-arg0",
+ "codex-backend-client",
  "codex-common",
  "codex-core",
  "codex-feedback",
  "codex-file-search",
  "codex-login",
- "codex-ollama",
  "codex-protocol",
  "codex-windows-sandbox",
  "color-eyre",
@@ -1466,6 +1558,7 @@ dependencies = [
  "ratatui",
  "ratatui-macros",
  "regex-lite",
+ "reqwest",
  "serde",
  "serde_json",
  "serial_test",
@@ -1543,21 +1636,12 @@ dependencies = [
 name = "codex-utils-string"
 version = "0.0.0"
 
-[[package]]
-name = "codex-utils-tokenizer"
-version = "0.0.0"
-dependencies = [
- "anyhow",
- "pretty_assertions",
- "thiserror 2.0.17",
- "tiktoken-rs",
-]
-
 [[package]]
 name = "codex-windows-sandbox"
 version = "0.1.0"
 dependencies = [
  "anyhow",
+ "codex-protocol",
  "dirs-next",
  "dunce",
  "rand 0.8.5",
@@ -1699,6 +1783,7 @@ dependencies = [
  "notify",
  "regex-lite",
  "serde_json",
+ "shlex",
  "tempfile",
  "tokio",
  "walkdir",
@@ -2373,17 +2458,6 @@ dependencies = [
  "once_cell",
 ]
 
-[[package]]
-name = "fancy-regex"
-version = "0.13.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "531e46835a22af56d1e3b66f04844bed63158bc094a628bec1d321d9b4c44bf2"
-dependencies = [
- "bit-set",
- "regex-automata",
- "regex-syntax 0.8.5",
-]
-
 [[package]]
 name = "fastrand"
 version = "2.3.0"
@@ -3262,6 +3336,7 @@ checksum = "46fdb647ebde000f43b5b53f773c30cf9b0cb4300453208713fa38b2c70935a0"
 dependencies = [
  "console",
  "once_cell",
+ "serde",
  "similar",
 ]
 
@@ -3671,11 +3746,13 @@ dependencies = [
  "assert_cmd",
  "codex-core",
  "codex-mcp-server",
+ "core_test_support",
  "mcp-types",
  "os_info",
  "pretty_assertions",
  "serde",
  "serde_json",
+ "shlex",
  "tokio",
  "wiremock",
 ]
@@ -4708,7 +4785,7 @@ dependencies = [
  "pin-project-lite",
  "quinn-proto",
  "quinn-udp",
- "rustc-hash 2.1.1",
+ "rustc-hash",
  "rustls",
  "socket2 0.6.0",
  "thiserror 2.0.17",
@@ -4728,7 +4805,7 @@ dependencies = [
  "lru-slab",
  "rand 0.9.2",
  "ring",
- "rustc-hash 2.1.1",
+ "rustc-hash",
  "rustls",
  "rustls-pki-types",
  "slab",
@@ -5073,12 +5150,6 @@ version = "0.1.25"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "989e6739f80c4ad5b13e0fd7fe89531180375b18520cc8c82080e4dc4035b84f"
 
-[[package]]
-name = "rustc-hash"
-version = "1.1.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "08d43f7aa6b08d49f382cde6a7982047c3426db949b1424bc4b7ec9ae12c6ce2"
-
 [[package]]
 name = "rustc-hash"
 version = "2.1.1"
@@ -5126,6 +5197,7 @@ version = "0.23.29"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "2491382039b29b9b11ff08b76ff6c97cf287671dbb74f0be44bda389fffe9bd1"
 dependencies = [
+ "log",
  "once_cell",
  "ring",
  "rustls-pki-types",
@@ -6160,6 +6232,39 @@ version = "0.5.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "8f50febec83f5ee1df3015341d8bd429f2d1cc62bcba7ea2076759d315084683"
 
+[[package]]
+name = "test-case"
+version = "3.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "eb2550dd13afcd286853192af8601920d959b14c401fcece38071d53bf0768a8"
+dependencies = [
+ "test-case-macros",
+]
+
+[[package]]
+name = "test-case-core"
+version = "3.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "adcb7fd841cd518e279be3d5a3eb0636409487998a4aff22f3de87b81e88384f"
+dependencies = [
+ "cfg-if",
+ "proc-macro2",
+ "quote",
+ "syn 2.0.104",
+]
+
+[[package]]
+name = "test-case-macros"
+version = "3.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5c89e72a01ed4c579669add59014b9a524d609c0c88c6a585ce37485879f6ffb"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.104",
+ "test-case-core",
+]
+
 [[package]]
 name = "test-log"
 version = "0.2.18"
@@ -6265,21 +6370,6 @@ dependencies = [
  "zune-jpeg",
 ]
 
-[[package]]
-name = "tiktoken-rs"
-version = "0.7.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "25563eeba904d770acf527e8b370fe9a5547bacd20ff84a0b6c3bc41288e5625"
-dependencies = [
- "anyhow",
- "base64",
- "bstr",
- "fancy-regex",
- "lazy_static",
- "regex",
- "rustc-hash 1.1.0",
-]
-
 [[package]]
 name = "time"
 version = "0.3.44"
@@ -6520,8 +6610,10 @@ dependencies = [
  "percent-encoding",
  "pin-project",
  "prost",
+ "rustls-native-certs",
  "socket2 0.5.10",
  "tokio",
+ "tokio-rustls",
  "tokio-stream",
  "tower",
  "tower-layer",

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 +41,6 @@ members = [
     "utils/pty",
     "utils/readiness",
     "utils/string",
-    "utils/tokenizer",
 ]
 resolver = "2"
 
@@ -63,11 +66,13 @@ codex-chatgpt = { path = "chatgpt" }
 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 +89,6 @@ 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" }
@@ -105,6 +109,7 @@ axum = { version = "0.8", default-features = false }
 base64 = "0.22.1"
 bytes = "1.10.1"
 chrono = "0.4.42"
+chardetng = "0.1.17"
 clap = "4"
 clap_complete = "4"
 color-eyre = "0.6.3"
@@ -117,19 +122,20 @@ dotenvy = "0.15.7"
 dunce = "1.0.4"
 env-flags = "0.1.1"
 env_logger = "0.11.5"
+encoding_rs = "0.8.35"
 escargot = "0.5"
 eventsource-stream = "0.2.3"
 futures = { version = "0.3", default-features = false }
 http = "1.3.1"
 icu_decimal = "2.1"
-icu_provider = { version = "2.1", features = ["sync"] }
 icu_locale_core = "2.1"
+icu_provider = { version = "2.1", features = ["sync"] }
 ignore = "0.4.23"
 image = { version = "^0.25.8", default-features = false }
 indexmap = "2.12.0"
 insta = "1.43.2"
 itertools = "0.14.0"
-keyring = "3.6"
+keyring = { version = "3.6", default-features = false }
 landlock = "0.4.1"
 lazy_static = "1"
 libc = "0.2.175"
@@ -140,6 +146,7 @@ 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 +155,6 @@ 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"
@@ -159,11 +165,11 @@ rand = "0.9"
 ratatui = "0.29.0"
 ratatui-macros = "0.6.0"
 regex-lite = "0.1.7"
+regex = "1.11.1"
 reqwest = "0.12"
 rmcp = { version = "0.8.5", default-features = false }
 schemars = "0.8.22"
 seccompiler = "0.5.0"
-sentry = "0.34.0"
 serde = "1"
 serde_json = "1"
 serde_with = "3.14"
@@ -172,6 +178,7 @@ 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"
@@ -258,11 +265,11 @@ ignored = [
     "icu_provider",
     "openssl-sys",
     "codex-utils-readiness",
-    "codex-utils-tokenizer",
 ]
 
 [profile.release]
-lto = "fat"
+opt-level = "z"        # or "s" (z is smaller)
+lto = "thin"           # "fat" can be smaller sometimes; test both
 # Because we bundle some of these executables with the TypeScript CLI, we
 # remove everything to make the binary as small as possible.
 strip = "symbols"
@@ -278,8 +285,8 @@ opt-level = 0
 [patch.crates-io]
 # Uncomment to debug local changes.
 # ratatui = { path = "../../ratatui" }
-ratatui = { git = "https://github.com/nornagon/ratatui", branch = "nornagon-v0.29.0-patch" }
 crossterm = { git = "https://github.com/nornagon/crossterm", branch = "nornagon/color-query" }
+ratatui = { git = "https://github.com/nornagon/ratatui", branch = "nornagon-v0.29.0-patch" }
 
 # Uncomment to debug local changes.
 # rmcp = { path = "../../rust-sdk/crates/rmcp" }

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

@@ -15,11 +15,11 @@ 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)?;
@@ -77,17 +99,28 @@ 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)
@@ -120,10 +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();
@@ -152,13 +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();
@@ -171,6 +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);
         }
@@ -181,6 +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;
                 }
@@ -386,14 +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()
         {
@@ -734,7 +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;
@@ -137,6 +129,10 @@ client_request_definitions! {
         params: v2::TurnInterruptParams,
         response: v2::TurnInterruptResponse,
     },
+    ReviewStart => "review/start" {
+        params: v2::ReviewStartParams,
+        response: v2::TurnStartResponse,
+    },
 
     ModelList => "model/list" {
         params: v2::ModelListParams,
@@ -277,34 +273,36 @@ macro_rules! server_request_definitions {
     (
         $(
             $(#[$variant_meta:meta])*
-            $variant:ident
+            $variant:ident $(=> $wire:literal)? {
+                params: $params:ty,
+                response: $response:ty,
+            }
         ),* $(,)?
     ) => {
-        paste! {
-            /// Request initiated from the server and sent to the client.
-            #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
-            #[serde(tag = "method", rename_all = "camelCase")]
-            pub enum ServerRequest {
-                $(
-                    $(#[$variant_meta])*
-                    $variant {
-                        #[serde(rename = "id")]
-                        request_id: RequestId,
-                        params: [<$variant Params>],
-                    },
-                )*
-            }
+        /// Request initiated from the server and sent to the client.
+        #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+        #[serde(tag = "method", rename_all = "camelCase")]
+        pub enum ServerRequest {
+            $(
+                $(#[$variant_meta])*
+                $(#[serde(rename = $wire)] #[ts(rename = $wire)])?
+                $variant {
+                    #[serde(rename = "id")]
+                    request_id: RequestId,
+                    params: $params,
+                },
+            )*
+        }
 
-            #[derive(Debug, Clone, PartialEq, JsonSchema)]
-            pub enum ServerRequestPayload {
-                $( $variant([<$variant Params>]), )*
-            }
+        #[derive(Debug, Clone, PartialEq, JsonSchema)]
+        pub enum ServerRequestPayload {
+            $( $variant($params), )*
+        }
 
-            impl ServerRequestPayload {
-                pub fn request_with_id(self, request_id: RequestId) -> ServerRequest {
-                    match self {
-                        $(Self::$variant(params) => ServerRequest::$variant { request_id, params },)*
-                    }
+        impl ServerRequestPayload {
+            pub fn request_with_id(self, request_id: RequestId) -> ServerRequest {
+                match self {
+                    $(Self::$variant(params) => ServerRequest::$variant { request_id, params },)*
                 }
             }
         }
@@ -312,9 +310,9 @@ macro_rules! server_request_definitions {
         pub fn export_server_responses(
             out_dir: &::std::path::Path,
         ) -> ::std::result::Result<(), ::ts_rs::ExportError> {
-            paste! {
-                $(<[<$variant Response>] as ::ts_rs::TS>::export_all_to(out_dir)?;)*
-            }
+            $(
+                <$response as ::ts_rs::TS>::export_all_to(out_dir)?;
+            )*
             Ok(())
         }
 
@@ -323,9 +321,12 @@ macro_rules! server_request_definitions {
             out_dir: &Path,
         ) -> ::anyhow::Result<Vec<GeneratedSchema>> {
             let mut schemas = Vec::new();
-            paste! {
-                $(schemas.push(crate::export::write_json_schema::<[<$variant Response>]>(out_dir, stringify!([<$variant Response>]))?);)*
-            }
+            $(
+                schemas.push(crate::export::write_json_schema::<$response>(
+                    out_dir,
+                    concat!(stringify!($variant), "Response"),
+                )?);
+            )*
             Ok(schemas)
         }
 
@@ -334,9 +335,12 @@ 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)
         }
     };
@@ -374,7 +378,7 @@ 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)?)
             }
         }
@@ -426,49 +430,34 @@ 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)]
@@ -498,6 +487,7 @@ pub struct FuzzyFileSearchResponse {
 
 server_notification_definitions! {
     /// NEW NOTIFICATIONS
+    Error => "error" (v2::ErrorNotification),
     ThreadStarted => "thread/started" (v2::ThreadStartedNotification),
     TurnStarted => "turn/started" (v2::TurnStartedNotification),
     TurnCompleted => "turn/completed" (v2::TurnCompletedNotification),
@@ -508,6 +498,12 @@ server_notification_definitions! {
     McpToolCallProgress => "item/mcpToolCall/progress" (v2::McpToolCallProgressNotification),
     AccountUpdated => "account/updated" (v2::AccountUpdatedNotification),
     AccountRateLimitsUpdated => "account/rateLimits/updated" (v2::AccountRateLimitsUpdatedNotification),
+    ReasoningSummaryTextDelta => "item/reasoning/summaryTextDelta" (v2::ReasoningSummaryTextDeltaNotification),
+    ReasoningSummaryPartAdded => "item/reasoning/summaryPartAdded" (v2::ReasoningSummaryPartAddedNotification),
+    ReasoningTextDelta => "item/reasoning/textDelta" (v2::ReasoningTextDeltaNotification),
+
+    /// 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")]
@@ -530,17 +526,20 @@ 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,
@@ -558,7 +557,7 @@ mod tests {
                 "method": "newConversation",
                 "id": 42,
                 "params": {
-                    "model": "gpt-5-codex",
+                    "model": "gpt-5.1-codex-max",
                     "modelProvider": null,
                     "profile": null,
                     "cwd": null,
@@ -613,7 +612,7 @@ 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/mod.rs

@@ -2,5 +2,6 @@
 // Exposes protocol pieces used by `lib.rs` via `pub use protocol::common::*;`.
 
 pub mod common;
+pub mod thread_history;
 pub mod v1;
 pub mod v2;

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

@@ -0,0 +1,409 @@
+use crate::protocol::v2::ThreadItem;
+use crate::protocol::v2::Turn;
+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(),
+            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>,
+    status: TurnStatus,
+}
+
+impl From<PendingTurn> for Turn {
+    fn from(value: PendingTurn) -> Self {
+        Self {
+            id: value.id,
+            items: value.items,
+            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

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

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

@@ -4,22 +4,29 @@ use std::path::PathBuf;
 use crate::protocol::common::AuthMode;
 use codex_protocol::ConversationId;
 use codex_protocol::account::PlanType;
+use codex_protocol::approvals::SandboxCommandAssessment as CoreSandboxCommandAssessment;
 use codex_protocol::config_types::ReasoningEffort;
 use codex_protocol::config_types::ReasoningSummary;
 use codex_protocol::items::AgentMessageContent as CoreAgentMessageContent;
 use codex_protocol::items::TurnItem as CoreTurnItem;
+use codex_protocol::models::ResponseItem;
+use codex_protocol::parse_command::ParsedCommand as CoreParsedCommand;
+use codex_protocol::protocol::CodexErrorInfo as CoreCodexErrorInfo;
+use codex_protocol::protocol::CreditsSnapshot as CoreCreditsSnapshot;
 use codex_protocol::protocol::RateLimitSnapshot as CoreRateLimitSnapshot;
 use codex_protocol::protocol::RateLimitWindow as CoreRateLimitWindow;
+use codex_protocol::protocol::SessionSource as CoreSessionSource;
 use codex_protocol::user_input::UserInput as CoreUserInput;
 use mcp_types::ContentBlock as McpContentBlock;
 use schemars::JsonSchema;
 use serde::Deserialize;
 use serde::Serialize;
 use serde_json::Value as JsonValue;
+use thiserror::Error;
 use ts_rs::TS;
 
 // Macro to declare a camelCased API v2 enum mirroring a core enum which
-// tends to use kebab-case.
+// tends to use either snake_case or kebab-case.
 macro_rules! v2_enum_from_core {
     (
         pub enum $Name:ident from $Src:path { $( $Variant:ident ),+ $(,)? }
@@ -43,6 +50,72 @@ macro_rules! v2_enum_from_core {
     };
 }
 
+/// This translation layer make sure that we expose codex error code in camel case.
+///
+/// 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.
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub enum CodexErrorInfo {
+    ContextWindowExceeded,
+    UsageLimitExceeded,
+    HttpConnectionFailed {
+        #[serde(rename = "httpStatusCode")]
+        #[ts(rename = "httpStatusCode")]
+        http_status_code: Option<u16>,
+    },
+    /// Failed to connect to the response SSE stream.
+    ResponseStreamConnectionFailed {
+        #[serde(rename = "httpStatusCode")]
+        #[ts(rename = "httpStatusCode")]
+        http_status_code: Option<u16>,
+    },
+    InternalServerError,
+    Unauthorized,
+    BadRequest,
+    SandboxError,
+    /// The response SSE stream disconnected in the middle of a turn before completion.
+    ResponseStreamDisconnected {
+        #[serde(rename = "httpStatusCode")]
+        #[ts(rename = "httpStatusCode")]
+        http_status_code: Option<u16>,
+    },
+    /// Reached the retry limit for responses.
+    ResponseTooManyFailedAttempts {
+        #[serde(rename = "httpStatusCode")]
+        #[ts(rename = "httpStatusCode")]
+        http_status_code: Option<u16>,
+    },
+    Other,
+}
+
+impl From<CoreCodexErrorInfo> for CodexErrorInfo {
+    fn from(value: CoreCodexErrorInfo) -> Self {
+        match value {
+            CoreCodexErrorInfo::ContextWindowExceeded => CodexErrorInfo::ContextWindowExceeded,
+            CoreCodexErrorInfo::UsageLimitExceeded => CodexErrorInfo::UsageLimitExceeded,
+            CoreCodexErrorInfo::HttpConnectionFailed { http_status_code } => {
+                CodexErrorInfo::HttpConnectionFailed { http_status_code }
+            }
+            CoreCodexErrorInfo::ResponseStreamConnectionFailed { http_status_code } => {
+                CodexErrorInfo::ResponseStreamConnectionFailed { http_status_code }
+            }
+            CoreCodexErrorInfo::InternalServerError => CodexErrorInfo::InternalServerError,
+            CoreCodexErrorInfo::Unauthorized => CodexErrorInfo::Unauthorized,
+            CoreCodexErrorInfo::BadRequest => CodexErrorInfo::BadRequest,
+            CoreCodexErrorInfo::SandboxError => CodexErrorInfo::SandboxError,
+            CoreCodexErrorInfo::ResponseStreamDisconnected { http_status_code } => {
+                CodexErrorInfo::ResponseStreamDisconnected { http_status_code }
+            }
+            CoreCodexErrorInfo::ResponseTooManyFailedAttempts { http_status_code } => {
+                CodexErrorInfo::ResponseTooManyFailedAttempts { http_status_code }
+            }
+            CoreCodexErrorInfo::Other => CodexErrorInfo::Other,
+        }
+    }
+}
+
 v2_enum_from_core!(
     pub enum AskForApproval from codex_protocol::protocol::AskForApproval {
         UnlessTrusted, OnFailure, OnRequest, Never
@@ -55,13 +128,32 @@ v2_enum_from_core!(
     }
 );
 
+v2_enum_from_core!(
+    pub enum CommandRiskLevel from codex_protocol::approvals::SandboxRiskLevel {
+        Low,
+        Medium,
+        High
+    }
+);
+
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq, JsonSchema, TS)]
-#[serde(tag = "mode", rename_all = "camelCase")]
-#[ts(tag = "mode")]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub enum ApprovalDecision {
+    Accept,
+    Decline,
+    Cancel,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq, JsonSchema, TS)]
+#[serde(tag = "type", rename_all = "camelCase")]
+#[ts(tag = "type")]
 #[ts(export_to = "v2/")]
 pub enum SandboxPolicy {
     DangerFullAccess,
     ReadOnly,
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
     WorkspaceWrite {
         #[serde(default)]
         writable_roots: Vec<PathBuf>,
@@ -118,6 +210,148 @@ impl From<codex_protocol::protocol::SandboxPolicy> for SandboxPolicy {
     }
 }
 
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct SandboxCommandAssessment {
+    pub description: String,
+    pub risk_level: CommandRiskLevel,
+}
+
+impl SandboxCommandAssessment {
+    pub fn into_core(self) -> CoreSandboxCommandAssessment {
+        CoreSandboxCommandAssessment {
+            description: self.description,
+            risk_level: self.risk_level.to_core(),
+        }
+    }
+}
+
+impl From<CoreSandboxCommandAssessment> for SandboxCommandAssessment {
+    fn from(value: CoreSandboxCommandAssessment) -> Self {
+        Self {
+            description: value.description,
+            risk_level: CommandRiskLevel::from(value.risk_level),
+        }
+    }
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(tag = "type", rename_all = "camelCase")]
+#[ts(tag = "type")]
+#[ts(export_to = "v2/")]
+pub enum CommandAction {
+    Read {
+        command: String,
+        name: String,
+        path: PathBuf,
+    },
+    ListFiles {
+        command: String,
+        path: Option<String>,
+    },
+    Search {
+        command: String,
+        query: Option<String>,
+        path: Option<String>,
+    },
+    Unknown {
+        command: String,
+    },
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(rename_all = "camelCase", export_to = "v2/")]
+#[derive(Default)]
+pub enum SessionSource {
+    Cli,
+    #[serde(rename = "vscode")]
+    #[ts(rename = "vscode")]
+    #[default]
+    VsCode,
+    Exec,
+    AppServer,
+    #[serde(other)]
+    Unknown,
+}
+
+impl From<CoreSessionSource> for SessionSource {
+    fn from(value: CoreSessionSource) -> Self {
+        match value {
+            CoreSessionSource::Cli => SessionSource::Cli,
+            CoreSessionSource::VSCode => SessionSource::VsCode,
+            CoreSessionSource::Exec => SessionSource::Exec,
+            CoreSessionSource::Mcp => SessionSource::AppServer,
+            CoreSessionSource::SubAgent(_) => SessionSource::Unknown,
+            CoreSessionSource::Unknown => SessionSource::Unknown,
+        }
+    }
+}
+
+impl From<SessionSource> for CoreSessionSource {
+    fn from(value: SessionSource) -> Self {
+        match value {
+            SessionSource::Cli => CoreSessionSource::Cli,
+            SessionSource::VsCode => CoreSessionSource::VSCode,
+            SessionSource::Exec => CoreSessionSource::Exec,
+            SessionSource::AppServer => CoreSessionSource::Mcp,
+            SessionSource::Unknown => CoreSessionSource::Unknown,
+        }
+    }
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct GitInfo {
+    pub sha: Option<String>,
+    pub branch: Option<String>,
+    pub origin_url: Option<String>,
+}
+
+impl CommandAction {
+    pub fn into_core(self) -> CoreParsedCommand {
+        match self {
+            CommandAction::Read {
+                command: cmd,
+                name,
+                path,
+            } => CoreParsedCommand::Read { cmd, name, path },
+            CommandAction::ListFiles { command: cmd, path } => {
+                CoreParsedCommand::ListFiles { cmd, path }
+            }
+            CommandAction::Search {
+                command: cmd,
+                query,
+                path,
+            } => CoreParsedCommand::Search { cmd, query, path },
+            CommandAction::Unknown { command: cmd } => CoreParsedCommand::Unknown { cmd },
+        }
+    }
+}
+
+impl From<CoreParsedCommand> for CommandAction {
+    fn from(value: CoreParsedCommand) -> Self {
+        match value {
+            CoreParsedCommand::Read { cmd, name, path } => CommandAction::Read {
+                command: cmd,
+                name,
+                path,
+            },
+            CoreParsedCommand::ListFiles { cmd, path } => {
+                CommandAction::ListFiles { command: cmd, path }
+            }
+            CoreParsedCommand::Search { cmd, query, path } => CommandAction::Search {
+                command: cmd,
+                query,
+                path,
+            },
+            CoreParsedCommand::Unknown { cmd } => CommandAction::Unknown { command: cmd },
+        }
+    }
+}
+
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(tag = "type", rename_all = "camelCase")]
 #[ts(tag = "type")]
@@ -278,7 +512,7 @@ pub struct ThreadStartParams {
     pub cwd: Option<String>,
     pub approval_policy: Option<AskForApproval>,
     pub sandbox: Option<SandboxMode>,
-    pub config: Option<HashMap<String, serde_json::Value>>,
+    pub config: Option<HashMap<String, JsonValue>>,
     pub base_instructions: Option<String>,
     pub developer_instructions: Option<String>,
 }
@@ -288,13 +522,47 @@ pub struct ThreadStartParams {
 #[ts(export_to = "v2/")]
 pub struct ThreadStartResponse {
     pub thread: Thread,
+    pub model: String,
+    pub model_provider: String,
+    pub cwd: PathBuf,
+    pub approval_policy: AskForApproval,
+    pub sandbox: SandboxPolicy,
+    pub reasoning_effort: Option<ReasoningEffort>,
 }
 
-#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[derive(Serialize, Deserialize, Debug, Default, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
 #[ts(export_to = "v2/")]
+/// There are three ways to resume a thread:
+/// 1. By thread_id: load the thread from disk by thread_id and resume it.
+/// 2. By history: instantiate the thread from memory and resume it.
+/// 3. By path: load the thread from disk by path and resume it.
+///
+/// The precedence is: history > path > thread_id.
+/// If using history or path, the thread_id param will be ignored.
+///
+/// Prefer using thread_id whenever possible.
 pub struct ThreadResumeParams {
     pub thread_id: String,
+
+    /// [UNSTABLE] FOR CODEX CLOUD - DO NOT USE.
+    /// If specified, the thread will be resumed with the provided history
+    /// instead of loaded from disk.
+    pub history: Option<Vec<ResponseItem>>,
+
+    /// [UNSTABLE] Specify the rollout path to resume from.
+    /// If specified, the thread_id param will be ignored.
+    pub path: Option<PathBuf>,
+
+    /// Configuration overrides for the resumed thread, if any.
+    pub model: Option<String>,
+    pub model_provider: Option<String>,
+    pub cwd: Option<String>,
+    pub approval_policy: Option<AskForApproval>,
+    pub sandbox: Option<SandboxMode>,
+    pub config: Option<HashMap<String, serde_json::Value>>,
+    pub base_instructions: Option<String>,
+    pub developer_instructions: Option<String>,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
@@ -302,6 +570,12 @@ pub struct ThreadResumeParams {
 #[ts(export_to = "v2/")]
 pub struct ThreadResumeResponse {
     pub thread: Thread,
+    pub model: String,
+    pub model_provider: String,
+    pub cwd: PathBuf,
+    pub approval_policy: AskForApproval,
+    pub sandbox: SandboxPolicy,
+    pub reasoning_effort: Option<ReasoningEffort>,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
@@ -358,9 +632,24 @@ pub struct Thread {
     pub id: String,
     /// Usually the first user message in the thread, if available.
     pub preview: String,
+    /// Model provider used for this thread (for example, 'openai').
     pub model_provider: String,
     /// Unix timestamp (in seconds) when the thread was created.
     pub created_at: i64,
+    /// [UNSTABLE] Path to the thread on disk.
+    pub path: PathBuf,
+    /// Working directory captured for the thread.
+    pub cwd: PathBuf,
+    /// Version of the CLI that created the thread.
+    pub cli_version: String,
+    /// Origin of the thread (CLI, VSCode, codex exec, codex app-server, etc.).
+    pub source: SessionSource,
+    /// Optional Git metadata captured when the thread was created.
+    pub git_info: Option<GitInfo>,
+    /// Only populated on a `thread/resume` response.
+    /// For all other responses and notifications returning a Thread,
+    /// the turns field will be an empty list.
+    pub turns: Vec<Turn>,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
@@ -375,25 +664,37 @@ pub struct AccountUpdatedNotification {
 #[ts(export_to = "v2/")]
 pub struct Turn {
     pub id: String,
+    /// Only populated on a `thread/resume` response.
+    /// For all other responses and notifications returning a Turn,
+    /// the items field will be an empty list.
     pub items: Vec<ThreadItem>,
+    #[serde(flatten)]
     pub status: TurnStatus,
-    pub error: Option<TurnError>,
 }
 
-#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS, Error)]
 #[serde(rename_all = "camelCase")]
 #[ts(export_to = "v2/")]
+#[error("{message}")]
 pub struct TurnError {
     pub message: String,
+    pub codex_error_info: Option<CodexErrorInfo>,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
 #[ts(export_to = "v2/")]
+pub struct ErrorNotification {
+    pub error: TurnError,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(tag = "status", rename_all = "camelCase")]
+#[ts(tag = "status", export_to = "v2/")]
 pub enum TurnStatus {
     Completed,
     Interrupted,
-    Failed,
+    Failed { error: TurnError },
     InProgress,
 }
 
@@ -418,6 +719,45 @@ pub struct TurnStartParams {
     pub summary: Option<ReasoningSummary>,
 }
 
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ReviewStartParams {
+    pub thread_id: String,
+    pub target: ReviewTarget,
+
+    /// When true, also append the final review message to the original thread.
+    #[serde(default)]
+    pub append_to_original_thread: bool,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(tag = "type", rename_all = "camelCase")]
+#[ts(tag = "type", export_to = "v2/")]
+pub enum ReviewTarget {
+    /// Review the working tree: staged, unstaged, and untracked files.
+    UncommittedChanges,
+
+    /// Review changes between the current branch and the given base branch.
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
+    BaseBranch { branch: String },
+
+    /// Review the changes introduced by a specific commit.
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
+    Commit {
+        sha: String,
+        /// Optional human-readable label (e.g., commit subject) for UIs.
+        title: Option<String>,
+    },
+
+    /// Arbitrary instructions, equivalent to the old free-form prompt.
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
+    Custom { instructions: String },
+}
+
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
 #[ts(export_to = "v2/")]
@@ -475,31 +815,50 @@ impl From<CoreUserInput> for UserInput {
 #[ts(tag = "type")]
 #[ts(export_to = "v2/")]
 pub enum ThreadItem {
-    UserMessage {
-        id: String,
-        content: Vec<UserInput>,
-    },
-    AgentMessage {
-        id: String,
-        text: String,
-    },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
+    UserMessage { id: String, content: Vec<UserInput> },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
+    AgentMessage { id: String, text: String },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
     Reasoning {
         id: String,
-        text: String,
+        #[serde(default)]
+        summary: Vec<String>,
+        #[serde(default)]
+        content: Vec<String>,
     },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
     CommandExecution {
         id: String,
+        /// The command to be executed.
         command: String,
-        aggregated_output: String,
-        exit_code: Option<i32>,
+        /// The command's working directory.
+        cwd: PathBuf,
         status: CommandExecutionStatus,
+        /// A best-effort parsing of the command to understand the action(s) it will perform.
+        /// This returns a list of CommandAction objects because a single shell command may
+        /// be composed of many commands piped together.
+        command_actions: Vec<CommandAction>,
+        /// The command's output, aggregated from stdout and stderr.
+        aggregated_output: Option<String>,
+        /// The command's exit code.
+        exit_code: Option<i32>,
+        /// The duration of the command execution in milliseconds.
         duration_ms: Option<i64>,
     },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
     FileChange {
         id: String,
         changes: Vec<FileUpdateChange>,
         status: PatchApplyStatus,
     },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
     McpToolCall {
         id: String,
         server: String,
@@ -509,22 +868,18 @@ pub enum ThreadItem {
         result: Option<McpToolCallResult>,
         error: Option<McpToolCallError>,
     },
-    WebSearch {
-        id: String,
-        query: String,
-    },
-    TodoList {
-        id: String,
-        items: Vec<TodoItem>,
-    },
-    ImageView {
-        id: String,
-        path: String,
-    },
-    CodeReview {
-        id: String,
-        review: String,
-    },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
+    WebSearch { id: String, query: String },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
+    TodoList { id: String, items: Vec<TodoItem> },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
+    ImageView { id: String, path: String },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
+    CodeReview { id: String, review: String },
 }
 
 impl From<CoreTurnItem> for ThreadItem {
@@ -544,17 +899,11 @@ impl From<CoreTurnItem> for ThreadItem {
                     .collect::<String>();
                 ThreadItem::AgentMessage { id: agent.id, text }
             }
-            CoreTurnItem::Reasoning(reasoning) => {
-                let text = if !reasoning.summary_text.is_empty() {
-                    reasoning.summary_text.join("\n")
-                } else {
-                    reasoning.raw_content.join("\n")
-                };
-                ThreadItem::Reasoning {
-                    id: reasoning.id,
-                    text,
-                }
-            }
+            CoreTurnItem::Reasoning(reasoning) => ThreadItem::Reasoning {
+                id: reasoning.id,
+                summary: reasoning.summary_text,
+                content: reasoning.raw_content,
+            },
             CoreTurnItem::WebSearch(search) => ThreadItem::WebSearch {
                 id: search.id,
                 query: search.query,
@@ -570,6 +919,7 @@ pub enum CommandExecutionStatus {
     InProgress,
     Completed,
     Failed,
+    Declined,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
@@ -582,20 +932,23 @@ pub struct FileUpdateChange {
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
-#[serde(rename_all = "camelCase")]
+#[serde(tag = "type", rename_all = "camelCase")]
+#[ts(tag = "type")]
 #[ts(export_to = "v2/")]
 pub enum PatchChangeKind {
     Add,
     Delete,
-    Update,
+    Update { move_path: Option<PathBuf> },
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
 #[ts(export_to = "v2/")]
 pub enum PatchApplyStatus {
+    InProgress,
     Completed,
     Failed,
+    Declined,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
@@ -612,7 +965,7 @@ pub enum McpToolCallStatus {
 #[ts(export_to = "v2/")]
 pub struct McpToolCallResult {
     pub content: Vec<McpContentBlock>,
-    pub structured_content: JsonValue,
+    pub structured_content: Option<JsonValue>,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
@@ -661,8 +1014,6 @@ pub struct Usage {
 #[ts(export_to = "v2/")]
 pub struct TurnCompletedNotification {
     pub turn: Turn,
-    // TODO: should usage be stored on the Turn object, and we return that instead?
-    pub usage: Usage,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
@@ -688,6 +1039,32 @@ pub struct AgentMessageDeltaNotification {
     pub delta: String,
 }
 
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ReasoningSummaryTextDeltaNotification {
+    pub item_id: String,
+    pub delta: String,
+    pub summary_index: i64,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ReasoningSummaryPartAddedNotification {
+    pub item_id: String,
+    pub summary_index: i64,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ReasoningTextDeltaNotification {
+    pub item_id: String,
+    pub delta: String,
+    pub content_index: i64,
+}
+
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
 #[ts(export_to = "v2/")]
@@ -704,6 +1081,68 @@ pub struct McpToolCallProgressNotification {
     pub message: String,
 }
 
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct WindowsWorldWritableWarningNotification {
+    pub sample_paths: Vec<String>,
+    pub extra_count: usize,
+    pub failed_scan: bool,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct CommandExecutionRequestApprovalParams {
+    pub thread_id: String,
+    pub turn_id: String,
+    pub item_id: String,
+    /// Optional explanatory reason (e.g. request for network access).
+    pub reason: Option<String>,
+    /// Optional model-provided risk assessment describing the blocked command.
+    pub risk: Option<SandboxCommandAssessment>,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct CommandExecutionRequestAcceptSettings {
+    /// If true, automatically approve this command for the duration of the session.
+    #[serde(default)]
+    pub for_session: bool,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct CommandExecutionRequestApprovalResponse {
+    pub decision: ApprovalDecision,
+    /// Optional approval settings for when the decision is `accept`.
+    /// Ignored if the decision is `decline` or `cancel`.
+    #[serde(default)]
+    pub accept_settings: Option<CommandExecutionRequestAcceptSettings>,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct FileChangeRequestApprovalParams {
+    pub thread_id: String,
+    pub turn_id: String,
+    pub item_id: String,
+    /// Optional explanatory reason (e.g. request for extra write access).
+    pub reason: Option<String>,
+    /// [UNSTABLE] 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)]
+#[ts(export_to = "v2/")]
+pub struct FileChangeRequestApprovalResponse {
+    pub decision: ApprovalDecision,
+}
+
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
 #[ts(export_to = "v2/")]
@@ -717,6 +1156,7 @@ pub struct AccountRateLimitsUpdatedNotification {
 pub struct RateLimitSnapshot {
     pub primary: Option<RateLimitWindow>,
     pub secondary: Option<RateLimitWindow>,
+    pub credits: Option<CreditsSnapshot>,
 }
 
 impl From<CoreRateLimitSnapshot> for RateLimitSnapshot {
@@ -724,6 +1164,7 @@ impl From<CoreRateLimitSnapshot> for RateLimitSnapshot {
         Self {
             primary: value.primary.map(RateLimitWindow::from),
             secondary: value.secondary.map(RateLimitWindow::from),
+            credits: value.credits.map(CreditsSnapshot::from),
         }
     }
 }
@@ -747,6 +1188,25 @@ impl From<CoreRateLimitWindow> for RateLimitWindow {
     }
 }
 
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct CreditsSnapshot {
+    pub has_credits: bool,
+    pub unlimited: bool,
+    pub balance: Option<String>,
+}
+
+impl From<CoreCreditsSnapshot> for CreditsSnapshot {
+    fn from(value: CoreCreditsSnapshot) -> Self {
+        Self {
+            has_credits: value.has_credits,
+            unlimited: value.unlimited,
+            balance: value.balance,
+        }
+    }
+}
+
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
 #[ts(export_to = "v2/")]
@@ -769,6 +1229,7 @@ mod tests {
     use codex_protocol::items::WebSearchItem;
     use codex_protocol::user_input::UserInput as CoreUserInput;
     use pretty_assertions::assert_eq;
+    use serde_json::json;
     use std::path::PathBuf;
 
     #[test]
@@ -836,7 +1297,8 @@ mod tests {
             ThreadItem::from(reasoning_item),
             ThreadItem::Reasoning {
                 id: "reasoning-1".to_string(),
-                text: "line one\nline two".to_string(),
+                summary: vec!["line one".to_string(), "line two".to_string()],
+                content: vec![],
             }
         );
 
@@ -853,4 +1315,20 @@ mod tests {
             }
         );
     }
+
+    #[test]
+    fn codex_error_info_serializes_http_status_code_in_camel_case() {
+        let value = CodexErrorInfo::ResponseTooManyFailedAttempts {
+            http_status_code: Some(401),
+        };
+
+        assert_eq!(
+            serde_json::to_value(value).unwrap(),
+            json!({
+                "responseTooManyFailedAttempts": {
+                    "httpStatusCode": 401
+                }
+            })
+        );
+    }
 }

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

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

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

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

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

@@ -0,0 +1,809 @@
+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,
+    /// Trigger the ChatGPT login flow and wait for completion.
+    TestLogin,
+    /// Fetch the current account rate limits from the Codex app-server.
+    GetAccountRateLimits,
+}
+
+fn main() -> Result<()> {
+    let Cli { codex_bin, command } = Cli::parse();
+
+    match command {
+        CliCommand::SendMessage { user_message } => send_message(codex_bin, user_message),
+        CliCommand::SendMessageV2 { user_message } => send_message_v2(codex_bin, user_message),
+        CliCommand::TriggerCmdApproval { user_message } => {
+            trigger_cmd_approval(codex_bin, user_message)
+        }
+        CliCommand::TriggerPatchApproval { user_message } => {
+            trigger_patch_approval(codex_bin, user_message)
+        }
+        CliCommand::NoTriggerCmdApproval => no_trigger_cmd_approval(codex_bin),
+        CliCommand::TestLogin => test_login(codex_bin),
+        CliCommand::GetAccountRateLimits => get_account_rate_limits(codex_bin),
+    }
+}
+
+fn send_message(codex_bin: String, user_message: String) -> Result<()> {
+    let mut client = CodexClient::spawn(codex_bin)?;
+
+    let initialize = client.initialize()?;
+    println!("< initialize response: {initialize:?}");
+
+    let conversation = client.new_conversation()?;
+    println!("< newConversation response: {conversation:?}");
+
+    let subscription = client.add_conversation_listener(&conversation.conversation_id)?;
+    println!("< addConversationListener response: {subscription:?}");
+
+    let send_response = client.send_user_message(&conversation.conversation_id, &user_message)?;
+    println!("< sendUserMessage response: {send_response:?}");
+
+    client.stream_conversation(&conversation.conversation_id)?;
+
+    client.remove_conversation_listener(subscription.subscription_id)?;
+
+    Ok(())
+}
+
+fn send_message_v2(codex_bin: String, user_message: String) -> Result<()> {
+    send_message_v2_with_policies(codex_bin, user_message, None, None)
+}
+
+fn trigger_cmd_approval(codex_bin: String, user_message: Option<String>) -> Result<()> {
+    let default_prompt =
+        "Run `touch /tmp/should-trigger-approval` so I can confirm the file exists.";
+    let message = user_message.unwrap_or_else(|| default_prompt.to_string());
+    send_message_v2_with_policies(
+        codex_bin,
+        message,
+        Some(AskForApproval::OnRequest),
+        Some(SandboxPolicy::ReadOnly),
+    )
+}
+
+fn trigger_patch_approval(codex_bin: String, user_message: Option<String>) -> Result<()> {
+    let default_prompt =
+        "Create a file named APPROVAL_DEMO.txt containing a short hello message using apply_patch.";
+    let message = user_message.unwrap_or_else(|| default_prompt.to_string());
+    send_message_v2_with_policies(
+        codex_bin,
+        message,
+        Some(AskForApproval::OnRequest),
+        Some(SandboxPolicy::ReadOnly),
+    )
+}
+
+fn no_trigger_cmd_approval(codex_bin: String) -> Result<()> {
+    let prompt = "Run `touch should_not_trigger_approval.txt`";
+    send_message_v2_with_policies(codex_bin, prompt.to_string(), None, None)
+}
+
+fn send_message_v2_with_policies(
+    codex_bin: String,
+    user_message: String,
+    approval_policy: Option<AskForApproval>,
+    sandbox_policy: Option<SandboxPolicy>,
+) -> Result<()> {
+    let mut client = CodexClient::spawn(codex_bin)?;
+
+    let initialize = client.initialize()?;
+    println!("< initialize response: {initialize:?}");
+
+    let thread_response = client.thread_start(ThreadStartParams::default())?;
+    println!("< thread/start response: {thread_response:?}");
+    let mut turn_params = TurnStartParams {
+        thread_id: thread_response.thread.id.clone(),
+        input: vec![V2UserInput::Text { text: user_message }],
+        ..Default::default()
+    };
+    turn_params.approval_policy = approval_policy;
+    turn_params.sandbox_policy = sandbox_policy;
+
+    let turn_response = client.turn_start(turn_params)?;
+    println!("< turn/start response: {turn_response:?}");
+
+    client.stream_turn(&thread_response.thread.id, &turn_response.turn.id)?;
+
+    Ok(())
+}
+
+fn test_login(codex_bin: String) -> Result<()> {
+    let mut client = CodexClient::spawn(codex_bin)?;
+
+    let initialize = client.initialize()?;
+    println!("< initialize response: {initialize:?}");
+
+    let login_response = client.login_chat_gpt()?;
+    println!("< loginChatGpt response: {login_response:?}");
+    println!(
+        "Open the following URL in your browser to continue:\n{}",
+        login_response.auth_url
+    );
+
+    let completion = client.wait_for_login_completion(&login_response.login_id)?;
+    println!("< loginChatGptComplete notification: {completion:?}");
+
+    if completion.success {
+        println!("Login succeeded.");
+        Ok(())
+    } else {
+        bail!(
+            "login failed: {}",
+            completion
+                .error
+                .as_deref()
+                .unwrap_or("unknown error from loginChatGptComplete")
+        );
+    }
+}
+
+fn get_account_rate_limits(codex_bin: String) -> Result<()> {
+    let mut client = CodexClient::spawn(codex_bin)?;
+
+    let initialize = client.initialize()?;
+    println!("< initialize response: {initialize:?}");
+
+    let response = client.get_account_rate_limits()?;
+    println!("< account/rateLimits/read response: {response:?}");
+
+    Ok(())
+}
+
+struct CodexClient {
+    child: Child,
+    stdin: Option<ChildStdin>,
+    stdout: BufReader<ChildStdout>,
+    pending_notifications: VecDeque<JSONRPCNotification>,
+}
+
+impl CodexClient {
+    fn spawn(codex_bin: String) -> Result<Self> {
+        let mut codex_app_server = Command::new(&codex_bin)
+            .arg("app-server")
+            .stdin(Stdio::piped())
+            .stdout(Stdio::piped())
+            .stderr(Stdio::inherit())
+            .spawn()
+            .with_context(|| format!("failed to start `{codex_bin}` app-server"))?;
+
+        let stdin = codex_app_server
+            .stdin
+            .take()
+            .context("codex app-server stdin unavailable")?;
+        let stdout = codex_app_server
+            .stdout
+            .take()
+            .context("codex app-server stdout unavailable")?;
+
+        Ok(Self {
+            child: codex_app_server,
+            stdin: Some(stdin),
+            stdout: BufReader::new(stdout),
+            pending_notifications: VecDeque::new(),
+        })
+    }
+
+    fn initialize(&mut self) -> Result<InitializeResponse> {
+        let request_id = self.request_id();
+        let request = ClientRequest::Initialize {
+            request_id: request_id.clone(),
+            params: InitializeParams {
+                client_info: ClientInfo {
+                    name: "codex-toy-app-server".to_string(),
+                    title: Some("Codex Toy App Server".to_string()),
+                    version: env!("CARGO_PKG_VERSION").to_string(),
+                },
+            },
+        };
+
+        self.send_request(request, request_id, "initialize")
+    }
+
+    fn new_conversation(&mut self) -> Result<NewConversationResponse> {
+        let request_id = self.request_id();
+        let request = ClientRequest::NewConversation {
+            request_id: request_id.clone(),
+            params: NewConversationParams::default(),
+        };
+
+        self.send_request(request, request_id, "newConversation")
+    }
+
+    fn add_conversation_listener(
+        &mut self,
+        conversation_id: &ConversationId,
+    ) -> Result<AddConversationSubscriptionResponse> {
+        let request_id = self.request_id();
+        let request = ClientRequest::AddConversationListener {
+            request_id: request_id.clone(),
+            params: AddConversationListenerParams {
+                conversation_id: *conversation_id,
+                experimental_raw_events: false,
+            },
+        };
+
+        self.send_request(request, request_id, "addConversationListener")
+    }
+
+    fn remove_conversation_listener(&mut self, subscription_id: Uuid) -> Result<()> {
+        let request_id = self.request_id();
+        let request = ClientRequest::RemoveConversationListener {
+            request_id: request_id.clone(),
+            params: codex_app_server_protocol::RemoveConversationListenerParams { subscription_id },
+        };
+
+        self.send_request::<codex_app_server_protocol::RemoveConversationSubscriptionResponse>(
+            request,
+            request_id,
+            "removeConversationListener",
+        )?;
+
+        Ok(())
+    }
+
+    fn send_user_message(
+        &mut self,
+        conversation_id: &ConversationId,
+        message: &str,
+    ) -> Result<SendUserMessageResponse> {
+        let request_id = self.request_id();
+        let request = ClientRequest::SendUserMessage {
+            request_id: request_id.clone(),
+            params: SendUserMessageParams {
+                conversation_id: *conversation_id,
+                items: vec![InputItem::Text {
+                    text: message.to_string(),
+                }],
+            },
+        };
+
+        self.send_request(request, request_id, "sendUserMessage")
+    }
+
+    fn thread_start(&mut self, params: ThreadStartParams) -> Result<ThreadStartResponse> {
+        let request_id = self.request_id();
+        let request = ClientRequest::ThreadStart {
+            request_id: request_id.clone(),
+            params,
+        };
+
+        self.send_request(request, request_id, "thread/start")
+    }
+
+    fn turn_start(&mut self, params: TurnStartParams) -> Result<TurnStartResponse> {
+        let request_id = self.request_id();
+        let request = ClientRequest::TurnStart {
+            request_id: request_id.clone(),
+            params,
+        };
+
+        self.send_request(request, request_id, "turn/start")
+    }
+
+    fn login_chat_gpt(&mut self) -> Result<LoginChatGptResponse> {
+        let request_id = self.request_id();
+        let request = ClientRequest::LoginChatGpt {
+            request_id: request_id.clone(),
+            params: None,
+        };
+
+        self.send_request(request, request_id, "loginChatGpt")
+    }
+
+    fn get_account_rate_limits(&mut self) -> Result<GetAccountRateLimitsResponse> {
+        let request_id = self.request_id();
+        let request = ClientRequest::GetAccountRateLimits {
+            request_id: request_id.clone(),
+            params: None,
+        };
+
+        self.send_request(request, request_id, "account/rateLimits/read")
+    }
+
+    fn stream_conversation(&mut self, conversation_id: &ConversationId) -> Result<()> {
+        loop {
+            let notification = self.next_notification()?;
+
+            if !notification.method.starts_with("codex/event/") {
+                continue;
+            }
+
+            if let Some(event) = self.extract_event(notification, conversation_id)? {
+                match &event.msg {
+                    EventMsg::AgentMessage(event) => {
+                        println!("{}", event.message);
+                    }
+                    EventMsg::AgentMessageDelta(event) => {
+                        print!("{}", event.delta);
+                        std::io::stdout().flush().ok();
+                    }
+                    EventMsg::TaskComplete(event) => {
+                        println!("\n[task complete: {event:?}]");
+                        break;
+                    }
+                    EventMsg::TurnAborted(event) => {
+                        println!("\n[turn aborted: {:?}]", event.reason);
+                        break;
+                    }
+                    EventMsg::Error(event) => {
+                        println!("[error] {event:?}");
+                    }
+                    _ => {
+                        println!("[UNKNOWN EVENT] {:?}", event.msg);
+                    }
+                }
+            }
+        }
+
+        Ok(())
+    }
+
+    fn wait_for_login_completion(
+        &mut self,
+        expected_login_id: &Uuid,
+    ) -> Result<LoginChatGptCompleteNotification> {
+        loop {
+            let notification = self.next_notification()?;
+
+            if let Ok(server_notification) = ServerNotification::try_from(notification) {
+                match server_notification {
+                    ServerNotification::LoginChatGptComplete(completion) => {
+                        if &completion.login_id == expected_login_id {
+                            return Ok(completion);
+                        }
+
+                        println!(
+                            "[ignoring loginChatGptComplete for unexpected login_id: {}]",
+                            completion.login_id
+                        );
+                    }
+                    ServerNotification::AuthStatusChange(status) => {
+                        println!("< authStatusChange notification: {status:?}");
+                    }
+                    ServerNotification::AccountRateLimitsUpdated(snapshot) => {
+                        println!("< accountRateLimitsUpdated notification: {snapshot:?}");
+                    }
+                    ServerNotification::SessionConfigured(_) => {
+                        // SessionConfigured notifications are unrelated to login; skip.
+                    }
+                    _ => {}
+                }
+            }
+
+            // Not a server notification (likely a conversation event); keep waiting.
+        }
+    }
+
+    fn stream_turn(&mut self, thread_id: &str, turn_id: &str) -> Result<()> {
+        loop {
+            let notification = self.next_notification()?;
+
+            let Ok(server_notification) = ServerNotification::try_from(notification) else {
+                continue;
+            };
+
+            match server_notification {
+                ServerNotification::ThreadStarted(payload) => {
+                    if payload.thread.id == thread_id {
+                        println!("< thread/started notification: {:?}", payload.thread);
+                    }
+                }
+                ServerNotification::TurnStarted(payload) => {
+                    if payload.turn.id == turn_id {
+                        println!("< turn/started notification: {:?}", payload.turn.status);
+                    }
+                }
+                ServerNotification::AgentMessageDelta(delta) => {
+                    print!("{}", delta.delta);
+                    std::io::stdout().flush().ok();
+                }
+                ServerNotification::CommandExecutionOutputDelta(delta) => {
+                    print!("{}", delta.delta);
+                    std::io::stdout().flush().ok();
+                }
+                ServerNotification::ItemStarted(payload) => {
+                    println!("\n< item started: {:?}", payload.item);
+                }
+                ServerNotification::ItemCompleted(payload) => {
+                    println!("< item completed: {:?}", payload.item);
+                }
+                ServerNotification::TurnCompleted(payload) => {
+                    if payload.turn.id == turn_id {
+                        println!("\n< turn/completed notification: {:?}", payload.turn.status);
+                        if let TurnStatus::Failed { error } = &payload.turn.status {
+                            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

@@ -46,9 +46,11 @@ 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

@@ -2,6 +2,16 @@
 
 `codex app-server` is the interface Codex uses to power rich interfaces such as the [Codex VS Code extension](https://marketplace.visualstudio.com/items?itemName=openai.chatgpt). The message schema is currently unstable, but those who wish to build experimental UIs on top of Codex may find it valuable.
 
+## Table of Contents
+- [Protocol](#protocol)
+- [Message Schema](#message-schema)
+- [Lifecycle Overview](#lifecycle-overview)
+- [Initialization](#initialization)
+- [Core primitives](#core-primitives)
+- [Thread & turn endpoints](#thread--turn-endpoints)
+- [Events (work-in-progress)](#events-work-in-progress)
+- [Auth endpoints](#auth-endpoints)
+
 ## Protocol
 
 Similar to [MCP](https://modelcontextprotocol.io/), `codex app-server` supports bidirectional communication, streaming JSONL over stdio. The protocol is JSON-RPC 2.0, though the `"jsonrpc":"2.0"` header is omitted.
@@ -15,6 +25,14 @@ codex app-server generate-ts --out DIR
 codex app-server generate-json-schema --out DIR
 ```
 
+## Lifecycle Overview
+
+- Initialize once: Immediately after launching the codex app-server process, send an `initialize` request with your client metadata, then emit an `initialized` notification. Any other request before this handshake gets rejected.
+- Start (or resume) a thread: Call `thread/start` to open a fresh conversation. The response returns the thread object and you’ll also get a `thread/started` notification. If you’re continuing an existing conversation, call `thread/resume` with its ID instead.
+- Begin a turn: To send user input, call `turn/start` with the target `threadId` and the user's input. Optional fields let you override model, cwd, sandbox policy, etc. This immediately returns the new turn object and triggers a `turn/started` notification.
+- Stream events: After `turn/start`, keep reading JSON-RPC notifications on stdout. You’ll see `item/started`, `item/completed`, deltas like `item/agentMessage/delta`, tool progress, etc. These represent streaming model output plus any side effects (commands, tool calls, reasoning notes).
+- Finish the turn: When the model is done (or the turn is interrupted via making the `turn/interrupt` call), the server sends `turn/completed` with the final turn state and token usage.
+
 ## Initialization
 
 Clients must send a single `initialize` request before invoking any other method, then acknowledge with an `initialized` notification. The server returns the user agent string it will present to upstream services; subsequent requests issued before initialization receive a `"Not initialized"` error, and repeated `initialize` calls receive an `"Already initialized"` error.
@@ -47,6 +65,7 @@ The JSON-RPC API exposes dedicated methods for managing Codex conversations. Thr
 - `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 a `item/completed` notification with a `codeReview` item when results are ready.
 
 ### 1) Start or resume a thread
 
@@ -56,7 +75,7 @@ Start a fresh thread when you need a new Codex conversation.
 { "method": "thread/start", "id": 10, "params": {
     // Optionally set config settings. If not specified, will use the user's
     // current config settings.
-    "model": "gpt-5-codex",
+    "model": "gpt-5.1-codex",
     "cwd": "/Users/me/project",
     "approvalPolicy": "never",
     "sandbox": "workspaceWrite",
@@ -137,7 +156,7 @@ 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"
 } }
@@ -163,6 +182,142 @@ 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.
 
+### 6) 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.
+- `appendToOriginalThread` (bool, default `false`) — when `true`, Codex also records a final assistant-style message with the review summary in the original thread. When `false`, only the `codeReview` item is emitted for the review run and no extra message is added to the original thread.
+
+Example request/response:
+
+```json
+{ "method": "review/start", "id": 40, "params": {
+    "threadId": "thr_123",
+    "appendToOriginalThread": true,
+    "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
+} } }
+```
+
+Codex streams the usual `turn/started` notification followed by an `item/started`
+with the same `codeReview` item id so clients can show progress:
+
+```json
+{ "method": "item/started", "params": { "item": {
+    "type": "codeReview",
+    "id": "turn_900",
+    "review": "current changes"
+} } }
+```
+
+When the reviewer finishes, the server emits `item/completed` containing the same
+`codeReview` item with the final review text:
+
+```json
+{ "method": "item/completed", "params": { "item": {
+    "type": "codeReview",
+    "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::CodeReview` in the generated schema). Use this notification to render the reviewer output in your client.
+
+## Events (work-in-progress)
+
+Event notifications are the server-initiated event stream for thread lifecycles, turn lifecycles, and the items within them. After you start or resume a thread, keep reading stdout for `thread/started`, `turn/*`, and `item/*` notifications.
+
+### Turn events
+
+The app-server streams JSON-RPC notifications while a turn is running. Each turn starts with `turn/started` (initial `turn`) and ends with `turn/completed` (final `turn` plus token `usage`), and clients subscribe to the events they care about, rendering each item incrementally as updates arrive. The per-item lifecycle is always: `item/started` → zero or more item-specific deltas → `item/completed`.
+
+- `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? } }`.
+
+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.
+
+#### Thread items
+
+`ThreadItem` is the tagged union carried in turn responses and `item/*` notifications. Currently we support events for the following items:
+- `userMessage` — `{id, content}` where `content` is a list of user inputs (`text`, `image`, or `localImage`).
+- `agentMessage` — `{id, text}` containing the accumulated agent reply.
+- `reasoning` — `{id, summary, content}` where `summary` holds streamed reasoning summaries (applicable for most OpenAI models) and `content` holds raw reasoning blocks (applicable for e.g. open source models).
+- `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.
+
+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
+`fileChange` items contain a `changes` list with `{path, kind, diff}` entries (`kind` is `add`, `delete`, or `update` with an optional `movePath`). The `status` tracks whether apply succeeded (`completed`), failed, or was `declined`.
+
+### 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.

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

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

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

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

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

@@ -6,7 +6,6 @@ use crate::outgoing_message::OutgoingMessageSender;
 use codex_app_server_protocol::ClientInfo;
 use codex_app_server_protocol::ClientRequest;
 use codex_app_server_protocol::InitializeResponse;
-
 use codex_app_server_protocol::JSONRPCError;
 use codex_app_server_protocol::JSONRPCErrorError;
 use codex_app_server_protocol::JSONRPCNotification;
@@ -118,6 +117,7 @@ impl MessageProcessor {
                     self.outgoing.send_response(request_id, response).await;
 
                     self.initialized = true;
+
                     return;
                 }
             }

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

@@ -24,3 +24,5 @@ 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 +9,14 @@ 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_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

@@ -35,6 +35,7 @@ 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 +378,15 @@ 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,

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 +77,7 @@ pub fn create_apply_patch_sse_response(
                         {
                             "id": call_id,
                             "function": {
-                                "name": "shell",
+                                "name": "shell_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;
@@ -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

@@ -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/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/mod.rs

@@ -1,6 +1,7 @@
 mod account;
 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

@@ -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 +1,279 @@
+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::ReviewStartParams;
+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::TurnStartResponse;
+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(),
+            append_to_original_thread: true,
+            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 TurnStartResponse { turn } = to_response::<TurnStartResponse>(review_resp)?;
+    let turn_id = turn.id.clone();
+    assert_eq!(turn.status, TurnStatus::InProgress);
+    assert_eq!(turn.items.len(), 1);
+    match &turn.items[0] {
+        ThreadItem::UserMessage { content, .. } => {
+            assert_eq!(content.len(), 1);
+            assert!(matches!(
+                &content[0],
+                codex_app_server_protocol::UserInput::Text { .. }
+            ));
+        }
+        other => panic!("expected user message, got {other:?}"),
+    }
+
+    let _started: JSONRPCNotification = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("turn/started"),
+    )
+    .await??;
+    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::CodeReview { id, review } => {
+            assert_eq!(id, turn_id);
+            assert_eq!(review, "commit 1234567");
+        }
+        other => panic!("expected code review item, got {other:?}"),
+    }
+
+    let mut review_body: Option<String> = None;
+    for _ in 0..5 {
+        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::CodeReview { id, review } => {
+                assert_eq!(id, turn_id);
+                review_body = Some(review);
+                break;
+            }
+            ThreadItem::UserMessage { .. } => continue,
+            other => panic!("unexpected item/completed payload: {other:?}"),
+        }
+    }
+
+    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,
+            append_to_original_thread: true,
+            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_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,
+            append_to_original_thread: true,
+            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,
+            append_to_original_thread: true,
+            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
@@ -99,7 +102,18 @@ 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.turn.status, TurnStatus::Interrupted);
+
     Ok(())
 }
 

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

@@ -1,25 +1,35 @@
 use anyhow::Result;
 use app_test_support::McpProcess;
+use app_test_support::create_apply_patch_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::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 core_test_support::skip_if_no_network;
 use pretty_assertions::assert_eq;
 use std::path::Path;
@@ -57,7 +67,7 @@ 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
@@ -118,13 +128,17 @@ 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.turn.status, TurnStatus::Completed);
 
     Ok(())
 }
@@ -157,7 +171,7 @@ 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 +205,7 @@ 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 +216,7 @@ 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 +247,9 @@ 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 +272,10 @@ 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 +288,11 @@ 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 +315,151 @@ 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 +471,6 @@ 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 +484,15 @@ 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 +517,7 @@ 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 +571,36 @@ 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 +611,308 @@ 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(()));
+    if cfg!(windows) {
+        // TODO apply_patch approvals are not parsed from powershell commands yet
+        return 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 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(()));
+    if cfg!(windows) {
+        // TODO apply_patch approvals are not parsed from powershell commands yet
+        return 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(())
+}
+
 // Helper to create a config.toml pointing at the mock model server.
 fn create_config_toml(
     codex_home: &Path,

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

@@ -30,6 +30,7 @@ pub use standalone_executable::main;
 pub const APPLY_PATCH_TOOL_INSTRUCTIONS: &str = include_str!("../apply_patch_tool_instructions.md");
 
 const APPLY_PATCH_COMMANDS: [&str; 2] = ["apply_patch", "applypatch"];
+const APPLY_PATCH_SHELLS: [&str; 3] = ["bash", "zsh", "sh"];
 
 #[derive(Debug, Error, PartialEq)]
 pub enum ApplyPatchError {
@@ -96,6 +97,13 @@ pub struct ApplyPatchArgs {
     pub workdir: Option<String>,
 }
 
+fn shell_supports_apply_patch(shell: &str) -> bool {
+    std::path::Path::new(shell)
+        .file_name()
+        .and_then(|name| name.to_str())
+        .is_some_and(|name| APPLY_PATCH_SHELLS.contains(&name))
+}
+
 pub fn maybe_parse_apply_patch(argv: &[String]) -> MaybeApplyPatch {
     match argv {
         // Direct invocation: apply_patch <patch>
@@ -104,7 +112,7 @@ pub fn maybe_parse_apply_patch(argv: &[String]) -> MaybeApplyPatch {
             Err(e) => MaybeApplyPatch::PatchParseError(e),
         },
         // Bash heredoc form: (optional `cd <path> &&`) apply_patch <<'EOF' ...
-        [bash, flag, script] if bash == "bash" && flag == "-lc" => {
+        [shell, flag, script] if shell_supports_apply_patch(shell) && flag == "-lc" => {
             match extract_apply_patch_from_bash(script) {
                 Ok((body, workdir)) => match parse_patch(&body) {
                     Ok(mut source) => {
@@ -224,12 +232,12 @@ pub fn maybe_parse_apply_patch_verified(argv: &[String], cwd: &Path) -> MaybeApp
                 );
             }
         }
-        [bash, flag, script] if bash == "bash" && flag == "-lc" => {
-            if parse_patch(script).is_ok() {
-                return MaybeApplyPatchVerified::CorrectnessError(
-                    ApplyPatchError::ImplicitInvocation,
-                );
-            }
+        [shell, flag, script]
+            if shell_supports_apply_patch(shell)
+                && flag == "-lc"
+                && parse_patch(script).is_ok() =>
+        {
+            return MaybeApplyPatchVerified::CorrectnessError(ApplyPatchError::ImplicitInvocation);
         }
         _ => {}
     }

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/cli/Cargo.toml

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

codex-rs/cli/src/debug_sandbox.rs

@@ -138,11 +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();
@@ -153,13 +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;

codex-rs/cli/src/main.rs

@@ -18,6 +18,7 @@ use codex_cli::login::run_logout;
 use codex_cloud_tasks::Cli as CloudTasksCli;
 use codex_common::CliConfigOverrides;
 use codex_exec::Cli as ExecCli;
+use codex_execpolicy::ExecPolicyCheckCommand;
 use codex_responses_api_proxy::Args as ResponsesApiProxyArgs;
 use codex_tui::AppExitInfo;
 use codex_tui::Cli as TuiCli;
@@ -93,6 +94,10 @@ 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 +139,10 @@ 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 +167,19 @@ 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 +345,10 @@ 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`.
@@ -448,6 +474,7 @@ 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 +482,7 @@ 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 +571,9 @@ 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 +642,7 @@ 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 +651,7 @@ 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 +735,21 @@ 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 +802,9 @@ 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 +816,7 @@ 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 +825,7 @@ 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 +834,14 @@ 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 +859,7 @@ mod tests {
                 "--ask-for-approval",
                 "on-request",
                 "-m",
-                "gpt-5-test",
+                "gpt-5.1-test",
                 "-p",
                 "my-profile",
                 "-C",
@@ -819,7 +870,7 @@ 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

@@ -79,6 +79,7 @@ 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,

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/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-backend-openapi-models/src/models/credit_status_details.rs

@@ -0,0 +1,52 @@
+/*
+ * codex-backend
+ *
+ * codex-backend
+ *
+ * The version of the OpenAPI document: 0.0.1
+ *
+ * Generated by: https://openapi-generator.tech
+ */
+use serde::Deserialize;
+use serde::Serialize;
+
+#[derive(Clone, Default, Debug, PartialEq, Serialize, Deserialize)]
+pub struct CreditStatusDetails {
+    #[serde(rename = "has_credits")]
+    pub has_credits: bool,
+    #[serde(rename = "unlimited")]
+    pub unlimited: bool,
+    #[serde(
+        rename = "balance",
+        default,
+        with = "::serde_with::rust::double_option",
+        skip_serializing_if = "Option::is_none"
+    )]
+    pub balance: Option<Option<String>>,
+    #[serde(
+        rename = "approx_local_messages",
+        default,
+        with = "::serde_with::rust::double_option",
+        skip_serializing_if = "Option::is_none"
+    )]
+    pub approx_local_messages: Option<Option<Vec<serde_json::Value>>>,
+    #[serde(
+        rename = "approx_cloud_messages",
+        default,
+        with = "::serde_with::rust::double_option",
+        skip_serializing_if = "Option::is_none"
+    )]
+    pub approx_cloud_messages: Option<Option<Vec<serde_json::Value>>>,
+}
+
+impl CreditStatusDetails {
+    pub fn new(has_credits: bool, unlimited: bool) -> CreditStatusDetails {
+        CreditStatusDetails {
+            has_credits,
+            unlimited,
+            balance: None,
+            approx_local_messages: None,
+            approx_cloud_messages: None,
+        }
+    }
+}

codex-rs/codex-backend-openapi-models/src/models/mod.rs

@@ -32,3 +32,6 @@ pub use self::rate_limit_status_details::RateLimitStatusDetails;
 
 pub mod rate_limit_window_snapshot;
 pub use self::rate_limit_window_snapshot::RateLimitWindowSnapshot;
+
+pub mod credit_status_details;
+pub use self::credit_status_details::CreditStatusDetails;

codex-rs/codex-backend-openapi-models/src/models/rate_limit_status_payload.rs

@@ -23,6 +23,13 @@ pub struct RateLimitStatusPayload {
         skip_serializing_if = "Option::is_none"
     )]
     pub rate_limit: Option<Option<Box<models::RateLimitStatusDetails>>>,
+    #[serde(
+        rename = "credits",
+        default,
+        with = "::serde_with::rust::double_option",
+        skip_serializing_if = "Option::is_none"
+    )]
+    pub credits: Option<Option<Box<models::CreditStatusDetails>>>,
 }
 
 impl RateLimitStatusPayload {
@@ -30,12 +37,15 @@ impl RateLimitStatusPayload {
         RateLimitStatusPayload {
             plan_type,
             rate_limit: None,
+            credits: None,
         }
     }
 }
 
 #[derive(Clone, Copy, Debug, Eq, PartialEq, Ord, PartialOrd, Hash, Serialize, Deserialize)]
 pub enum PlanType {
+    #[serde(rename = "guest")]
+    Guest,
     #[serde(rename = "free")]
     Free,
     #[serde(rename = "go")]
@@ -44,6 +54,8 @@ pub enum PlanType {
     Plus,
     #[serde(rename = "pro")]
     Pro,
+    #[serde(rename = "free_workspace")]
+    FreeWorkspace,
     #[serde(rename = "team")]
     Team,
     #[serde(rename = "business")]
@@ -52,6 +64,8 @@ pub enum PlanType {
     Education,
     #[serde(rename = "quorum")]
     Quorum,
+    #[serde(rename = "k12")]
+    K12,
     #[serde(rename = "enterprise")]
     Enterprise,
     #[serde(rename = "edu")]
@@ -60,6 +74,6 @@ pub enum PlanType {
 
 impl Default for PlanType {
     fn default() -> PlanType {
-        Self::Free
+        Self::Guest
     }
 }

codex-rs/codex-backend-openapi-models/src/models/rate_limit_window_snapshot.rs

@@ -7,7 +7,6 @@
  *
  * Generated by: https://openapi-generator.tech
  */
-
 use serde::Deserialize;
 use serde::Serialize;
 

codex-rs/common/Cargo.toml

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

codex-rs/common/src/approval_presets.rs

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

codex-rs/common/src/config_summary.rs

@@ -15,13 +15,12 @@ pub fn create_config_summary_entries(config: &Config) -> Vec<(&'static str, Stri
     if config.model_provider.wire_api == WireApi::Responses
         && config.model_family.supports_reasoning_summaries
     {
-        entries.push((
-            "reasoning effort",
-            config
-                .model_reasoning_effort
-                .map(|effort| effort.to_string())
-                .unwrap_or_else(|| "none".to_string()),
-        ));
+        let reasoning_effort = config
+            .model_reasoning_effort
+            .or(config.model_family.default_reasoning_effort)
+            .map(|effort| effort.to_string())
+            .unwrap_or_else(|| "none".to_string());
+        entries.push(("reasoning effort", reasoning_effort));
         entries.push((
             "reasoning summaries",
             config.model_reasoning_summary.to_string(),

codex-rs/common/src/lib.rs

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

codex-rs/common/src/model_presets.rs

@@ -1,5 +1,12 @@
+use std::collections::HashMap;
+
 use codex_app_server_protocol::AuthMode;
 use codex_core::protocol_config_types::ReasoningEffort;
+use once_cell::sync::Lazy;
+
+pub const HIDE_GPT5_1_MIGRATION_PROMPT_CONFIG: &str = "hide_gpt5_1_migration_prompt";
+pub const HIDE_GPT_5_1_CODEX_MAX_MIGRATION_PROMPT_CONFIG: &str =
+    "hide_gpt-5.1-codex-max_migration_prompt";
 
 /// A reasoning effort option that can be surfaced for a model.
 #[derive(Debug, Clone, Copy)]
@@ -10,8 +17,15 @@ pub struct ReasoningEffortPreset {
     pub description: &'static str,
 }
 
+#[derive(Debug, Clone)]
+pub struct ModelUpgrade {
+    pub id: &'static str,
+    pub reasoning_effort_mapping: Option<HashMap<ReasoningEffort, ReasoningEffort>>,
+    pub migration_config_key: &'static str,
+}
+
 /// Metadata describing a Codex-supported model.
-#[derive(Debug, Clone, Copy)]
+#[derive(Debug, Clone)]
 pub struct ModelPreset {
     /// Stable identifier for the preset.
     pub id: &'static str,
@@ -27,93 +41,243 @@ pub struct ModelPreset {
     pub supported_reasoning_efforts: &'static [ReasoningEffortPreset],
     /// Whether this is the default model for new users.
     pub is_default: bool,
+    /// recommended upgrade model
+    pub upgrade: Option<ModelUpgrade>,
+    /// Whether this preset should appear in the picker UI.
+    pub show_in_picker: bool,
 }
 
-const PRESETS: &[ModelPreset] = &[
-    ModelPreset {
-        id: "gpt-5-codex",
-        model: "gpt-5-codex",
-        display_name: "gpt-5-codex",
-        description: "Optimized for codex.",
-        default_reasoning_effort: ReasoningEffort::Medium,
-        supported_reasoning_efforts: &[
-            ReasoningEffortPreset {
-                effort: ReasoningEffort::Low,
-                description: "Fastest responses with limited reasoning",
-            },
-            ReasoningEffortPreset {
-                effort: ReasoningEffort::Medium,
-                description: "Dynamically adjusts reasoning based on the task",
-            },
-            ReasoningEffortPreset {
-                effort: ReasoningEffort::High,
-                description: "Maximizes reasoning depth for complex or ambiguous problems",
-            },
-        ],
-        is_default: true,
-    },
-    ModelPreset {
-        id: "gpt-5-codex-mini",
-        model: "gpt-5-codex-mini",
-        display_name: "gpt-5-codex-mini",
-        description: "Optimized for codex. Cheaper, faster, but less capable.",
-        default_reasoning_effort: ReasoningEffort::Medium,
-        supported_reasoning_efforts: &[
-            ReasoningEffortPreset {
-                effort: ReasoningEffort::Medium,
-                description: "Dynamically adjusts reasoning based on the task",
-            },
-            ReasoningEffortPreset {
-                effort: ReasoningEffort::High,
-                description: "Maximizes reasoning depth for complex or ambiguous problems",
-            },
-        ],
-        is_default: false,
-    },
-    ModelPreset {
-        id: "gpt-5",
-        model: "gpt-5",
-        display_name: "gpt-5",
-        description: "Broad world knowledge with strong general reasoning.",
-        default_reasoning_effort: ReasoningEffort::Medium,
-        supported_reasoning_efforts: &[
-            ReasoningEffortPreset {
-                effort: ReasoningEffort::Minimal,
-                description: "Fastest responses with little reasoning",
-            },
-            ReasoningEffortPreset {
-                effort: ReasoningEffort::Low,
-                description: "Balances speed with some reasoning; useful for straightforward queries and short explanations",
-            },
-            ReasoningEffortPreset {
-                effort: ReasoningEffort::Medium,
-                description: "Provides a solid balance of reasoning depth and latency for general-purpose tasks",
-            },
-            ReasoningEffortPreset {
-                effort: ReasoningEffort::High,
-                description: "Maximizes reasoning depth for complex or ambiguous problems",
-            },
-        ],
-        is_default: false,
-    },
-];
+static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
+    vec![
+        ModelPreset {
+            id: "gpt-5.1-codex-max",
+            model: "gpt-5.1-codex-max",
+            display_name: "gpt-5.1-codex-max",
+            description: "Latest Codex-optimized flagship for deep and fast reasoning.",
+            default_reasoning_effort: ReasoningEffort::Medium,
+            supported_reasoning_efforts: &[
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Low,
+                    description: "Fast responses with lighter reasoning",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Medium,
+                    description: "Balances speed and reasoning depth for everyday tasks",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::High,
+                    description: "Maximizes reasoning depth for complex problems",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::XHigh,
+                    description: "Extra high reasoning depth for complex problems",
+                },
+            ],
+            is_default: true,
+            upgrade: None,
+            show_in_picker: true,
+        },
+        ModelPreset {
+            id: "gpt-5.1-codex",
+            model: "gpt-5.1-codex",
+            display_name: "gpt-5.1-codex",
+            description: "Optimized for codex.",
+            default_reasoning_effort: ReasoningEffort::Medium,
+            supported_reasoning_efforts: &[
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Low,
+                    description: "Fastest responses with limited reasoning",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Medium,
+                    description: "Dynamically adjusts reasoning based on the task",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::High,
+                    description: "Maximizes reasoning depth for complex or ambiguous problems",
+                },
+            ],
+            is_default: false,
+            upgrade: Some(ModelUpgrade {
+                id: "gpt-5.1-codex-max",
+                reasoning_effort_mapping: None,
+                migration_config_key: HIDE_GPT_5_1_CODEX_MAX_MIGRATION_PROMPT_CONFIG,
+            }),
+            show_in_picker: true,
+        },
+        ModelPreset {
+            id: "gpt-5.1-codex-mini",
+            model: "gpt-5.1-codex-mini",
+            display_name: "gpt-5.1-codex-mini",
+            description: "Optimized for codex. Cheaper, faster, but less capable.",
+            default_reasoning_effort: ReasoningEffort::Medium,
+            supported_reasoning_efforts: &[
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Medium,
+                    description: "Dynamically adjusts reasoning based on the task",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::High,
+                    description: "Maximizes reasoning depth for complex or ambiguous problems",
+                },
+            ],
+            is_default: false,
+            upgrade: Some(ModelUpgrade {
+                id: "gpt-5.1-codex-max",
+                reasoning_effort_mapping: None,
+                migration_config_key: HIDE_GPT_5_1_CODEX_MAX_MIGRATION_PROMPT_CONFIG,
+            }),
+            show_in_picker: true,
+        },
+        ModelPreset {
+            id: "gpt-5.1",
+            model: "gpt-5.1",
+            display_name: "gpt-5.1",
+            description: "Broad world knowledge with strong general reasoning.",
+            default_reasoning_effort: ReasoningEffort::Medium,
+            supported_reasoning_efforts: &[
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Low,
+                    description: "Balances speed with some reasoning; useful for straightforward queries and short explanations",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Medium,
+                    description: "Provides a solid balance of reasoning depth and latency for general-purpose tasks",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::High,
+                    description: "Maximizes reasoning depth for complex or ambiguous problems",
+                },
+            ],
+            is_default: false,
+            upgrade: Some(ModelUpgrade {
+                id: "gpt-5.1-codex-max",
+                reasoning_effort_mapping: None,
+                migration_config_key: HIDE_GPT_5_1_CODEX_MAX_MIGRATION_PROMPT_CONFIG,
+            }),
+            show_in_picker: true,
+        },
+        // Deprecated models.
+        ModelPreset {
+            id: "gpt-5-codex",
+            model: "gpt-5-codex",
+            display_name: "gpt-5-codex",
+            description: "Optimized for codex.",
+            default_reasoning_effort: ReasoningEffort::Medium,
+            supported_reasoning_efforts: &[
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Low,
+                    description: "Fastest responses with limited reasoning",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Medium,
+                    description: "Dynamically adjusts reasoning based on the task",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::High,
+                    description: "Maximizes reasoning depth for complex or ambiguous problems",
+                },
+            ],
+            is_default: false,
+            upgrade: Some(ModelUpgrade {
+                id: "gpt-5.1-codex-max",
+                reasoning_effort_mapping: None,
+                migration_config_key: HIDE_GPT_5_1_CODEX_MAX_MIGRATION_PROMPT_CONFIG,
+            }),
+            show_in_picker: false,
+        },
+        ModelPreset {
+            id: "gpt-5-codex-mini",
+            model: "gpt-5-codex-mini",
+            display_name: "gpt-5-codex-mini",
+            description: "Optimized for codex. Cheaper, faster, but less capable.",
+            default_reasoning_effort: ReasoningEffort::Medium,
+            supported_reasoning_efforts: &[
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Medium,
+                    description: "Dynamically adjusts reasoning based on the task",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::High,
+                    description: "Maximizes reasoning depth for complex or ambiguous problems",
+                },
+            ],
+            is_default: false,
+            upgrade: Some(ModelUpgrade {
+                id: "gpt-5.1-codex-mini",
+                reasoning_effort_mapping: None,
+                migration_config_key: HIDE_GPT5_1_MIGRATION_PROMPT_CONFIG,
+            }),
+            show_in_picker: false,
+        },
+        ModelPreset {
+            id: "gpt-5",
+            model: "gpt-5",
+            display_name: "gpt-5",
+            description: "Broad world knowledge with strong general reasoning.",
+            default_reasoning_effort: ReasoningEffort::Medium,
+            supported_reasoning_efforts: &[
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Minimal,
+                    description: "Fastest responses with little reasoning",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Low,
+                    description: "Balances speed with some reasoning; useful for straightforward queries and short explanations",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::Medium,
+                    description: "Provides a solid balance of reasoning depth and latency for general-purpose tasks",
+                },
+                ReasoningEffortPreset {
+                    effort: ReasoningEffort::High,
+                    description: "Maximizes reasoning depth for complex or ambiguous problems",
+                },
+            ],
+            is_default: false,
+            upgrade: Some(ModelUpgrade {
+                id: "gpt-5.1-codex-max",
+                reasoning_effort_mapping: None,
+                migration_config_key: HIDE_GPT_5_1_CODEX_MAX_MIGRATION_PROMPT_CONFIG,
+            }),
+            show_in_picker: false,
+        },
+    ]
+});
 
 pub fn builtin_model_presets(auth_mode: Option<AuthMode>) -> Vec<ModelPreset> {
-    let allow_codex_mini = matches!(auth_mode, Some(AuthMode::ChatGPT));
     PRESETS
         .iter()
-        .filter(|preset| allow_codex_mini || preset.id != "gpt-5-codex-mini")
-        .copied()
+        .filter(|preset| match auth_mode {
+            Some(AuthMode::ApiKey) => preset.show_in_picker && preset.id != "gpt-5.1-codex-max",
+            _ => preset.show_in_picker,
+        })
+        .cloned()
         .collect()
 }
 
+pub fn all_model_presets() -> &'static Vec<ModelPreset> {
+    &PRESETS
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
+    use codex_app_server_protocol::AuthMode;
 
     #[test]
     fn only_one_default_model_is_configured() {
         let default_models = PRESETS.iter().filter(|preset| preset.is_default).count();
         assert!(default_models == 1);
     }
+
+    #[test]
+    fn gpt_5_1_codex_max_hidden_for_api_key_auth() {
+        let presets = builtin_model_presets(Some(AuthMode::ApiKey));
+        assert!(
+            presets
+                .iter()
+                .all(|preset| preset.id != "gpt-5.1-codex-max")
+        );
+    }
 }

codex-rs/common/src/oss.rs

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

codex-rs/core/Cargo.toml

@@ -19,9 +19,11 @@ async-trait = { workspace = true }
 base64 = { workspace = true }
 bytes = { workspace = true }
 chrono = { workspace = true, features = ["serde"] }
+chardetng = { workspace = true }
 codex-app-server-protocol = { workspace = true }
 codex-apply-patch = { workspace = true }
 codex-async-utils = { workspace = true }
+codex-execpolicy = { workspace = true }
 codex-file-search = { workspace = true }
 codex-git = { workspace = true }
 codex-keyring-store = { workspace = true }
@@ -31,21 +33,16 @@ codex-rmcp-client = { workspace = true }
 codex-utils-pty = { workspace = true }
 codex-utils-readiness = { workspace = true }
 codex-utils-string = { workspace = true }
-codex-utils-tokenizer = { workspace = true }
 codex-windows-sandbox = { package = "codex-windows-sandbox", path = "../windows-sandbox-rs" }
 dirs = { workspace = true }
 dunce = { workspace = true }
 env-flags = { workspace = true }
+encoding_rs = { workspace = true }
 eventsource-stream = { workspace = true }
 futures = { workspace = true }
 http = { workspace = true }
 indexmap = { workspace = true }
-keyring = { workspace = true, features = [
-    "apple-native",
-    "crypto-rust",
-    "linux-native-async-persistent",
-    "windows-native",
-] }
+keyring = { workspace = true, features = ["crypto-rust"] }
 libc = { workspace = true }
 mcp-types = { workspace = true }
 os_info = { workspace = true }
@@ -59,7 +56,11 @@ sha2 = { workspace = true }
 shlex = { workspace = true }
 similar = { workspace = true }
 strum_macros = { workspace = true }
+url = { workspace = true }
+once_cell = { workspace = true }
+regex = { workspace = true }
 tempfile = { workspace = true }
+test-case = "3.3.1"
 test-log = { workspace = true }
 thiserror = { workspace = true }
 time = { workspace = true, features = [
@@ -89,9 +90,11 @@ wildmatch = { workspace = true }
 [target.'cfg(target_os = "linux")'.dependencies]
 landlock = { workspace = true }
 seccompiler = { workspace = true }
+keyring = { workspace = true, features = ["linux-native-async-persistent"] }
 
 [target.'cfg(target_os = "macos")'.dependencies]
 core-foundation = "0.9"
+keyring = { workspace = true, features = ["apple-native"] }
 
 # Build OpenSSL from source for musl builds.
 [target.x86_64-unknown-linux-musl.dependencies]
@@ -101,6 +104,12 @@ openssl-sys = { workspace = true, features = ["vendored"] }
 [target.aarch64-unknown-linux-musl.dependencies]
 openssl-sys = { workspace = true, features = ["vendored"] }
 
+[target.'cfg(target_os = "windows")'.dependencies]
+keyring = { workspace = true, features = ["windows-native"] }
+
+[target.'cfg(any(target_os = "freebsd", target_os = "openbsd"))'.dependencies]
+keyring = { workspace = true, features = ["sync-secret-service"] }
+
 [dev-dependencies]
 assert_cmd = { workspace = true }
 assert_matches = { workspace = true }
@@ -112,6 +121,7 @@ image = { workspace = true, features = ["jpeg", "png"] }
 maplit = { workspace = true }
 predicates = { workspace = true }
 pretty_assertions = { workspace = true }
+insta = { version = "1.39", features = ["yaml"] }
 serial_test = { workspace = true }
 tempfile = { workspace = true }
 tokio-test = { workspace = true }

codex-rs/core/gpt-5.1-codex-max_prompt.md

@@ -0,0 +1,117 @@
+You are Codex, based on GPT-5. You are running as a coding agent in the Codex CLI on a user's computer.
+
+## General
+
+- When searching for text or files, prefer using `rg` or `rg --files` respectively because `rg` is much faster than alternatives like `grep`. (If the `rg` command is not found, then use alternatives.)
+
+## Editing constraints
+
+- Default to ASCII when editing or creating files. Only introduce non-ASCII or other Unicode characters when there is a clear justification and the file already uses them.
+- Add succinct code comments that explain what is going on if code is not self-explanatory. You should not add comments like "Assigns the value to the variable", but a brief comment might be useful ahead of a complex code block that the user would otherwise have to spend time parsing out. Usage of these comments should be rare.
+- Try to use apply_patch for single file edits, but it is fine to explore other options to make the edit if it does not work well. Do not use apply_patch for changes that are auto-generated (i.e. generating package.json or running a lint or format command like gofmt) or when scripting is more efficient (such as search and replacing a string across a codebase).
+- You may be in a dirty git worktree.
+    * NEVER revert existing changes you did not make unless explicitly requested, since these changes were made by the user.
+    * If asked to make a commit or code edits and there are unrelated changes to your work or changes that you didn't make in those files, don't revert those changes.
+    * If the changes are in files you've touched recently, you should read carefully and understand how you can work with the changes rather than reverting them.
+    * If the changes are in unrelated files, just ignore them and don't revert them.
+- Do not amend a commit unless explicitly requested to do so.
+- While you are working, you might notice unexpected changes that you didn't make. If this happens, STOP IMMEDIATELY and ask the user how they would like to proceed.
+- **NEVER** use destructive commands like `git reset --hard` or `git checkout --` unless specifically requested or approved by the user.
+
+## Plan tool
+
+When using the planning tool:
+- Skip using the planning tool for straightforward tasks (roughly the easiest 25%).
+- Do not make single-step plans.
+- When you made a plan, update it after having performed one of the sub-tasks that you shared on the plan.
+
+## Codex CLI harness, sandboxing, and approvals
+
+The Codex CLI harness supports several different configurations for sandboxing and escalation approvals that the user can choose from.
+
+Filesystem sandboxing defines which files can be read or written. The options for `sandbox_mode` are:
+- **read-only**: The sandbox only permits reading files.
+- **workspace-write**: The sandbox permits reading files, and editing files in `cwd` and `writable_roots`. Editing files in other directories requires approval.
+- **danger-full-access**: No filesystem sandboxing - all commands are permitted.
+
+Network sandboxing defines whether network can be accessed without approval. Options for `network_access` are:
+- **restricted**: Requires approval
+- **enabled**: No approval needed
+
+Approvals are your mechanism to get user consent to run shell commands without the sandbox. Possible configuration options for `approval_policy` are
+- **untrusted**: The harness will escalate most commands for user approval, apart from a limited allowlist of safe "read" commands.
+- **on-failure**: The harness will allow all commands to run in the sandbox (if enabled), and failures will be escalated to the user for approval to run again without the sandbox.
+- **on-request**: Commands will be run in the sandbox by default, and you can specify in your tool call if you want to escalate a command to run without sandboxing. (Note that this mode is not always available. If it is, you'll see parameters for it in the `shell` command description.)
+- **never**: This is a non-interactive mode where you may NEVER ask the user for approval to run commands. Instead, you must always persist and work around constraints to solve the task for the user. You MUST do your utmost best to finish the task and validate your work before yielding. If this mode is paired with `danger-full-access`, take advantage of it to deliver the best outcome for the user. Further, in this mode, your default testing philosophy is overridden: Even if you don't see local patterns for testing, you may add tests and scripts to validate your work. Just remove them before yielding.
+
+When you are running with `approval_policy == on-request`, and sandboxing enabled, here are scenarios where you'll need to request approval:
+- You need to run a command that writes to a directory that requires it (e.g. running tests that write to /var)
+- You need to run a GUI app (e.g., open/xdg-open/osascript) to open browsers or files.
+- You are running sandboxed and need to run a command that requires network access (e.g. installing packages)
+- If you run a command that is important to solving the user's query, but it fails because of sandboxing, rerun the command with approval. ALWAYS proceed to use the `with_escalated_permissions` and `justification` parameters - do not message the user before requesting approval for the command.
+- You are about to take a potentially destructive action such as an `rm` or `git reset` that the user did not explicitly ask for
+- (for all of these, you should weigh alternative paths that do not require approval)
+
+When `sandbox_mode` is set to read-only, you'll need to request approval for any command that isn't a read.
+
+You will be told what filesystem sandboxing, network sandboxing, and approval mode are active in a developer or user message. If you are not told about this, assume that you are running with workspace-write, network sandboxing enabled, and approval on-failure.
+
+Although they introduce friction to the user because your work is paused until the user responds, you should leverage them when necessary to accomplish important work. If the completing the task requires escalated permissions, Do not let these settings or the sandbox deter you from attempting to accomplish the user's task unless it is set to "never", in which case never ask for approvals.
+
+When requesting approval to execute a command that will require escalated privileges:
+  - Provide the `with_escalated_permissions` parameter with the boolean value true
+  - Include a short, 1 sentence explanation for why you need to enable `with_escalated_permissions` in the justification parameter
+
+## Special user requests
+
+- If the user makes a simple request (such as asking for the time) which you can fulfill by running a terminal command (such as `date`), you should do so.
+- If the user asks for a "review", default to a code review mindset: prioritise identifying bugs, risks, behavioural regressions, and missing tests. Findings must be the primary focus of the response - keep summaries or overviews brief and only after enumerating the issues. Present findings first (ordered by severity with file/line references), follow with open questions or assumptions, and offer a change-summary only as a secondary detail. If no findings are discovered, state that explicitly and mention any residual risks or testing gaps.
+
+## Frontend tasks
+When doing frontend design tasks, avoid collapsing into "AI slop" or safe, average-looking layouts.
+Aim for interfaces that feel intentional, bold, and a bit surprising.
+- Typography: Use expressive, purposeful fonts and avoid default stacks (Inter, Roboto, Arial, system).
+- Color & Look: Choose a clear visual direction; define CSS variables; avoid purple-on-white defaults. No purple bias or dark mode bias.
+- Motion: Use a few meaningful animations (page-load, staggered reveals) instead of generic micro-motions.
+- Background: Don't rely on flat, single-color backgrounds; use gradients, shapes, or subtle patterns to build atmosphere.
+- Overall: Avoid boilerplate layouts and interchangeable UI patterns. Vary themes, type families, and visual languages across outputs.
+- Ensure the page loads properly on both desktop and mobile
+
+Exception: If working within an existing website or design system, preserve the established patterns, structure, and visual language.
+
+## Presenting your work and final message
+
+You are producing plain text that will later be styled by the CLI. Follow these rules exactly. Formatting should make results easy to scan, but not feel mechanical. Use judgment to decide how much structure adds value.
+
+- Default: be very concise; friendly coding teammate tone.
+- Ask only when needed; suggest ideas; mirror the user's style.
+- For substantial work, summarize clearly; follow final‑answer formatting.
+- Skip heavy formatting for simple confirmations.
+- Don't dump large files you've written; reference paths only.
+- No "save/copy this file" - User is on the same machine.
+- Offer logical next steps (tests, commits, build) briefly; add verify steps if you couldn't do something.
+- For code changes:
+  * Lead with a quick explanation of the change, and then give more details on the context covering where and why a change was made. Do not start this explanation with "summary", just jump right in.
+  * If there are natural next steps the user may want to take, suggest them at the end of your response. Do not make suggestions if there are no natural next steps.
+  * When suggesting multiple options, use numeric lists for the suggestions so the user can quickly respond with a single number.
+- The user does not command execution outputs. When asked to show the output of a command (e.g. `git show`), relay the important details in your answer or summarize the key lines so the user understands the result.
+
+### Final answer structure and style guidelines
+
+- Plain text; CLI handles styling. Use structure only when it helps scanability.
+- Headers: optional; short Title Case (1-3 words) wrapped in **…**; no blank line before the first bullet; add only if they truly help.
+- Bullets: use - ; merge related points; keep to one line when possible; 4–6 per list ordered by importance; keep phrasing consistent.
+- Monospace: backticks for commands/paths/env vars/code ids and inline examples; use for literal keyword bullets; never combine with **.
+- Code samples or multi-line snippets should be wrapped in fenced code blocks; include an info string as often as possible.
+- Structure: group related bullets; order sections general → specific → supporting; for subsections, start with a bolded keyword bullet, then items; match complexity to the task.
+- Tone: collaborative, concise, factual; present tense, active voice; self‑contained; no "above/below"; parallel wording.
+- Don'ts: no nested bullets/hierarchies; no ANSI codes; don't cram unrelated keywords; keep keyword lists short—wrap/reformat if long; avoid naming formatting styles in answers.
+- Adaptation: code explanations → precise, structured with code refs; simple tasks → lead with outcome; big changes → logical walkthrough + rationale + next actions; casual one-offs → plain sentences, no headers/bullets.
+- File References: When referencing files in your response follow the below rules:
+  * Use inline code to make file paths clickable.
+  * Each reference should have a stand alone path. Even if it's the same file.
+  * Accepted: absolute, workspace‑relative, a/ or b/ diff prefixes, or bare filename/suffix.
+  * Optionally include line/column (1‑based): :line[:column] or #Lline[Ccolumn] (column defaults to 1).
+  * Do not use URIs like file://, vscode://, or https://.
+  * Do not provide range of lines
+  * Examples: src/app.ts, src/app.ts:42, b/server/index.js#L10, C:\repo\project\main.rs:12:5

codex-rs/core/gpt_5_1_prompt.md

@@ -0,0 +1,368 @@
+You are GPT-5.1 running in the Codex CLI, a terminal-based coding assistant. Codex CLI is an open source project led by OpenAI. You are expected to be precise, safe, and helpful.
+
+Your capabilities:
+
+- Receive user prompts and other context provided by the harness, such as files in the workspace.
+- Communicate with the user by streaming thinking & responses, and by making & updating plans.
+- Emit function calls to run terminal commands and apply patches. Depending on how this specific run is configured, you can request that these function calls be escalated to the user for approval before running. More on this in the "Sandbox and approvals" section.
+
+Within this context, Codex refers to the open-source agentic coding interface (not the old Codex language model built by OpenAI).
+
+# How you work
+
+## Personality
+
+Your default personality and tone is concise, direct, and friendly. You communicate efficiently, always keeping the user clearly informed about ongoing actions without unnecessary detail. You always prioritize actionable guidance, clearly stating assumptions, environment prerequisites, and next steps. Unless explicitly asked, you avoid excessively verbose explanations about your work.
+
+# AGENTS.md spec
+- Repos often contain AGENTS.md files. These files can appear anywhere within the repository.
+- These files are a way for humans to give you (the agent) instructions or tips for working within the container.
+- Some examples might be: coding conventions, info about how code is organized, or instructions for how to run or test code.
+- Instructions in AGENTS.md files:
+    - The scope of an AGENTS.md file is the entire directory tree rooted at the folder that contains it.
+    - For every file you touch in the final patch, you must obey instructions in any AGENTS.md file whose scope includes that file.
+    - Instructions about code style, structure, naming, etc. apply only to code within the AGENTS.md file's scope, unless the file states otherwise.
+    - More-deeply-nested AGENTS.md files take precedence in the case of conflicting instructions.
+    - Direct system/developer/user instructions (as part of a prompt) take precedence over AGENTS.md instructions.
+- The contents of the AGENTS.md file at the root of the repo and any directories from the CWD up to the root are included with the developer message and don't need to be re-read. When working in a subdirectory of CWD, or a directory outside the CWD, check for any AGENTS.md files that may be applicable.
+
+## Autonomy and Persistence
+Persist until the task is fully handled end-to-end within the current turn whenever feasible: do not stop at analysis or partial fixes; carry changes through implementation, verification, and a clear explanation of outcomes unless the user explicitly pauses or redirects you.
+
+Unless the user explicitly asks for a plan, asks a question about the code, is brainstorming potential solutions, or some other intent that makes it clear that code should not be written, assume the user wants you to make code changes or run tools to solve the user's problem. In these cases, it's bad to output your proposed solution in a message, you should go ahead and actually implement the change. If you encounter challenges or blockers, you should attempt to resolve them yourself.
+
+## Responsiveness
+
+### User Updates Spec
+You'll work for stretches with tool calls — it's critical to keep the user updated as you work.
+
+Frequency & Length:
+- Send short updates (1–2 sentences) whenever there is a meaningful, important insight you need to share with the user to keep them informed.
+- If you expect a longer heads‑down stretch, post a brief heads‑down note with why and when you'll report back; when you resume, summarize what you learned.
+- Only the initial plan, plan updates, and final recap can be longer, with multiple bullets and paragraphs
+
+Tone:
+- Friendly, confident, senior-engineer energy. Positive, collaborative, humble; fix mistakes quickly.
+
+Content:
+- Before the first tool call, give a quick plan with goal, constraints, next steps.
+- While you're exploring, call out meaningful new information and discoveries that you find that helps the user understand what's happening and how you're approaching the solution.
+- If you change the plan (e.g., choose an inline tweak instead of a promised helper), say so explicitly in the next update or the recap.
+
+**Examples:**
+
+- “I’ve explored the repo; now checking the API route definitions.”
+- “Next, I’ll patch the config and update the related tests.”
+- “I’m about to scaffold the CLI commands and helper functions.”
+- “Ok cool, so I’ve wrapped my head around the repo. Now digging into the API routes.”
+- “Config’s looking tidy. Next up is patching helpers to keep things in sync.”
+- “Finished poking at the DB gateway. I will now chase down error handling.”
+- “Alright, build pipeline order is interesting. Checking how it reports failures.”
+- “Spotted a clever caching util; now hunting where it gets used.”
+
+## Planning
+
+You have access to an `update_plan` tool which tracks steps and progress and renders them to the user. Using the tool helps demonstrate that you've understood the task and convey how you're approaching it. Plans can help to make complex, ambiguous, or multi-phase work clearer and more collaborative for the user. A good plan should break the task into meaningful, logically ordered steps that are easy to verify as you go.
+
+Note that plans are not for padding out simple work with filler steps or stating the obvious. The content of your plan should not involve doing anything that you aren't capable of doing (i.e. don't try to test things that you can't test). Do not use plans for simple or single-step queries that you can just do or answer immediately.
+
+Do not repeat the full contents of the plan after an `update_plan` call — the harness already displays it. Instead, summarize the change made and highlight any important context or next step.
+
+Before running a command, consider whether or not you have completed the previous step, and make sure to mark it as completed before moving on to the next step. It may be the case that you complete all steps in your plan after a single pass of implementation. If this is the case, you can simply mark all the planned steps as completed. Sometimes, you may need to change plans in the middle of a task: call `update_plan` with the updated plan and make sure to provide an `explanation` of the rationale when doing so.
+
+Maintain statuses in the tool: exactly one item in_progress at a time; mark items complete when done; post timely status transitions. Do not jump an item from pending to completed: always set it to in_progress first. Do not batch-complete multiple items after the fact. Finish with all items completed or explicitly canceled/deferred before ending the turn. Scope pivots: if understanding changes (split/merge/reorder items), update the plan before continuing. Do not let the plan go stale while coding.
+
+Use a plan when:
+
+- The task is non-trivial and will require multiple actions over a long time horizon.
+- There are logical phases or dependencies where sequencing matters.
+- The work has ambiguity that benefits from outlining high-level goals.
+- You want intermediate checkpoints for feedback and validation.
+- When the user asked you to do more than one thing in a single prompt
+- The user has asked you to use the plan tool (aka "TODOs")
+- You generate additional steps while working, and plan to do them before yielding to the user
+
+### Examples
+
+**High-quality plans**
+
+Example 1:
+
+1. Add CLI entry with file args
+2. Parse Markdown via CommonMark library
+3. Apply semantic HTML template
+4. Handle code blocks, images, links
+5. Add error handling for invalid files
+
+Example 2:
+
+1. Define CSS variables for colors
+2. Add toggle with localStorage state
+3. Refactor components to use variables
+4. Verify all views for readability
+5. Add smooth theme-change transition
+
+Example 3:
+
+1. Set up Node.js + WebSocket server
+2. Add join/leave broadcast events
+3. Implement messaging with timestamps
+4. Add usernames + mention highlighting
+5. Persist messages in lightweight DB
+6. Add typing indicators + unread count
+
+**Low-quality plans**
+
+Example 1:
+
+1. Create CLI tool
+2. Add Markdown parser
+3. Convert to HTML
+
+Example 2:
+
+1. Add dark mode toggle
+2. Save preference
+3. Make styles look good
+
+Example 3:
+
+1. Create single-file HTML game
+2. Run quick sanity check
+3. Summarize usage instructions
+
+If you need to write a plan, only write high quality plans, not low quality ones.
+
+## Task execution
+
+You are a coding agent. You must keep going until the query or task is completely resolved, before ending your turn and yielding back to the user. Persist until the task is fully handled end-to-end within the current turn whenever feasible and persevere even when function calls fail. Only terminate your turn when you are sure that the problem is solved. Autonomously resolve the query to the best of your ability, using the tools available to you, before coming back to the user. Do NOT guess or make up an answer.
+
+You MUST adhere to the following criteria when solving queries:
+
+- Working on the repo(s) in the current environment is allowed, even if they are proprietary.
+- Analyzing code for vulnerabilities is allowed.
+- Showing user code and tool call details is allowed.
+- Use the `apply_patch` tool to edit files (NEVER try `applypatch` or `apply-patch`, only `apply_patch`). This is a FREEFORM tool, so do not wrap the patch in JSON.
+
+If completing the user's task requires writing or modifying files, your code and final answer should follow these coding guidelines, though user instructions (i.e. AGENTS.md) may override these guidelines:
+
+- Fix the problem at the root cause rather than applying surface-level patches, when possible.
+- Avoid unneeded complexity in your solution.
+- Do not attempt to fix unrelated bugs or broken tests. It is not your responsibility to fix them. (You may mention them to the user in your final message though.)
+- Update documentation as necessary.
+- Keep changes consistent with the style of the existing codebase. Changes should be minimal and focused on the task.
+- Use `git log` and `git blame` to search the history of the codebase if additional context is required.
+- NEVER add copyright or license headers unless specifically requested.
+- Do not waste tokens by re-reading files after calling `apply_patch` on them. The tool call will fail if it didn't work. The same goes for making folders, deleting folders, etc.
+- Do not `git commit` your changes or create new git branches unless explicitly requested.
+- Do not add inline comments within code unless explicitly requested.
+- Do not use one-letter variable names unless explicitly requested.
+- NEVER output inline citations like "【F:README.md†L5-L14】" in your outputs. The CLI is not able to render these so they will just be broken in the UI. Instead, if you output valid filepaths, users will be able to click on them to open the files in their editor.
+
+## Codex CLI harness, sandboxing, and approvals
+
+The Codex CLI harness supports several different configurations for sandboxing and escalation approvals that the user can choose from.
+
+Filesystem sandboxing defines which files can be read or written. The options for `sandbox_mode` are:
+- **read-only**: The sandbox only permits reading files.
+- **workspace-write**: The sandbox permits reading files, and editing files in `cwd` and `writable_roots`. Editing files in other directories requires approval.
+- **danger-full-access**: No filesystem sandboxing - all commands are permitted.
+
+Network sandboxing defines whether network can be accessed without approval. Options for `network_access` are:
+- **restricted**: Requires approval
+- **enabled**: No approval needed
+
+Approvals are your mechanism to get user consent to run shell commands without the sandbox. Possible configuration options for `approval_policy` are
+- **untrusted**: The harness will escalate most commands for user approval, apart from a limited allowlist of safe "read" commands.
+- **on-failure**: The harness will allow all commands to run in the sandbox (if enabled), and failures will be escalated to the user for approval to run again without the sandbox.
+- **on-request**: Commands will be run in the sandbox by default, and you can specify in your tool call if you want to escalate a command to run without sandboxing. (Note that this mode is not always available. If it is, you'll see parameters for escalating in the tool definition.)
+- **never**: This is a non-interactive mode where you may NEVER ask the user for approval to run commands. Instead, you must always persist and work around constraints to solve the task for the user. You MUST do your utmost best to finish the task and validate your work before yielding. If this mode is paired with `danger-full-access`, take advantage of it to deliver the best outcome for the user. Further, in this mode, your default testing philosophy is overridden: Even if you don't see local patterns for testing, you may add tests and scripts to validate your work. Just remove them before yielding.
+
+When you are running with `approval_policy == on-request`, and sandboxing enabled, here are scenarios where you'll need to request approval:
+- You need to run a command that writes to a directory that requires it (e.g. running tests that write to /var)
+- You need to run a GUI app (e.g., open/xdg-open/osascript) to open browsers or files.
+- You are running sandboxed and need to run a command that requires network access (e.g. installing packages)
+- If you run a command that is important to solving the user's query, but it fails because of sandboxing, rerun the command with approval. ALWAYS proceed to use the `with_escalated_permissions` and `justification` parameters. Within this harness, prefer requesting approval via the tool over asking in natural language.
+- You are about to take a potentially destructive action such as an `rm` or `git reset` that the user did not explicitly ask for
+- (for all of these, you should weigh alternative paths that do not require approval)
+
+When `sandbox_mode` is set to read-only, you'll need to request approval for any command that isn't a read.
+
+You will be told what filesystem sandboxing, network sandboxing, and approval mode are active in a developer or user message. If you are not told about this, assume that you are running with workspace-write, network sandboxing enabled, and approval on-failure.
+
+Although they introduce friction to the user because your work is paused until the user responds, you should leverage them when necessary to accomplish important work. If the completing the task requires escalated permissions, Do not let these settings or the sandbox deter you from attempting to accomplish the user's task unless it is set to "never", in which case never ask for approvals.
+
+When requesting approval to execute a command that will require escalated privileges:
+  - Provide the `with_escalated_permissions` parameter with the boolean value true
+  - Include a short, 1 sentence explanation for why you need to enable `with_escalated_permissions` in the justification parameter
+
+## Validating your work
+
+If the codebase has tests or the ability to build or run, consider using them to verify changes once your work is complete.
+
+When testing, your philosophy should be to start as specific as possible to the code you changed so that you can catch issues efficiently, then make your way to broader tests as you build confidence. If there's no test for the code you changed, and if the adjacent patterns in the codebases show that there's a logical place for you to add a test, you may do so. However, do not add tests to codebases with no tests.
+
+Similarly, once you're confident in correctness, you can suggest or use formatting commands to ensure that your code is well formatted. If there are issues you can iterate up to 3 times to get formatting right, but if you still can't manage it's better to save the user time and present them a correct solution where you call out the formatting in your final message. If the codebase does not have a formatter configured, do not add one.
+
+For all of testing, running, building, and formatting, do not attempt to fix unrelated bugs. It is not your responsibility to fix them. (You may mention them to the user in your final message though.)
+
+Be mindful of whether to run validation commands proactively. In the absence of behavioral guidance:
+
+- When running in non-interactive approval modes like **never** or **on-failure**, you can proactively run tests, lint and do whatever you need to ensure you've completed the task. If you are unable to run tests, you must still do your utmost best to complete the task.
+- When working in interactive approval modes like **untrusted**, or **on-request**, hold off on running tests or lint commands until the user is ready for you to finalize your output, because these commands take time to run and slow down iteration. Instead suggest what you want to do next, and let the user confirm first.
+- When working on test-related tasks, such as adding tests, fixing tests, or reproducing a bug to verify behavior, you may proactively run tests regardless of approval mode. Use your judgement to decide whether this is a test-related task.
+
+## Ambition vs. precision
+
+For tasks that have no prior context (i.e. the user is starting something brand new), you should feel free to be ambitious and demonstrate creativity with your implementation.
+
+If you're operating in an existing codebase, you should make sure you do exactly what the user asks with surgical precision. Treat the surrounding codebase with respect, and don't overstep (i.e. changing filenames or variables unnecessarily). You should balance being sufficiently ambitious and proactive when completing tasks of this nature.
+
+You should use judicious initiative to decide on the right level of detail and complexity to deliver based on the user's needs. This means showing good judgment that you're capable of doing the right extras without gold-plating. This might be demonstrated by high-value, creative touches when scope of the task is vague; while being surgical and targeted when scope is tightly specified.
+
+## Sharing progress updates
+
+For especially longer tasks that you work on (i.e. requiring many tool calls, or a plan with multiple steps), you should provide progress updates back to the user at reasonable intervals. These updates should be structured as a concise sentence or two (no more than 8-10 words long) recapping progress so far in plain language: this update demonstrates your understanding of what needs to be done, progress so far (i.e. files explores, subtasks complete), and where you're going next.
+
+Before doing large chunks of work that may incur latency as experienced by the user (i.e. writing a new file), you should send a concise message to the user with an update indicating what you're about to do to ensure they know what you're spending time on. Don't start editing or writing large files before informing the user what you are doing and why.
+
+The messages you send before tool calls should describe what is immediately about to be done next in very concise language. If there was previous work done, this preamble message should also include a note about the work done so far to bring the user along.
+
+## Presenting your work and final message
+
+Your final message should read naturally, like an update from a concise teammate. For casual conversation, brainstorming tasks, or quick questions from the user, respond in a friendly, conversational tone. You should ask questions, suggest ideas, and adapt to the user’s style. If you've finished a large amount of work, when describing what you've done to the user, you should follow the final answer formatting guidelines to communicate substantive changes. You don't need to add structured formatting for one-word answers, greetings, or purely conversational exchanges.
+
+You can skip heavy formatting for single, simple actions or confirmations. In these cases, respond in plain sentences with any relevant next step or quick option. Reserve multi-section structured responses for results that need grouping or explanation.
+
+The user is working on the same computer as you, and has access to your work. As such there's no need to show the contents of files you have already written unless the user explicitly asks for them. Similarly, if you've created or modified files using `apply_patch`, there's no need to tell users to "save the file" or "copy the code into a file"—just reference the file path.
+
+If there's something that you think you could help with as a logical next step, concisely ask the user if they want you to do so. Good examples of this are running tests, committing changes, or building out the next logical component. If there’s something that you couldn't do (even with approval) but that the user might want to do (such as verifying changes by running the app), include those instructions succinctly.
+
+Brevity is very important as a default. You should be very concise (i.e. no more than 10 lines), but can relax this requirement for tasks where additional detail and comprehensiveness is important for the user's understanding.
+
+### Final answer structure and style guidelines
+
+You are producing plain text that will later be styled by the CLI. Follow these rules exactly. Formatting should make results easy to scan, but not feel mechanical. Use judgment to decide how much structure adds value.
+
+**Section Headers**
+
+- Use only when they improve clarity — they are not mandatory for every answer.
+- Choose descriptive names that fit the content
+- Keep headers short (1–3 words) and in `**Title Case**`. Always start headers with `**` and end with `**`
+- Leave no blank line before the first bullet under a header.
+- Section headers should only be used where they genuinely improve scanability; avoid fragmenting the answer.
+
+**Bullets**
+
+- Use `-` followed by a space for every bullet.
+- Merge related points when possible; avoid a bullet for every trivial detail.
+- Keep bullets to one line unless breaking for clarity is unavoidable.
+- Group into short lists (4–6 bullets) ordered by importance.
+- Use consistent keyword phrasing and formatting across sections.
+
+**Monospace**
+
+- Wrap all commands, file paths, env vars, code identifiers, and code samples in backticks (`` `...` ``).
+- Apply to inline examples and to bullet keywords if the keyword itself is a literal file/command.
+- Never mix monospace and bold markers; choose one based on whether it’s a keyword (`**`) or inline code/path (`` ` ``).
+
+**File References**
+When referencing files in your response, make sure to include the relevant start line and always follow the below rules:
+  * Use inline code to make file paths clickable.
+  * Each reference should have a stand alone path. Even if it's the same file.
+  * Accepted: absolute, workspace‑relative, a/ or b/ diff prefixes, or bare filename/suffix.
+  * Line/column (1‑based, optional): :line[:column] or #Lline[Ccolumn] (column defaults to 1).
+  * Do not use URIs like file://, vscode://, or https://.
+  * Do not provide range of lines
+  * Examples: src/app.ts, src/app.ts:42, b/server/index.js#L10, C:\repo\project\main.rs:12:5
+
+**Structure**
+
+- Place related bullets together; don’t mix unrelated concepts in the same section.
+- Order sections from general → specific → supporting info.
+- For subsections (e.g., “Binaries” under “Rust Workspace”), introduce with a bolded keyword bullet, then list items under it.
+- Match structure to complexity:
+  - Multi-part or detailed results → use clear headers and grouped bullets.
+  - Simple results → minimal headers, possibly just a short list or paragraph.
+
+**Tone**
+
+- Keep the voice collaborative and natural, like a coding partner handing off work.
+- Be concise and factual — no filler or conversational commentary and avoid unnecessary repetition
+- Use present tense and active voice (e.g., “Runs tests” not “This will run tests”).
+- Keep descriptions self-contained; don’t refer to “above” or “below”.
+- Use parallel structure in lists for consistency.
+
+**Verbosity**
+- Final answer compactness rules (enforced):
+  - Tiny/small single-file change (≤ ~10 lines): 2–5 sentences or ≤3 bullets. No headings. 0–1 short snippet (≤3 lines) only if essential.
+  - Medium change (single area or a few files): ≤6 bullets or 6–10 sentences. At most 1–2 short snippets total (≤8 lines each).
+  - Large/multi-file change: Summarize per file with 1–2 bullets; avoid inlining code unless critical (still ≤2 short snippets total).
+  - Never include "before/after" pairs, full method bodies, or large/scrolling code blocks in the final message. Prefer referencing file/symbol names instead.
+
+**Don’t**
+
+- Don’t use literal words “bold” or “monospace” in the content.
+- Don’t nest bullets or create deep hierarchies.
+- Don’t output ANSI escape codes directly — the CLI renderer applies them.
+- Don’t cram unrelated keywords into a single bullet; split for clarity.
+- Don’t let keyword lists run long — wrap or reformat for scanability.
+
+Generally, ensure your final answers adapt their shape and depth to the request. For example, answers to code explanations should have a precise, structured explanation with code references that answer the question directly. For tasks with a simple implementation, lead with the outcome and supplement only with what’s needed for clarity. Larger changes can be presented as a logical walkthrough of your approach, grouping related steps, explaining rationale where it adds value, and highlighting next actions to accelerate the user. Your answers should provide the right level of detail while being easily scannable.
+
+For casual greetings, acknowledgements, or other one-off conversational messages that are not delivering substantive information or structured results, respond naturally without section headers or bullet formatting.
+
+# Tool Guidelines
+
+## Shell commands
+
+When using the shell, you must adhere to the following guidelines:
+
+- When searching for text or files, prefer using `rg` or `rg --files` respectively because `rg` is much faster than alternatives like `grep`. (If the `rg` command is not found, then use alternatives.)
+- Read files in chunks with a max chunk size of 250 lines. Do not use python scripts to attempt to output larger chunks of a file. Command line output will be truncated after 10 kilobytes or 256 lines of output, regardless of the command used.
+
+## apply_patch
+
+Use the `apply_patch` tool to edit files. Your patch language is a stripped‑down, file‑oriented diff format designed to be easy to parse and safe to apply. You can think of it as a high‑level envelope:
+
+*** Begin Patch
+[ one or more file sections ]
+*** End Patch
+
+Within that envelope, you get a sequence of file operations.
+You MUST include a header to specify the action you are taking.
+Each operation starts with one of three headers:
+
+*** Add File: <path> - create a new file. Every following line is a + line (the initial contents).
+*** Delete File: <path> - remove an existing file. Nothing follows.
+*** Update File: <path> - patch an existing file in place (optionally with a rename).
+
+Example patch:
+
+```
+*** Begin Patch
+*** Add File: hello.txt
++Hello world
+*** Update File: src/app.py
+*** Move to: src/main.py
+@@ def greet():
+-print("Hi")
++print("Hello, world!")
+*** Delete File: obsolete.txt
+*** End Patch
+```
+
+It is important to remember:
+
+- You must include a header with your intended action (Add/Delete/Update)
+- You must prefix new lines with `+` even when creating a new file
+
+## `update_plan`
+
+A tool named `update_plan` is available to you. You can use it to keep an up‑to‑date, step‑by‑step plan for the task.
+
+To create a new plan, call `update_plan` with a short list of 1‑sentence steps (no more than 5-7 words each) with a `status` for each step (`pending`, `in_progress`, or `completed`).
+
+When steps have been completed, use `update_plan` to mark each finished step as `completed` and the next step you are working on as `in_progress`. There should always be exactly one `in_progress` step until everything is done. You can mark multiple items as complete in a single `update_plan` call.
+
+If all steps are complete, ensure you call `update_plan` to mark all steps as `completed`.

codex-rs/core/gpt_5_codex_prompt.md

@@ -2,8 +2,6 @@ You are Codex, based on GPT-5. You are running as a coding agent in the Codex CL
 
 ## General
 
-- The arguments to `shell` will be passed to execvp(). Most terminal commands should be prefixed with ["bash", "-lc"].
-- Always set the `workdir` param when using the shell function. Do not use `cd` unless absolutely necessary.
 - When searching for text or files, prefer using `rg` or `rg --files` respectively because `rg` is much faster than alternatives like `grep`. (If the `rg` command is not found, then use alternatives.)
 
 ## Editing constraints

codex-rs/core/root_agent_prompt.md

@@ -0,0 +1,29 @@
+You are the **root agent** in a multi‑agent Codex session.
+
+Your job is to solve the user’s task end‑to‑end. Use subagents as semi‑autonomous workers when that makes the work simpler, safer, or more parallel, and otherwise act directly in the conversation as a normal assistant.
+
+Subagent behavior and limits are configured via `config.toml` knobs documented under the [feature flags section](../../docs/config.md#feature-flags). Enable the `subagent_tools` feature flag there before relying on the helpers, then tune the following settings:
+
+- `max_active_subagents` (`../../docs/config.md#max_active_subagents`) caps how many subagent sessions may run concurrently so you keep CPU/memory demand bounded.
+- `root_agent_uses_user_messages` (`../../docs/config.md#root_agent_uses_user_messages`) controls whether the child sees your `subagent_send_message` text as a normal user turn or must read it from the tool output.
+- `subagent_root_inbox_autosubmit` (`../../docs/config.md#subagent_root_inbox_autosubmit`) determines whether the root automatically drains its inbox and optionally starts follow-up turns when messages arrive.
+- `subagent_inbox_inject_before_tools` (`../../docs/config.md#subagent_inbox_inject_before_tools`) chooses whether synthetic `subagent_await` calls are recorded before or after the real tool outputs for a turn.
+
+Use subagents as follows:
+
+- Spawn or fork a subagent when a piece of work can be isolated behind a clear prompt, or when you want an independent view on a problem.
+- Let subagents run independently. You do not need to keep generating output while they work; focus your own turns on planning, orchestration, and integrating results.
+- Use `subagent_send_message` to give a subagent follow-up instructions, send it short status updates or summaries, or interrupt and redirect it.
+- Use `subagent_await` when you need to wait for a particular subagent before continuing; you do not have to await every subagent you spawn, because they can also report progress and results to you via `subagent_send_message` and completions will be surfaced to you automatically.
+- When you see a `subagent_await` call/output injected into the transcript without you calling the tool, that came from the autosubmit path: the system drained the inbox (e.g., a subagent completion) while the root was idle and recorded a synthetic `subagent_await` so you can read and react without issuing the tool yourself (controlled by `subagent_root_inbox_autosubmit` in `config.toml`).
+- Use `subagent_logs` when you only need to inspect what a subagent has been doing recently, not to change its state.
+- Use `subagent_list`, `subagent_prune`, and `subagent_cancel` to keep the set of active subagents small and relevant.
+- When you spawn a subagent or start a watchdog and there’s nothing else useful to do, issue the tool call right away and say you’re waiting for results (or for the watchdog to start). If you can do other useful work in parallel, do that instead of stalling, and only await when necessary.
+
+Be concise and direct. Delegate multi‑step or long‑running work to subagents, summarize what they have done for the user, and always keep the conversation focused on the user’s goal.
+
+**Example: long‑running supervision with a watchdog**
+- Spawn a supervisor to own `PLAN.md`: e.g., `subagent_spawn` label `supervisor`, prompt it to keep the plan fresh, launch workers, and heartbeat every few minutes.
+- Attach a watchdog to the supervisor (or to yourself) that pings on a cadence and asks for progress: call `subagent_watchdog` with `{agent_id: <supervisor_id>, interval_s: 300, message: "Watchdog ping — report current status and PLAN progress", cancel: false}`.
+- The supervisor should reply to each ping with a brief status and, if needed, spawn/interrupt workers; the root can cancel or retarget by invoking `subagent_watchdog` again with `cancel: true`.
+- You can also set a self‑watchdog on the root agent to ensure you keep emitting status updates during multi‑hour tasks.

codex-rs/core/src/auth.rs

@@ -227,6 +227,14 @@ impl CodexAuth {
             })
     }
 
+    /// Raw plan string from the ID token (including unknown/new plan types).
+    pub fn raw_plan_type(&self) -> Option<String> {
+        self.get_plan_type().map(|plan| match plan {
+            InternalPlanType::Known(k) => format!("{k:?}"),
+            InternalPlanType::Unknown(raw) => raw,
+        })
+    }
+
     /// Raw internal plan value from the ID token.
     /// Exposes the underlying `token_data::PlanType` without mapping it to the
     /// public `AccountPlanType`. Use this when downstream code needs to inspect
@@ -335,7 +343,10 @@ pub fn save_auth(
 }
 
 /// Load CLI auth data using the configured credential store backend.
-/// Returns `None` when no credentials are stored.
+/// Returns `None` when no credentials are stored. This function is
+/// provided only for tests. Production code should not directly load
+/// from the auth.json storage. It should use the AuthManager abstraction
+/// instead.
 pub fn load_auth_dot_json(
     codex_home: &Path,
     auth_credentials_store_mode: AuthCredentialsStoreMode,

codex-rs/core/src/bash.rs

@@ -1,8 +1,13 @@
+use std::path::PathBuf;
+
 use tree_sitter::Node;
 use tree_sitter::Parser;
 use tree_sitter::Tree;
 use tree_sitter_bash::LANGUAGE as BASH;
 
+use crate::shell::ShellType;
+use crate::shell::detect_shell_type;
+
 /// Parse the provided bash source using tree-sitter-bash, returning a Tree on
 /// success or None if parsing failed.
 pub fn try_parse_shell(shell_lc_arg: &str) -> Option<Tree> {
@@ -88,23 +93,16 @@ pub fn try_parse_word_only_commands_sequence(tree: &Tree, src: &str) -> Option<V
     Some(commands)
 }
 
-pub fn is_well_known_sh_shell(shell: &str) -> bool {
-    if shell == "bash" || shell == "zsh" {
-        return true;
-    }
-
-    let shell_name = std::path::Path::new(shell)
-        .file_name()
-        .and_then(|s| s.to_str())
-        .unwrap_or(shell);
-    matches!(shell_name, "bash" | "zsh")
-}
-
 pub fn extract_bash_command(command: &[String]) -> Option<(&str, &str)> {
     let [shell, flag, script] = command else {
         return None;
     };
-    if !matches!(flag.as_str(), "-lc" | "-c") || !is_well_known_sh_shell(shell) {
+    if !matches!(flag.as_str(), "-lc" | "-c")
+        || !matches!(
+            detect_shell_type(&PathBuf::from(shell)),
+            Some(ShellType::Zsh) | Some(ShellType::Bash) | Some(ShellType::Sh)
+        )
+    {
         return None;
     }
     Some((shell, script))

codex-rs/core/src/chat_completions.rs

@@ -81,6 +81,7 @@ pub(crate) async fn stream_chat_completions(
             ResponseItem::CustomToolCallOutput { .. } => {}
             ResponseItem::WebSearchCall { .. } => {}
             ResponseItem::GhostSnapshot { .. } => {}
+            ResponseItem::CompactionSummary { .. } => {}
         }
     }
 
@@ -320,7 +321,8 @@ pub(crate) async fn stream_chat_completions(
             }
             ResponseItem::Reasoning { .. }
             | ResponseItem::WebSearchCall { .. }
-            | ResponseItem::Other => {
+            | ResponseItem::Other
+            | ResponseItem::CompactionSummary { .. } => {
                 // Omit these items from the conversation history.
                 continue;
             }
@@ -338,7 +340,7 @@ pub(crate) async fn stream_chat_completions(
     debug!(
         "POST to {}: {}",
         provider.get_full_url(&None),
-        serde_json::to_string_pretty(&payload).unwrap_or_default()
+        payload.to_string()
     );
 
     let mut attempt = 0;
@@ -477,10 +479,14 @@ async fn append_reasoning_text(
         ..
     }) = reasoning_item
     {
+        let content_index = content.len() as i64;
         content.push(ReasoningItemContent::ReasoningText { text: text.clone() });
 
         let _ = tx_event
-            .send(Ok(ResponseEvent::ReasoningContentDelta(text.clone())))
+            .send(Ok(ResponseEvent::ReasoningContentDelta {
+                delta: text.clone(),
+                content_index,
+            }))
             .await;
     }
 }
@@ -669,7 +675,9 @@ async fn process_chat_sse<S>(
             }
 
             // Emit end-of-turn when finish_reason signals completion.
-            if let Some(finish_reason) = choice.get("finish_reason").and_then(|v| v.as_str()) {
+            if let Some(finish_reason) = choice.get("finish_reason").and_then(|v| v.as_str())
+                && !finish_reason.is_empty()
+            {
                 match finish_reason {
                     "tool_calls" if fn_call_state.active => {
                         // First, flush the terminal raw reasoning so UIs can finalize
@@ -898,20 +906,26 @@ where
                         continue;
                     }
                 }
-                Poll::Ready(Some(Ok(ResponseEvent::ReasoningContentDelta(delta)))) => {
+                Poll::Ready(Some(Ok(ResponseEvent::ReasoningContentDelta {
+                    delta,
+                    content_index,
+                }))) => {
                     // Always accumulate reasoning deltas so we can emit a final Reasoning item at Completed.
                     this.cumulative_reasoning.push_str(&delta);
                     if matches!(this.mode, AggregateMode::Streaming) {
                         // In streaming mode, also forward the delta immediately.
-                        return Poll::Ready(Some(Ok(ResponseEvent::ReasoningContentDelta(delta))));
+                        return Poll::Ready(Some(Ok(ResponseEvent::ReasoningContentDelta {
+                            delta,
+                            content_index,
+                        })));
                     } else {
                         continue;
                     }
                 }
-                Poll::Ready(Some(Ok(ResponseEvent::ReasoningSummaryDelta(_)))) => {
+                Poll::Ready(Some(Ok(ResponseEvent::ReasoningSummaryDelta { .. }))) => {
                     continue;
                 }
-                Poll::Ready(Some(Ok(ResponseEvent::ReasoningSummaryPartAdded))) => {
+                Poll::Ready(Some(Ok(ResponseEvent::ReasoningSummaryPartAdded { .. }))) => {
                     continue;
                 }
                 Poll::Ready(Some(Ok(ResponseEvent::OutputItemAdded(item)))) => {

codex-rs/core/src/client.rs

@@ -26,6 +26,7 @@ use tokio::sync::mpsc;
 use tokio::time::timeout;
 use tokio_util::io::ReaderStream;
 use tracing::debug;
+use tracing::enabled;
 use tracing::trace;
 use tracing::warn;
 
@@ -35,10 +36,10 @@ use crate::auth::RefreshTokenError;
 use crate::chat_completions::AggregateStreamExt;
 use crate::chat_completions::stream_chat_completions;
 use crate::client_common::Prompt;
+use crate::client_common::Reasoning;
 use crate::client_common::ResponseEvent;
 use crate::client_common::ResponseStream;
 use crate::client_common::ResponsesApiRequest;
-use crate::client_common::create_reasoning_param_for_request;
 use crate::client_common::create_text_param_for_request;
 use crate::config::Config;
 use crate::default_client::CodexHttpClient;
@@ -55,6 +56,7 @@ use crate::model_family::ModelFamily;
 use crate::model_provider_info::ModelProviderInfo;
 use crate::model_provider_info::WireApi;
 use crate::openai_model_info::get_model_info;
+use crate::protocol::CreditsSnapshot;
 use crate::protocol::RateLimitSnapshot;
 use crate::protocol::RateLimitWindow;
 use crate::protocol::TokenUsage;
@@ -78,6 +80,18 @@ struct Error {
     resets_at: Option<i64>,
 }
 
+#[derive(Debug, Serialize)]
+struct CompactHistoryRequest<'a> {
+    model: &'a str,
+    input: &'a [ResponseItem],
+    instructions: &'a str,
+}
+
+#[derive(Debug, Deserialize)]
+struct CompactHistoryResponse {
+    output: Vec<ResponseItem>,
+}
+
 #[derive(Debug, Clone)]
 pub struct ModelClient {
     config: Arc<Config>,
@@ -199,12 +213,18 @@ impl ModelClient {
         let auth_manager = self.auth_manager.clone();
 
         let full_instructions = prompt.get_full_instructions(&self.config.model_family);
-        let tools_json = create_tools_json_for_responses_api(&prompt.tools)?;
-        let reasoning = create_reasoning_param_for_request(
-            &self.config.model_family,
-            self.effort,
-            self.summary,
-        );
+        let tools_json: Vec<Value> = create_tools_json_for_responses_api(&prompt.tools)?;
+
+        let reasoning = if self.config.model_family.supports_reasoning_summaries {
+            Some(Reasoning {
+                effort: self
+                    .effort
+                    .or(self.config.model_family.default_reasoning_effort),
+                summary: Some(self.summary),
+            })
+        } else {
+            None
+        };
 
         let include: Vec<String> = if reasoning.is_some() {
             vec!["reasoning.encrypted_content".to_string()]
@@ -215,7 +235,9 @@ impl ModelClient {
         let input_with_instructions = prompt.get_formatted_input();
 
         let verbosity = if self.config.model_family.support_verbosity {
-            self.config.model_verbosity
+            self.config
+                .model_verbosity
+                .or(self.config.model_family.default_verbosity)
         } else {
             if self.config.model_verbosity.is_some() {
                 warn!(
@@ -294,10 +316,9 @@ impl ModelClient {
         let auth = auth_manager.as_ref().and_then(|m| m.auth());
 
         trace!(
-            "POST to {}: {:?}",
+            "POST to {}: {}",
             self.provider.get_full_url(&auth),
-            serde_json::to_string(payload_json)
-                .unwrap_or("<unable to serialize payload>".to_string())
+            payload_json.to_string()
         );
 
         let mut req_builder = self
@@ -500,6 +521,70 @@ impl ModelClient {
     pub fn get_auth_manager(&self) -> Option<Arc<AuthManager>> {
         self.auth_manager.clone()
     }
+
+    pub async fn compact_conversation_history(&self, prompt: &Prompt) -> Result<Vec<ResponseItem>> {
+        if prompt.input.is_empty() {
+            return Ok(Vec::new());
+        }
+        let auth_manager = self.auth_manager.clone();
+        let auth = auth_manager.as_ref().and_then(|m| m.auth());
+        let mut req_builder = self
+            .provider
+            .create_compact_request_builder(&self.client, &auth)
+            .await?;
+        if let SessionSource::SubAgent(sub) = &self.session_source {
+            let subagent = if let crate::protocol::SubAgentSource::Other(label) = sub {
+                label.clone()
+            } else {
+                serde_json::to_value(sub)
+                    .ok()
+                    .and_then(|v| v.as_str().map(std::string::ToString::to_string))
+                    .unwrap_or_else(|| "other".to_string())
+            };
+            req_builder = req_builder.header("x-openai-subagent", subagent);
+        }
+        if let Some(auth) = auth.as_ref()
+            && auth.mode == AuthMode::ChatGPT
+            && let Some(account_id) = auth.get_account_id()
+        {
+            req_builder = req_builder.header("chatgpt-account-id", account_id);
+        }
+        let payload = CompactHistoryRequest {
+            model: &self.config.model,
+            input: &prompt.input,
+            instructions: &prompt.get_full_instructions(&self.config.model_family),
+        };
+
+        if enabled!(tracing::Level::TRACE) {
+            trace!(
+                "POST to {}: {}",
+                self.provider
+                    .get_compact_url(&auth)
+                    .unwrap_or("<none>".to_string()),
+                serde_json::to_value(&payload).unwrap_or_default()
+            );
+        }
+
+        let response = req_builder
+            .json(&payload)
+            .send()
+            .await
+            .map_err(|source| CodexErr::ConnectionFailed(ConnectionFailedError { source }))?;
+        let status = response.status();
+        let body = response
+            .text()
+            .await
+            .map_err(|source| CodexErr::ConnectionFailed(ConnectionFailedError { source }))?;
+        if !status.is_success() {
+            return Err(CodexErr::UnexpectedStatus(UnexpectedResponseError {
+                status,
+                body,
+                request_id: None,
+            }));
+        }
+        let CompactHistoryResponse { output } = serde_json::from_str(&body)?;
+        Ok(output)
+    }
 }
 
 enum StreamAttemptError {
@@ -553,6 +638,8 @@ struct SseEvent {
     response: Option<Value>,
     item: Option<Value>,
     delta: Option<String>,
+    summary_index: Option<i64>,
+    content_index: Option<i64>,
 }
 
 #[derive(Debug, Deserialize)]
@@ -640,7 +727,13 @@ fn parse_rate_limit_snapshot(headers: &HeaderMap) -> Option<RateLimitSnapshot> {
         "x-codex-secondary-reset-at",
     );
 
-    Some(RateLimitSnapshot { primary, secondary })
+    let credits = parse_credits_snapshot(headers);
+
+    Some(RateLimitSnapshot {
+        primary,
+        secondary,
+        credits,
+    })
 }
 
 fn parse_rate_limit_window(
@@ -667,6 +760,20 @@ fn parse_rate_limit_window(
     })
 }
 
+fn parse_credits_snapshot(headers: &HeaderMap) -> Option<CreditsSnapshot> {
+    let has_credits = parse_header_bool(headers, "x-codex-credits-has-credits")?;
+    let unlimited = parse_header_bool(headers, "x-codex-credits-unlimited")?;
+    let balance = parse_header_str(headers, "x-codex-credits-balance")
+        .map(str::trim)
+        .filter(|value| !value.is_empty())
+        .map(std::string::ToString::to_string);
+    Some(CreditsSnapshot {
+        has_credits,
+        unlimited,
+        balance,
+    })
+}
+
 fn parse_header_f64(headers: &HeaderMap, name: &str) -> Option<f64> {
     parse_header_str(headers, name)?
         .parse::<f64>()
@@ -678,6 +785,17 @@ fn parse_header_i64(headers: &HeaderMap, name: &str) -> Option<i64> {
     parse_header_str(headers, name)?.parse::<i64>().ok()
 }
 
+fn parse_header_bool(headers: &HeaderMap, name: &str) -> Option<bool> {
+    let raw = parse_header_str(headers, name)?;
+    if raw.eq_ignore_ascii_case("true") || raw == "1" {
+        Some(true)
+    } else if raw.eq_ignore_ascii_case("false") || raw == "0" {
+        Some(false)
+    } else {
+        None
+    }
+}
+
 fn parse_header_str<'a>(headers: &'a HeaderMap, name: &str) -> Option<&'a str> {
     headers.get(name)?.to_str().ok()
 }
@@ -812,16 +930,22 @@ async fn process_sse<S>(
                 }
             }
             "response.reasoning_summary_text.delta" => {
-                if let Some(delta) = event.delta {
-                    let event = ResponseEvent::ReasoningSummaryDelta(delta);
+                if let (Some(delta), Some(summary_index)) = (event.delta, event.summary_index) {
+                    let event = ResponseEvent::ReasoningSummaryDelta {
+                        delta,
+                        summary_index,
+                    };
                     if tx_event.send(Ok(event)).await.is_err() {
                         return;
                     }
                 }
             }
             "response.reasoning_text.delta" => {
-                if let Some(delta) = event.delta {
-                    let event = ResponseEvent::ReasoningContentDelta(delta);
+                if let (Some(delta), Some(content_index)) = (event.delta, event.content_index) {
+                    let event = ResponseEvent::ReasoningContentDelta {
+                        delta,
+                        content_index,
+                    };
                     if tx_event.send(Ok(event)).await.is_err() {
                         return;
                     }
@@ -898,10 +1022,12 @@ async fn process_sse<S>(
                 }
             }
             "response.reasoning_summary_part.added" => {
-                // Boundary between reasoning summary sections (e.g., titles).
-                let event = ResponseEvent::ReasoningSummaryPartAdded;
-                if tx_event.send(Ok(event)).await.is_err() {
-                    return;
+                if let Some(summary_index) = event.summary_index {
+                    // Boundary between reasoning summary sections (e.g., titles).
+                    let event = ResponseEvent::ReasoningSummaryPartAdded { summary_index };
+                    if tx_event.send(Ok(event)).await.is_err() {
+                        return;
+                    }
                 }
             }
             "response.reasoning_summary_text.done" => {}
@@ -1208,7 +1334,7 @@ mod tests {
 
     #[tokio::test]
     async fn error_when_error_event() {
-        let raw_error = r#"{"type":"response.failed","sequence_number":3,"response":{"id":"resp_689bcf18d7f08194bf3440ba62fe05d803fee0cdac429894","object":"response","created_at":1755041560,"status":"failed","background":false,"error":{"code":"rate_limit_exceeded","message":"Rate limit reached for gpt-5 in organization org-AAA on tokens per min (TPM): Limit 30000, Used 22999, Requested 12528. Please try again in 11.054s. Visit https://platform.openai.com/account/rate-limits to learn more."}, "usage":null,"user":null,"metadata":{}}}"#;
+        let raw_error = r#"{"type":"response.failed","sequence_number":3,"response":{"id":"resp_689bcf18d7f08194bf3440ba62fe05d803fee0cdac429894","object":"response","created_at":1755041560,"status":"failed","background":false,"error":{"code":"rate_limit_exceeded","message":"Rate limit reached for gpt-5.1 in organization org-AAA on tokens per min (TPM): Limit 30000, Used 22999, Requested 12528. Please try again in 11.054s. Visit https://platform.openai.com/account/rate-limits to learn more."}, "usage":null,"user":null,"metadata":{}}}"#;
 
         let sse1 = format!("event: response.failed\ndata: {raw_error}\n\n");
         let provider = ModelProviderInfo {
@@ -1237,7 +1363,7 @@ mod tests {
             Err(CodexErr::Stream(msg, delay)) => {
                 assert_eq!(
                     msg,
-                    "Rate limit reached for gpt-5 in organization org-AAA on tokens per min (TPM): Limit 30000, Used 22999, Requested 12528. Please try again in 11.054s. Visit https://platform.openai.com/account/rate-limits to learn more."
+                    "Rate limit reached for gpt-5.1 in organization org-AAA on tokens per min (TPM): Limit 30000, Used 22999, Requested 12528. Please try again in 11.054s. Visit https://platform.openai.com/account/rate-limits to learn more."
                 );
                 assert_eq!(*delay, Some(Duration::from_secs_f64(11.054)));
             }
@@ -1456,7 +1582,7 @@ mod tests {
     fn test_try_parse_retry_after() {
         let err = Error {
             r#type: None,
-            message: Some("Rate limit reached for gpt-5 in organization org- on tokens per min (TPM): Limit 1, Used 1, Requested 19304. Please try again in 28ms. Visit https://platform.openai.com/account/rate-limits to learn more.".to_string()),
+            message: Some("Rate limit reached for gpt-5.1 in organization org- on tokens per min (TPM): Limit 1, Used 1, Requested 19304. Please try again in 28ms. Visit https://platform.openai.com/account/rate-limits to learn more.".to_string()),
             code: Some("rate_limit_exceeded".to_string()),
             plan_type: None,
             resets_at: None
@@ -1470,7 +1596,7 @@ mod tests {
     fn test_try_parse_retry_after_no_delay() {
         let err = Error {
             r#type: None,
-            message: Some("Rate limit reached for gpt-5 in organization <ORG> on tokens per min (TPM): Limit 30000, Used 6899, Requested 24050. Please try again in 1.898s. Visit https://platform.openai.com/account/rate-limits to learn more.".to_string()),
+            message: Some("Rate limit reached for gpt-5.1 in organization <ORG> on tokens per min (TPM): Limit 30000, Used 6899, Requested 24050. Please try again in 1.898s. Visit https://platform.openai.com/account/rate-limits to learn more.".to_string()),
             code: Some("rate_limit_exceeded".to_string()),
             plan_type: None,
             resets_at: None

codex-rs/core/src/client_common.rs

@@ -165,11 +165,9 @@ fn build_structured_output(parsed: &ExecOutputJson) -> String {
     ));
 
     let mut output = parsed.output.clone();
-    if let Some(total_lines) = extract_total_output_lines(&parsed.output) {
+    if let Some((stripped, total_lines)) = strip_total_output_header(&parsed.output) {
         sections.push(format!("Total output lines: {total_lines}"));
-        if let Some(stripped) = strip_total_output_header(&output) {
-            output = stripped.to_string();
-        }
+        output = stripped.to_string();
     }
 
     sections.push("Output:".to_string());
@@ -178,19 +176,12 @@ fn build_structured_output(parsed: &ExecOutputJson) -> String {
     sections.join("\n")
 }
 
-fn extract_total_output_lines(output: &str) -> Option<u32> {
-    let marker_start = output.find("[... omitted ")?;
-    let marker = &output[marker_start..];
-    let (_, after_of) = marker.split_once(" of ")?;
-    let (total_segment, _) = after_of.split_once(' ')?;
-    total_segment.parse::<u32>().ok()
-}
-
-fn strip_total_output_header(output: &str) -> Option<&str> {
+fn strip_total_output_header(output: &str) -> Option<(&str, u32)> {
     let after_prefix = output.strip_prefix("Total output lines: ")?;
-    let (_, remainder) = after_prefix.split_once('\n')?;
+    let (total_segment, remainder) = after_prefix.split_once('\n')?;
+    let total_lines = total_segment.parse::<u32>().ok()?;
     let remainder = remainder.strip_prefix('\n').unwrap_or(remainder);
-    Some(remainder)
+    Some((remainder, total_lines))
 }
 
 #[derive(Debug)]
@@ -203,9 +194,17 @@ pub enum ResponseEvent {
         token_usage: Option<TokenUsage>,
     },
     OutputTextDelta(String),
-    ReasoningSummaryDelta(String),
-    ReasoningContentDelta(String),
-    ReasoningSummaryPartAdded,
+    ReasoningSummaryDelta {
+        delta: String,
+        summary_index: i64,
+    },
+    ReasoningContentDelta {
+        delta: String,
+        content_index: i64,
+    },
+    ReasoningSummaryPartAdded {
+        summary_index: i64,
+    },
     RateLimits(RateLimitSnapshot),
 }
 
@@ -342,21 +341,6 @@ pub(crate) mod tools {
     }
 }
 
-pub(crate) fn create_reasoning_param_for_request(
-    model_family: &ModelFamily,
-    effort: Option<ReasoningEffortConfig>,
-    summary: ReasoningSummaryConfig,
-) -> Option<Reasoning> {
-    if !model_family.supports_reasoning_summaries {
-        return None;
-    }
-
-    Some(Reasoning {
-        effort,
-        summary: Some(summary),
-    })
-}
-
 pub(crate) fn create_text_param_for_request(
     verbosity: Option<VerbosityConfig>,
     output_schema: &Option<Value>,
@@ -421,6 +405,10 @@ mod tests {
                 slug: "gpt-5",
                 expects_apply_patch_instructions: true,
             },
+            InstructionsTestCase {
+                slug: "gpt-5.1",
+                expects_apply_patch_instructions: false,
+            },
             InstructionsTestCase {
                 slug: "codex-mini-latest",
                 expects_apply_patch_instructions: true,
@@ -430,7 +418,11 @@ mod tests {
                 expects_apply_patch_instructions: false,
             },
             InstructionsTestCase {
-                slug: "gpt-5-codex",
+                slug: "gpt-5.1-codex",
+                expects_apply_patch_instructions: false,
+            },
+            InstructionsTestCase {
+                slug: "gpt-5.1-codex-max",
                 expects_apply_patch_instructions: false,
             },
         ];
@@ -456,7 +448,7 @@ mod tests {
         let input: Vec<ResponseItem> = vec![];
         let tools: Vec<serde_json::Value> = vec![];
         let req = ResponsesApiRequest {
-            model: "gpt-5",
+            model: "gpt-5.1",
             instructions: "i",
             input: &input,
             tools: &tools,
@@ -497,7 +489,7 @@ mod tests {
             create_text_param_for_request(None, &Some(schema.clone())).expect("text controls");
 
         let req = ResponsesApiRequest {
-            model: "gpt-5",
+            model: "gpt-5.1",
             instructions: "i",
             input: &input,
             tools: &tools,
@@ -533,7 +525,7 @@ mod tests {
         let input: Vec<ResponseItem> = vec![];
         let tools: Vec<serde_json::Value> = vec![];
         let req = ResponsesApiRequest {
-            model: "gpt-5",
+            model: "gpt-5.1",
             instructions: "i",
             input: &input,
             tools: &tools,

codex-rs/core/src/codex_delegate.rs

@@ -4,6 +4,7 @@ use std::sync::atomic::AtomicU64;
 use async_channel::Receiver;
 use async_channel::Sender;
 use codex_async_utils::OrCancelExt;
+use codex_protocol::ConversationId;
 use codex_protocol::protocol::ApplyPatchApprovalRequestEvent;
 use codex_protocol::protocol::Event;
 use codex_protocol::protocol::EventMsg;
@@ -30,13 +31,16 @@ use codex_protocol::protocol::InitialHistory;
 /// The returned `events_rx` yields non-approval events emitted by the sub-agent.
 /// Approval requests are handled via `parent_session` and are not surfaced.
 /// The returned `ops_tx` allows the caller to submit additional `Op`s to the sub-agent.
+#[allow(clippy::too_many_arguments)]
 pub(crate) async fn run_codex_conversation_interactive(
     config: Config,
     auth_manager: Arc<AuthManager>,
     parent_session: Arc<Session>,
     parent_ctx: Arc<TurnContext>,
     cancel_token: CancellationToken,
+    desired_conversation_id: Option<ConversationId>,
     initial_history: Option<InitialHistory>,
+    source: SubAgentSource,
 ) -> Result<Codex, CodexErr> {
     let (tx_sub, rx_sub) = async_channel::bounded(SUBMISSION_CHANNEL_CAPACITY);
     let (tx_ops, rx_ops) = async_channel::bounded(SUBMISSION_CHANNEL_CAPACITY);
@@ -45,7 +49,8 @@ pub(crate) async fn run_codex_conversation_interactive(
         config,
         auth_manager,
         initial_history.unwrap_or(InitialHistory::New),
-        SessionSource::SubAgent(SubAgentSource::Review),
+        SessionSource::SubAgent(source),
+        desired_conversation_id,
     )
     .await?;
     let codex = Arc::new(codex);
@@ -81,6 +86,7 @@ pub(crate) async fn run_codex_conversation_interactive(
         next_id: AtomicU64::new(0),
         tx_sub: tx_ops,
         rx_event: rx_sub,
+        conversation_id: codex.conversation_id(),
     })
 }
 
@@ -105,13 +111,16 @@ pub(crate) async fn run_codex_conversation_one_shot(
         parent_session,
         parent_ctx,
         child_cancel.clone(),
+        None,
         initial_history,
+        SubAgentSource::Review,
     )
     .await?;
 
     // Send the initial input to kick off the one-shot turn.
     io.submit(Op::UserInput { items: input }).await?;
 
+    let conversation_id = io.conversation_id();
     // Bridge events so we can observe completion and shut down automatically.
     let (tx_bridge, rx_bridge) = async_channel::bounded(SUBMISSION_CHANNEL_CAPACITY);
     let ops_tx = io.tx_sub.clone();
@@ -146,6 +155,7 @@ pub(crate) async fn run_codex_conversation_one_shot(
         next_id: AtomicU64::new(0),
         rx_event: rx_bridge,
         tx_sub: tx_closed,
+        conversation_id,
     })
 }
 

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

@@ -1,14 +1,19 @@
 use codex_protocol::protocol::AskForApproval;
 use codex_protocol::protocol::SandboxPolicy;
 
+use crate::sandboxing::SandboxPermissions;
+
 use crate::bash::parse_shell_lc_plain_commands;
 use crate::is_safe_command::is_known_safe_command;
+#[cfg(windows)]
+#[path = "windows_dangerous_commands.rs"]
+mod windows_dangerous_commands;
 
 pub fn requires_initial_appoval(
     policy: AskForApproval,
     sandbox_policy: &SandboxPolicy,
     command: &[String],
-    with_escalated_permissions: bool,
+    sandbox_permissions: SandboxPermissions,
 ) -> bool {
     if is_known_safe_command(command) {
         return false;
@@ -24,8 +29,7 @@ pub fn requires_initial_appoval(
             // In restricted sandboxes (ReadOnly/WorkspaceWrite), do not prompt for
             // non‑escalated, non‑dangerous commands — let the sandbox enforce
             // restrictions (e.g., block network/write) without a user prompt.
-            let wants_escalation: bool = with_escalated_permissions;
-            if wants_escalation {
+            if sandbox_permissions.requires_escalated_permissions() {
                 return true;
             }
             command_might_be_dangerous(command)
@@ -35,6 +39,13 @@ pub fn requires_initial_appoval(
 }
 
 pub fn command_might_be_dangerous(command: &[String]) -> bool {
+    #[cfg(windows)]
+    {
+        if windows_dangerous_commands::is_dangerous_command_windows(command) {
+            return true;
+        }
+    }
+
     if is_dangerous_to_call_with_exec(command) {
         return true;
     }

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

@@ -1,4 +1,5 @@
 use crate::bash::parse_shell_lc_plain_commands;
+use crate::command_safety::windows_safe_commands::is_safe_command_windows;
 
 pub fn is_known_safe_command(command: &[String]) -> bool {
     let command: Vec<String> = command
@@ -11,12 +12,9 @@ pub fn is_known_safe_command(command: &[String]) -> bool {
             }
         })
         .collect();
-    #[cfg(target_os = "windows")]
-    {
-        use super::windows_safe_commands::is_safe_command_windows;
-        if is_safe_command_windows(&command) {
-            return true;
-        }
+
+    if is_safe_command_windows(&command) {
+        return true;
     }
 
     if is_safe_to_call_with_exec(&command) {
@@ -269,6 +267,20 @@ mod tests {
         }
     }
 
+    #[test]
+    fn windows_powershell_full_path_is_safe() {
+        if !cfg!(windows) {
+            // Windows only because on Linux path splitting doesn't handle `/` separators properly
+            return;
+        }
+
+        assert!(is_known_safe_command(&vec_str(&[
+            r"C:\Program Files\PowerShell\7\pwsh.exe",
+            "-Command",
+            "Get-Location",
+        ])));
+    }
+
     #[test]
     fn bash_lc_safe_examples() {
         assert!(is_known_safe_command(&vec_str(&["bash", "-lc", "ls"])));

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

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

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

@@ -0,0 +1,316 @@
+use std::path::Path;
+
+use once_cell::sync::Lazy;
+use regex::Regex;
+use shlex::split as shlex_split;
+use url::Url;
+
+pub fn is_dangerous_command_windows(command: &[String]) -> bool {
+    // Prefer structured parsing for PowerShell/CMD so we can spot URL-bearing
+    // invocations of ShellExecute-style entry points before falling back to
+    // simple argv heuristics.
+    if is_dangerous_powershell(command) {
+        return true;
+    }
+
+    if is_dangerous_cmd(command) {
+        return true;
+    }
+
+    is_direct_gui_launch(command)
+}
+
+fn is_dangerous_powershell(command: &[String]) -> bool {
+    let Some((exe, rest)) = command.split_first() else {
+        return false;
+    };
+    if !is_powershell_executable(exe) {
+        return false;
+    }
+    // Parse the PowerShell invocation to get a flat token list we can scan for
+    // dangerous cmdlets/COM calls plus any URL-looking arguments. This is a
+    // best-effort shlex split of the script text, not a full PS parser.
+    let Some(parsed) = parse_powershell_invocation(rest) else {
+        return false;
+    };
+
+    let tokens_lc: Vec<String> = parsed
+        .tokens
+        .iter()
+        .map(|t| t.trim_matches('\'').trim_matches('"').to_ascii_lowercase())
+        .collect();
+    let has_url = args_have_url(&parsed.tokens);
+
+    if has_url
+        && tokens_lc.iter().any(|t| {
+            matches!(
+                t.as_str(),
+                "start-process" | "start" | "saps" | "invoke-item" | "ii"
+            ) || t.contains("start-process")
+                || t.contains("invoke-item")
+        })
+    {
+        return true;
+    }
+
+    if has_url
+        && tokens_lc
+            .iter()
+            .any(|t| t.contains("shellexecute") || t.contains("shell.application"))
+    {
+        return true;
+    }
+
+    if let Some(first) = tokens_lc.first() {
+        // Legacy ShellExecute path via url.dll
+        if first == "rundll32"
+            && tokens_lc
+                .iter()
+                .any(|t| t.contains("url.dll,fileprotocolhandler"))
+            && has_url
+        {
+            return true;
+        }
+        if first == "mshta" && has_url {
+            return true;
+        }
+        if is_browser_executable(first) && has_url {
+            return true;
+        }
+        if matches!(first.as_str(), "explorer" | "explorer.exe") && has_url {
+            return true;
+        }
+    }
+
+    false
+}
+
+fn is_dangerous_cmd(command: &[String]) -> bool {
+    let Some((exe, rest)) = command.split_first() else {
+        return false;
+    };
+    let Some(base) = executable_basename(exe) else {
+        return false;
+    };
+    if base != "cmd" && base != "cmd.exe" {
+        return false;
+    }
+
+    let mut iter = rest.iter();
+    for arg in iter.by_ref() {
+        let lower = arg.to_ascii_lowercase();
+        match lower.as_str() {
+            "/c" | "/r" | "-c" => break,
+            _ if lower.starts_with('/') => continue,
+            // Unknown tokens before the command body => bail.
+            _ => return false,
+        }
+    }
+
+    let Some(first_cmd) = iter.next() else {
+        return false;
+    };
+    // Classic `cmd /c start https://...` ShellExecute path.
+    if !first_cmd.eq_ignore_ascii_case("start") {
+        return false;
+    }
+    let remaining: Vec<String> = iter.cloned().collect();
+    args_have_url(&remaining)
+}
+
+fn is_direct_gui_launch(command: &[String]) -> bool {
+    let Some((exe, rest)) = command.split_first() else {
+        return false;
+    };
+    let Some(base) = executable_basename(exe) else {
+        return false;
+    };
+
+    // Explorer/rundll32/mshta or direct browser exe with a URL anywhere in args.
+    if matches!(base.as_str(), "explorer" | "explorer.exe") && args_have_url(rest) {
+        return true;
+    }
+    if matches!(base.as_str(), "mshta" | "mshta.exe") && args_have_url(rest) {
+        return true;
+    }
+    if (base == "rundll32" || base == "rundll32.exe")
+        && rest.iter().any(|t| {
+            t.to_ascii_lowercase()
+                .contains("url.dll,fileprotocolhandler")
+        })
+        && args_have_url(rest)
+    {
+        return true;
+    }
+    if is_browser_executable(&base) && args_have_url(rest) {
+        return true;
+    }
+
+    false
+}
+
+fn args_have_url(args: &[String]) -> bool {
+    args.iter().any(|arg| looks_like_url(arg))
+}
+
+fn looks_like_url(token: &str) -> bool {
+    // Strip common PowerShell punctuation around inline URLs (quotes, parens, trailing semicolons).
+    // Capture the middle token after trimming leading quotes/parens/whitespace and trailing semicolons/closing parens.
+    static RE: Lazy<Option<Regex>> =
+        Lazy::new(|| Regex::new(r#"^[ "'\(\s]*([^\s"'\);]+)[\s;\)]*$"#).ok());
+    // If the token embeds a URL alongside other text (e.g., Start-Process('https://...'))
+    // as a single shlex token, grab the substring starting at the first URL prefix.
+    let urlish = token
+        .find("https://")
+        .or_else(|| token.find("http://"))
+        .map(|idx| &token[idx..])
+        .unwrap_or(token);
+
+    let candidate = RE
+        .as_ref()
+        .and_then(|re| re.captures(urlish))
+        .and_then(|caps| caps.get(1))
+        .map(|m| m.as_str())
+        .unwrap_or(urlish);
+    let Ok(url) = Url::parse(candidate) else {
+        return false;
+    };
+    matches!(url.scheme(), "http" | "https")
+}
+
+fn executable_basename(exe: &str) -> Option<String> {
+    Path::new(exe)
+        .file_name()
+        .and_then(|osstr| osstr.to_str())
+        .map(str::to_ascii_lowercase)
+}
+
+fn is_powershell_executable(exe: &str) -> bool {
+    matches!(
+        executable_basename(exe).as_deref(),
+        Some("powershell") | Some("powershell.exe") | Some("pwsh") | Some("pwsh.exe")
+    )
+}
+
+fn is_browser_executable(name: &str) -> bool {
+    matches!(
+        name,
+        "chrome"
+            | "chrome.exe"
+            | "msedge"
+            | "msedge.exe"
+            | "firefox"
+            | "firefox.exe"
+            | "iexplore"
+            | "iexplore.exe"
+    )
+}
+
+struct ParsedPowershell {
+    tokens: Vec<String>,
+}
+
+fn parse_powershell_invocation(args: &[String]) -> Option<ParsedPowershell> {
+    if args.is_empty() {
+        return None;
+    }
+
+    let mut idx = 0;
+    while idx < args.len() {
+        let arg = &args[idx];
+        let lower = arg.to_ascii_lowercase();
+        match lower.as_str() {
+            "-command" | "/command" | "-c" => {
+                let script = args.get(idx + 1)?;
+                if idx + 2 != args.len() {
+                    return None;
+                }
+                let tokens = shlex_split(script)?;
+                return Some(ParsedPowershell { tokens });
+            }
+            _ if lower.starts_with("-command:") || lower.starts_with("/command:") => {
+                if idx + 1 != args.len() {
+                    return None;
+                }
+                let (_, script) = arg.split_once(':')?;
+                let tokens = shlex_split(script)?;
+                return Some(ParsedPowershell { tokens });
+            }
+            "-nologo" | "-noprofile" | "-noninteractive" | "-mta" | "-sta" => {
+                idx += 1;
+            }
+            _ if lower.starts_with('-') => {
+                idx += 1;
+            }
+            _ => {
+                let rest = args[idx..].to_vec();
+                return Some(ParsedPowershell { tokens: rest });
+            }
+        }
+    }
+
+    None
+}
+
+#[cfg(test)]
+mod tests {
+    use super::is_dangerous_command_windows;
+
+    fn vec_str(items: &[&str]) -> Vec<String> {
+        items.iter().map(std::string::ToString::to_string).collect()
+    }
+
+    #[test]
+    fn powershell_start_process_url_is_dangerous() {
+        assert!(is_dangerous_command_windows(&vec_str(&[
+            "powershell",
+            "-NoLogo",
+            "-Command",
+            "Start-Process 'https://example.com'"
+        ])));
+    }
+
+    #[test]
+    fn powershell_start_process_url_with_trailing_semicolon_is_dangerous() {
+        assert!(is_dangerous_command_windows(&vec_str(&[
+            "powershell",
+            "-Command",
+            "Start-Process('https://example.com');"
+        ])));
+    }
+
+    #[test]
+    fn powershell_start_process_local_is_not_flagged() {
+        assert!(!is_dangerous_command_windows(&vec_str(&[
+            "powershell",
+            "-Command",
+            "Start-Process notepad.exe"
+        ])));
+    }
+
+    #[test]
+    fn cmd_start_with_url_is_dangerous() {
+        assert!(is_dangerous_command_windows(&vec_str(&[
+            "cmd",
+            "/c",
+            "start",
+            "https://example.com"
+        ])));
+    }
+
+    #[test]
+    fn msedge_with_url_is_dangerous() {
+        assert!(is_dangerous_command_windows(&vec_str(&[
+            "msedge.exe",
+            "https://example.com"
+        ])));
+    }
+
+    #[test]
+    fn explorer_with_directory_is_not_flagged() {
+        assert!(!is_dangerous_command_windows(&vec_str(&[
+            "explorer.exe",
+            "."
+        ])));
+    }
+}

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

@@ -1,4 +1,5 @@
 use shlex::split as shlex_split;
+use std::path::Path;
 
 /// On Windows, we conservatively allow only clearly read-only PowerShell invocations
 /// that match a small safelist. Anything else (including direct CMD commands) is unsafe.
@@ -131,8 +132,14 @@ fn split_into_commands(tokens: Vec<String>) -> Option<Vec<Vec<String>>> {
 
 /// Returns true when the executable name is one of the supported PowerShell binaries.
 fn is_powershell_executable(exe: &str) -> bool {
+    let executable_name = Path::new(exe)
+        .file_name()
+        .and_then(|osstr| osstr.to_str())
+        .unwrap_or(exe)
+        .to_ascii_lowercase();
+
     matches!(
-        exe.to_ascii_lowercase().as_str(),
+        executable_name.as_str(),
         "powershell" | "powershell.exe" | "pwsh" | "pwsh.exe"
     )
 }
@@ -313,6 +320,27 @@ mod tests {
         ])));
     }
 
+    #[test]
+    fn accepts_full_path_powershell_invocations() {
+        if !cfg!(windows) {
+            // Windows only because on Linux path splitting doesn't handle `/` separators properly
+            return;
+        }
+
+        assert!(is_safe_command_windows(&vec_str(&[
+            r"C:\Program Files\PowerShell\7\pwsh.exe",
+            "-NoProfile",
+            "-Command",
+            "Get-ChildItem -Path .",
+        ])));
+
+        assert!(is_safe_command_windows(&vec_str(&[
+            r"C:\Windows\System32\WindowsPowerShell\v1.0\powershell.exe",
+            "-Command",
+            "Get-Content Cargo.toml",
+        ])));
+    }
+
     #[test]
     fn allows_read_only_pipelines_and_git_usage() {
         assert!(is_safe_command_windows(&vec_str(&[

codex-rs/core/src/compact.rs

@@ -7,15 +7,18 @@ use crate::codex::TurnContext;
 use crate::codex::get_last_assistant_message_from_turn;
 use crate::error::CodexErr;
 use crate::error::Result as CodexResult;
+use crate::features::Feature;
 use crate::protocol::AgentMessageEvent;
 use crate::protocol::CompactedItem;
-use crate::protocol::ErrorEvent;
 use crate::protocol::EventMsg;
 use crate::protocol::TaskStartedEvent;
 use crate::protocol::TurnContextItem;
 use crate::protocol::WarningEvent;
-use crate::truncate::truncate_middle;
+use crate::truncate::TruncationPolicy;
+use crate::truncate::approx_token_count;
+use crate::truncate::truncate_text;
 use crate::util::backoff;
+use codex_app_server_protocol::AuthMode;
 use codex_protocol::items::TurnItem;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ResponseInputItem;
@@ -26,14 +29,25 @@ use futures::prelude::*;
 use tracing::error;
 
 pub const SUMMARIZATION_PROMPT: &str = include_str!("../templates/compact/prompt.md");
+pub const SUMMARY_PREFIX: &str = include_str!("../templates/compact/summary_prefix.md");
 const COMPACT_USER_MESSAGE_MAX_TOKENS: usize = 20_000;
 
+pub(crate) async fn should_use_remote_compact_task(session: &Session) -> bool {
+    session
+        .services
+        .auth_manager
+        .auth()
+        .is_some_and(|auth| auth.mode == AuthMode::ChatGPT)
+        && session.enabled(Feature::RemoteCompaction).await
+}
+
 pub(crate) async fn run_inline_auto_compact_task(
     sess: Arc<Session>,
     turn_context: Arc<TurnContext>,
 ) {
     let prompt = turn_context.compact_prompt().to_string();
     let input = vec![UserInput::Text { text: prompt }];
+
     run_compact_task_inner(sess, turn_context, input).await;
 }
 
@@ -41,13 +55,12 @@ pub(crate) async fn run_compact_task(
     sess: Arc<Session>,
     turn_context: Arc<TurnContext>,
     input: Vec<UserInput>,
-) -> Option<String> {
+) {
     let start_event = EventMsg::TaskStarted(TaskStartedEvent {
         model_context_window: turn_context.client.get_model_context_window(),
     });
     sess.send_event(&turn_context, start_event).await;
     run_compact_task_inner(sess.clone(), turn_context, input).await;
-    None
 }
 
 async fn run_compact_task_inner(
@@ -58,7 +71,10 @@ async fn run_compact_task_inner(
     let initial_input_for_turn: ResponseInputItem = ResponseInputItem::from(input);
 
     let mut history = sess.clone_history().await;
-    history.record_items(&[initial_input_for_turn.into()]);
+    history.record_items(
+        &[initial_input_for_turn.into()],
+        turn_context.truncation_policy,
+    );
 
     let mut truncated_count = 0usize;
 
@@ -111,9 +127,7 @@ async fn run_compact_task_inner(
                     continue;
                 }
                 sess.set_total_tokens_full(turn_context.as_ref()).await;
-                let event = EventMsg::Error(ErrorEvent {
-                    message: e.to_string(),
-                });
+                let event = EventMsg::Error(e.to_error_event(None));
                 sess.send_event(&turn_context, event).await;
                 return;
             }
@@ -124,14 +138,13 @@ async fn run_compact_task_inner(
                     sess.notify_stream_error(
                         turn_context.as_ref(),
                         format!("Reconnecting... {retries}/{max_retries}"),
+                        e,
                     )
                     .await;
                     tokio::time::sleep(delay).await;
                     continue;
                 } else {
-                    let event = EventMsg::Error(ErrorEvent {
-                        message: e.to_string(),
-                    });
+                    let event = EventMsg::Error(e.to_error_event(None));
                     sess.send_event(&turn_context, event).await;
                     return;
                 }
@@ -140,7 +153,9 @@ async fn run_compact_task_inner(
     }
 
     let history_snapshot = sess.clone_history().await.get_history();
-    let summary_text = get_last_assistant_message_from_turn(&history_snapshot).unwrap_or_default();
+    let summary_suffix =
+        get_last_assistant_message_from_turn(&history_snapshot).unwrap_or_default();
+    let summary_text = format!("{SUMMARY_PREFIX}\n{summary_suffix}");
     let user_messages = collect_user_messages(&history_snapshot);
 
     let initial_context = sess.build_initial_context(turn_context.as_ref());
@@ -152,9 +167,11 @@ async fn run_compact_task_inner(
         .collect();
     new_history.extend(ghost_snapshots);
     sess.replace_history(new_history).await;
+    sess.recompute_token_usage(&turn_context).await;
 
     let rollout_item = RolloutItem::Compacted(CompactedItem {
         message: summary_text.clone(),
+        replacement_history: None,
     });
     sess.persist_rollout_items(&[rollout_item]).await;
 
@@ -192,12 +209,22 @@ pub(crate) fn collect_user_messages(items: &[ResponseItem]) -> Vec<String> {
     items
         .iter()
         .filter_map(|item| match crate::event_mapping::parse_turn_item(item) {
-            Some(TurnItem::UserMessage(user)) => Some(user.message()),
+            Some(TurnItem::UserMessage(user)) => {
+                if is_summary_message(&user.message()) {
+                    None
+                } else {
+                    Some(user.message())
+                }
+            }
             _ => None,
         })
         .collect()
 }
 
+pub(crate) fn is_summary_message(message: &str) -> bool {
+    message.starts_with(format!("{SUMMARY_PREFIX}\n").as_str())
+}
+
 pub(crate) fn build_compacted_history(
     initial_context: Vec<ResponseItem>,
     user_messages: &[String],
@@ -207,7 +234,7 @@ pub(crate) fn build_compacted_history(
         initial_context,
         user_messages,
         summary_text,
-        COMPACT_USER_MESSAGE_MAX_TOKENS * 4,
+        COMPACT_USER_MESSAGE_MAX_TOKENS,
     )
 }
 
@@ -215,20 +242,21 @@ fn build_compacted_history_with_limit(
     mut history: Vec<ResponseItem>,
     user_messages: &[String],
     summary_text: &str,
-    max_bytes: usize,
+    max_tokens: usize,
 ) -> Vec<ResponseItem> {
     let mut selected_messages: Vec<String> = Vec::new();
-    if max_bytes > 0 {
-        let mut remaining = max_bytes;
+    if max_tokens > 0 {
+        let mut remaining = max_tokens;
         for message in user_messages.iter().rev() {
             if remaining == 0 {
                 break;
             }
-            if message.len() <= remaining {
+            let tokens = approx_token_count(message);
+            if tokens <= remaining {
                 selected_messages.push(message.clone());
-                remaining = remaining.saturating_sub(message.len());
+                remaining = remaining.saturating_sub(tokens);
             } else {
-                let (truncated, _) = truncate_middle(message, remaining);
+                let truncated = truncate_text(message, TruncationPolicy::Tokens(remaining));
                 selected_messages.push(truncated);
                 break;
             }
@@ -277,7 +305,8 @@ async fn drain_to_completed(
         };
         match event {
             Ok(ResponseEvent::OutputItemDone(item)) => {
-                sess.record_into_history(std::slice::from_ref(&item)).await;
+                sess.record_into_history(std::slice::from_ref(&item), turn_context)
+                    .await;
             }
             Ok(ResponseEvent::RateLimits(snapshot)) => {
                 sess.update_rate_limits(turn_context, snapshot).await;
@@ -295,6 +324,7 @@ async fn drain_to_completed(
 
 #[cfg(test)]
 mod tests {
+
     use super::*;
     use pretty_assertions::assert_eq;
 
@@ -386,16 +416,16 @@ mod tests {
     }
 
     #[test]
-    fn build_compacted_history_truncates_overlong_user_messages() {
+    fn build_token_limited_compacted_history_truncates_overlong_user_messages() {
         // Use a small truncation limit so the test remains fast while still validating
         // that oversized user content is truncated.
-        let max_bytes = 128;
-        let big = "X".repeat(max_bytes + 50);
+        let max_tokens = 16;
+        let big = "word ".repeat(200);
         let history = super::build_compacted_history_with_limit(
             Vec::new(),
             std::slice::from_ref(&big),
             "SUMMARY",
-            max_bytes,
+            max_tokens,
         );
         assert_eq!(history.len(), 2);
 
@@ -428,7 +458,7 @@ mod tests {
     }
 
     #[test]
-    fn build_compacted_history_appends_summary_message() {
+    fn build_token_limited_compacted_history_appends_summary_message() {
         let initial_context: Vec<ResponseItem> = Vec::new();
         let user_messages = vec!["first user message".to_string()];
         let summary_text = "summary text";

codex-rs/core/src/compact_remote.rs

@@ -0,0 +1,83 @@
+use std::sync::Arc;
+
+use crate::Prompt;
+use crate::codex::Session;
+use crate::codex::TurnContext;
+use crate::error::Result as CodexResult;
+use crate::protocol::AgentMessageEvent;
+use crate::protocol::CompactedItem;
+use crate::protocol::EventMsg;
+use crate::protocol::RolloutItem;
+use crate::protocol::TaskStartedEvent;
+use codex_protocol::models::ResponseItem;
+
+pub(crate) async fn run_inline_remote_auto_compact_task(
+    sess: Arc<Session>,
+    turn_context: Arc<TurnContext>,
+) {
+    run_remote_compact_task_inner(&sess, &turn_context).await;
+}
+
+pub(crate) async fn run_remote_compact_task(sess: Arc<Session>, turn_context: Arc<TurnContext>) {
+    let start_event = EventMsg::TaskStarted(TaskStartedEvent {
+        model_context_window: turn_context.client.get_model_context_window(),
+    });
+    sess.send_event(&turn_context, start_event).await;
+
+    run_remote_compact_task_inner(&sess, &turn_context).await;
+}
+
+async fn run_remote_compact_task_inner(sess: &Arc<Session>, turn_context: &Arc<TurnContext>) {
+    if let Err(err) = run_remote_compact_task_inner_impl(sess, turn_context).await {
+        let event = EventMsg::Error(
+            err.to_error_event(Some("Error running remote compact task".to_string())),
+        );
+        sess.send_event(turn_context, event).await;
+    }
+}
+
+async fn run_remote_compact_task_inner_impl(
+    sess: &Arc<Session>,
+    turn_context: &Arc<TurnContext>,
+) -> CodexResult<()> {
+    let mut history = sess.clone_history().await;
+    let prompt = Prompt {
+        input: history.get_history_for_prompt(),
+        tools: vec![],
+        parallel_tool_calls: false,
+        base_instructions_override: turn_context.base_instructions.clone(),
+        output_schema: None,
+    };
+
+    let mut new_history = turn_context
+        .client
+        .compact_conversation_history(&prompt)
+        .await?;
+    // Required to keep `/undo` available after compaction
+    let ghost_snapshots: Vec<ResponseItem> = history
+        .get_history()
+        .iter()
+        .filter(|item| matches!(item, ResponseItem::GhostSnapshot { .. }))
+        .cloned()
+        .collect();
+
+    if !ghost_snapshots.is_empty() {
+        new_history.extend(ghost_snapshots);
+    }
+    sess.replace_history(new_history.clone()).await;
+    sess.recompute_token_usage(turn_context).await;
+
+    let compacted_item = CompactedItem {
+        message: String::new(),
+        replacement_history: Some(new_history),
+    };
+    sess.persist_rollout_items(&[RolloutItem::Compacted(compacted_item)])
+        .await;
+
+    let event = EventMsg::AgentMessage(AgentMessageEvent {
+        message: "Compact task completed".to_string(),
+    });
+    sess.send_event(turn_context, event).await;
+
+    Ok(())
+}

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

@@ -3,6 +3,7 @@ use crate::config::types::McpServerConfig;
 use crate::config::types::Notice;
 use anyhow::Context;
 use codex_protocol::config_types::ReasoningEffort;
+use codex_protocol::config_types::TrustLevel;
 use std::collections::BTreeMap;
 use std::path::Path;
 use std::path::PathBuf;
@@ -29,11 +30,13 @@ pub enum ConfigEdit {
     SetNoticeHideRateLimitModelNudge(bool),
     /// Toggle the Windows onboarding acknowledgement flag.
     SetWindowsWslSetupAcknowledged(bool),
+    /// Toggle the model migration prompt acknowledgement flag.
+    SetNoticeHideModelMigrationPrompt(String, bool),
     /// Replace the entire `[mcp_servers]` table.
     ReplaceMcpServers(BTreeMap<String, McpServerConfig>),
-    /// Set trust_level = "trusted" under `[projects."<path>"]`,
+    /// Set trust_level under `[projects."<path>"]`,
     /// migrating inline tables to explicit tables.
-    SetProjectTrusted(PathBuf),
+    SetProjectTrustLevel { path: PathBuf, level: TrustLevel },
     /// Set the value stored at the exact dotted path.
     SetPath {
         segments: Vec<String>,
@@ -253,6 +256,13 @@ impl ConfigDocument {
                 &[Notice::TABLE_KEY, "hide_rate_limit_model_nudge"],
                 value(*acknowledged),
             )),
+            ConfigEdit::SetNoticeHideModelMigrationPrompt(migration_config, acknowledged) => {
+                Ok(self.write_value(
+                    Scope::Global,
+                    &[Notice::TABLE_KEY, migration_config.as_str()],
+                    value(*acknowledged),
+                ))
+            }
             ConfigEdit::SetWindowsWslSetupAcknowledged(acknowledged) => Ok(self.write_value(
                 Scope::Global,
                 &["windows_wsl_setup_acknowledged"],
@@ -261,10 +271,14 @@ impl ConfigDocument {
             ConfigEdit::ReplaceMcpServers(servers) => Ok(self.replace_mcp_servers(servers)),
             ConfigEdit::SetPath { segments, value } => Ok(self.insert(segments, value.clone())),
             ConfigEdit::ClearPath { segments } => Ok(self.clear_owned(segments)),
-            ConfigEdit::SetProjectTrusted(project_path) => {
+            ConfigEdit::SetProjectTrustLevel { path, level } => {
                 // Delegate to the existing, tested logic in config.rs to
                 // ensure tables are explicit and migration is preserved.
-                crate::config::set_project_trusted_inner(&mut self.doc, project_path.as_path())?;
+                crate::config::set_project_trust_level_inner(
+                    &mut self.doc,
+                    path.as_path(),
+                    *level,
+                )?;
                 Ok(true)
             }
         }
@@ -499,6 +513,15 @@ impl ConfigEditsBuilder {
         self
     }
 
+    pub fn set_hide_model_migration_prompt(mut self, model: &str, acknowledged: bool) -> Self {
+        self.edits
+            .push(ConfigEdit::SetNoticeHideModelMigrationPrompt(
+                model.to_string(),
+                acknowledged,
+            ));
+        self
+    }
+
     pub fn set_windows_wsl_setup_acknowledged(mut self, acknowledged: bool) -> Self {
         self.edits
             .push(ConfigEdit::SetWindowsWslSetupAcknowledged(acknowledged));
@@ -511,9 +534,24 @@ impl ConfigEditsBuilder {
         self
     }
 
-    pub fn set_project_trusted<P: Into<PathBuf>>(mut self, project_path: P) -> Self {
-        self.edits
-            .push(ConfigEdit::SetProjectTrusted(project_path.into()));
+    pub fn set_project_trust_level<P: Into<PathBuf>>(
+        mut self,
+        project_path: P,
+        trust_level: TrustLevel,
+    ) -> Self {
+        self.edits.push(ConfigEdit::SetProjectTrustLevel {
+            path: project_path.into(),
+            level: trust_level,
+        });
+        self
+    }
+
+    /// Enable or disable a feature flag by key under the `[features]` table.
+    pub fn set_feature_enabled(mut self, key: &str, enabled: bool) -> Self {
+        self.edits.push(ConfigEdit::SetPath {
+            segments: vec!["features".to_string(), key.to_string()],
+            value: value(enabled),
+        });
         self
     }
 
@@ -551,7 +589,7 @@ mod tests {
             codex_home,
             None,
             &[ConfigEdit::SetModel {
-                model: Some("gpt-5-codex".to_string()),
+                model: Some("gpt-5.1-codex".to_string()),
                 effort: Some(ReasoningEffort::High),
             }],
         )
@@ -559,7 +597,7 @@ mod tests {
 
         let contents =
             std::fs::read_to_string(codex_home.join(CONFIG_TOML_FILE)).expect("read config");
-        let expected = r#"model = "gpt-5-codex"
+        let expected = r#"model = "gpt-5.1-codex"
 model_reasoning_effort = "high"
 "#;
         assert_eq!(contents, expected);
@@ -689,7 +727,7 @@ model = "o5-preview"
         std::fs::write(
             codex_home.join(CONFIG_TOML_FILE),
             r#"[profiles."team a"]
-model = "gpt-5-codex"
+model = "gpt-5.1-codex"
 "#,
         )
         .expect("seed");
@@ -770,6 +808,65 @@ existing = "value"
         let expected = r#"[notice]
 existing = "value"
 hide_rate_limit_model_nudge = true
+"#;
+        assert_eq!(contents, expected);
+    }
+    #[test]
+    fn blocking_set_hide_gpt5_1_migration_prompt_preserves_table() {
+        let tmp = tempdir().expect("tmpdir");
+        let codex_home = tmp.path();
+        std::fs::write(
+            codex_home.join(CONFIG_TOML_FILE),
+            r#"[notice]
+existing = "value"
+"#,
+        )
+        .expect("seed");
+        apply_blocking(
+            codex_home,
+            None,
+            &[ConfigEdit::SetNoticeHideModelMigrationPrompt(
+                "hide_gpt5_1_migration_prompt".to_string(),
+                true,
+            )],
+        )
+        .expect("persist");
+
+        let contents =
+            std::fs::read_to_string(codex_home.join(CONFIG_TOML_FILE)).expect("read config");
+        let expected = r#"[notice]
+existing = "value"
+hide_gpt5_1_migration_prompt = true
+"#;
+        assert_eq!(contents, expected);
+    }
+
+    #[test]
+    fn blocking_set_hide_gpt_5_1_codex_max_migration_prompt_preserves_table() {
+        let tmp = tempdir().expect("tmpdir");
+        let codex_home = tmp.path();
+        std::fs::write(
+            codex_home.join(CONFIG_TOML_FILE),
+            r#"[notice]
+existing = "value"
+"#,
+        )
+        .expect("seed");
+        apply_blocking(
+            codex_home,
+            None,
+            &[ConfigEdit::SetNoticeHideModelMigrationPrompt(
+                "hide_gpt-5.1-codex-max_migration_prompt".to_string(),
+                true,
+            )],
+        )
+        .expect("persist");
+
+        let contents =
+            std::fs::read_to_string(codex_home.join(CONFIG_TOML_FILE)).expect("read config");
+        let expected = r#"[notice]
+existing = "value"
+"hide_gpt-5.1-codex-max_migration_prompt" = true
 "#;
         assert_eq!(contents, expected);
     }
@@ -910,14 +1007,14 @@ B = \"2\"
         let codex_home = tmp.path().to_path_buf();
 
         ConfigEditsBuilder::new(&codex_home)
-            .set_model(Some("gpt-5-codex"), Some(ReasoningEffort::High))
+            .set_model(Some("gpt-5.1-codex"), Some(ReasoningEffort::High))
             .apply()
             .await
             .expect("persist");
 
         let contents =
             std::fs::read_to_string(codex_home.join(CONFIG_TOML_FILE)).expect("read config");
-        let expected = r#"model = "gpt-5-codex"
+        let expected = r#"model = "gpt-5.1-codex"
 model_reasoning_effort = "high"
 "#;
         assert_eq!(contents, expected);
@@ -939,11 +1036,11 @@ model_reasoning_effort = "low"
             std::fs::read_to_string(codex_home.join(CONFIG_TOML_FILE)).expect("read config");
         assert_eq!(contents, initial_expected);
 
-        let updated_expected = r#"model = "gpt-5-codex"
+        let updated_expected = r#"model = "gpt-5.1-codex"
 model_reasoning_effort = "high"
 "#;
         ConfigEditsBuilder::new(codex_home)
-            .set_model(Some("gpt-5-codex"), Some(ReasoningEffort::High))
+            .set_model(Some("gpt-5.1-codex"), Some(ReasoningEffort::High))
             .apply_blocking()
             .expect("persist update");
         contents = std::fs::read_to_string(codex_home.join(CONFIG_TOML_FILE)).expect("read config");

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

@@ -30,9 +30,11 @@ pub struct ConfigProfile {
     pub experimental_sandbox_command_assessment: Option<bool>,
     pub tools_web_search: Option<bool>,
     pub tools_view_image: Option<bool>,
+    pub max_active_subagents: Option<usize>,
     /// Optional feature toggles scoped to this profile.
     #[serde(default)]
     pub features: Option<crate::features::FeaturesToml>,
+    pub oss_provider: Option<String>,
 }
 
 impl From<ConfigProfile> for codex_app_server_protocol::Profile {

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

@@ -282,6 +282,14 @@ pub enum OtelHttpProtocol {
     Json,
 }
 
+#[derive(Deserialize, Debug, Clone, PartialEq, Default)]
+#[serde(rename_all = "kebab-case")]
+pub struct OtelTlsConfig {
+    pub ca_certificate: Option<PathBuf>,
+    pub client_certificate: Option<PathBuf>,
+    pub client_private_key: Option<PathBuf>,
+}
+
 /// Which OTEL exporter to use.
 #[derive(Deserialize, Debug, Clone, PartialEq)]
 #[serde(rename_all = "kebab-case")]
@@ -289,12 +297,18 @@ pub enum OtelExporterKind {
     None,
     OtlpHttp {
         endpoint: String,
+        #[serde(default)]
         headers: HashMap<String, String>,
         protocol: OtelHttpProtocol,
+        #[serde(default)]
+        tls: Option<OtelTlsConfig>,
     },
     OtlpGrpc {
         endpoint: String,
+        #[serde(default)]
         headers: HashMap<String, String>,
+        #[serde(default)]
+        tls: Option<OtelTlsConfig>,
     },
 }
 
@@ -338,7 +352,7 @@ pub enum Notifications {
 
 impl Default for Notifications {
     fn default() -> Self {
-        Self::Enabled(false)
+        Self::Enabled(true)
     }
 }
 
@@ -346,9 +360,18 @@ impl Default for Notifications {
 #[derive(Deserialize, Debug, Clone, PartialEq, Default)]
 pub struct Tui {
     /// Enable desktop notifications from the TUI when the terminal is unfocused.
-    /// Defaults to `false`.
+    /// Defaults to `true`.
     #[serde(default)]
     pub notifications: Notifications,
+
+    /// Enable animations (welcome screen, shimmer effects, spinners).
+    /// Defaults to `true`.
+    #[serde(default = "default_true")]
+    pub animations: bool,
+}
+
+const fn default_true() -> bool {
+    true
 }
 
 /// Settings for notices we display to users via the tui and app-server clients
@@ -362,6 +385,11 @@ pub struct Notice {
     pub hide_world_writable_warning: Option<bool>,
     /// Tracks whether the user opted out of the rate limit model switch reminder.
     pub hide_rate_limit_model_nudge: Option<bool>,
+    /// Tracks whether the user has seen the model migration prompt
+    pub hide_gpt5_1_migration_prompt: Option<bool>,
+    /// Tracks whether the user has seen the gpt-5.1-codex-max migration prompt
+    #[serde(rename = "hide_gpt-5.1-codex-max_migration_prompt")]
+    pub hide_gpt_5_1_codex_max_migration_prompt: Option<bool>,
 }
 
 impl Notice {

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

@@ -1,13 +1,15 @@
+use crate::codex::TurnContext;
+use crate::context_manager::normalize;
+use crate::truncate::TruncationPolicy;
+use crate::truncate::approx_token_count;
+use crate::truncate::truncate_function_output_items_with_policy;
+use crate::truncate::truncate_text;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::protocol::TokenUsage;
 use codex_protocol::protocol::TokenUsageInfo;
 use std::ops::Deref;
 
-use crate::context_manager::normalize;
-use crate::context_manager::truncate::format_output_for_model_body;
-use crate::context_manager::truncate::globally_truncate_function_output_items;
-
 /// Transcript of conversation history
 #[derive(Debug, Clone, Default)]
 pub(crate) struct ContextManager {
@@ -28,6 +30,10 @@ impl ContextManager {
         self.token_info.clone()
     }
 
+    pub(crate) fn set_token_info(&mut self, info: Option<TokenUsageInfo>) {
+        self.token_info = info;
+    }
+
     pub(crate) fn set_token_usage_full(&mut self, context_window: i64) {
         match &mut self.token_info {
             Some(info) => info.fill_to_context_window(context_window),
@@ -38,7 +44,7 @@ impl ContextManager {
     }
 
     /// `items` is ordered from oldest to newest.
-    pub(crate) fn record_items<I>(&mut self, items: I)
+    pub(crate) fn record_items<I>(&mut self, items: I, policy: TruncationPolicy)
     where
         I: IntoIterator,
         I::Item: std::ops::Deref<Target = ResponseItem>,
@@ -50,7 +56,7 @@ impl ContextManager {
                 continue;
             }
 
-            let processed = Self::process_item(&item);
+            let processed = self.process_item(item_ref, policy);
             self.items.push(processed);
         }
     }
@@ -68,6 +74,23 @@ impl ContextManager {
         history
     }
 
+    // Estimate token usage using byte-based heuristics from the truncation helpers.
+    // This is a coarse lower bound, not a tokenizer-accurate count.
+    pub(crate) fn estimate_token_count(&self, turn_context: &TurnContext) -> Option<i64> {
+        let model_family = turn_context.client.get_model_family();
+        let base_tokens =
+            i64::try_from(approx_token_count(model_family.base_instructions.as_str()))
+                .unwrap_or(i64::MAX);
+
+        let items_tokens = self.items.iter().fold(0i64, |acc, item| {
+            let serialized = serde_json::to_string(item).unwrap_or_default();
+            let item_tokens = i64::try_from(approx_token_count(&serialized)).unwrap_or(i64::MAX);
+            acc.saturating_add(item_tokens)
+        });
+
+        Some(base_tokens.saturating_add(items_tokens))
+    }
+
     pub(crate) fn remove_first_item(&mut self) {
         if !self.items.is_empty() {
             // Remove the oldest item (front of the list). Items are ordered from
@@ -116,14 +139,18 @@ impl ContextManager {
         items.retain(|item| !matches!(item, ResponseItem::GhostSnapshot { .. }));
     }
 
-    fn process_item(item: &ResponseItem) -> ResponseItem {
+    fn process_item(&self, item: &ResponseItem, policy: TruncationPolicy) -> ResponseItem {
+        let policy_with_serialization_budget = policy.mul(1.2);
         match item {
             ResponseItem::FunctionCallOutput { call_id, output } => {
-                let truncated = format_output_for_model_body(output.content.as_str());
-                let truncated_items = output
-                    .content_items
-                    .as_ref()
-                    .map(|items| globally_truncate_function_output_items(items));
+                let truncated =
+                    truncate_text(output.content.as_str(), policy_with_serialization_budget);
+                let truncated_items = output.content_items.as_ref().map(|items| {
+                    truncate_function_output_items_with_policy(
+                        items,
+                        policy_with_serialization_budget,
+                    )
+                });
                 ResponseItem::FunctionCallOutput {
                     call_id: call_id.clone(),
                     output: FunctionCallOutputPayload {
@@ -134,7 +161,7 @@ impl ContextManager {
                 }
             }
             ResponseItem::CustomToolCallOutput { call_id, output } => {
-                let truncated = format_output_for_model_body(output);
+                let truncated = truncate_text(output, policy_with_serialization_budget);
                 ResponseItem::CustomToolCallOutput {
                     call_id: call_id.clone(),
                     output: truncated,
@@ -146,6 +173,7 @@ impl ContextManager {
             | ResponseItem::FunctionCall { .. }
             | ResponseItem::WebSearchCall { .. }
             | ResponseItem::CustomToolCall { .. }
+            | ResponseItem::CompactionSummary { .. }
             | ResponseItem::GhostSnapshot { .. }
             | ResponseItem::Other => item.clone(),
         }
@@ -163,7 +191,8 @@ fn is_api_message(message: &ResponseItem) -> bool {
         | ResponseItem::CustomToolCallOutput { .. }
         | ResponseItem::LocalShellCall { .. }
         | ResponseItem::Reasoning { .. }
-        | ResponseItem::WebSearchCall { .. } => true,
+        | ResponseItem::WebSearchCall { .. }
+        | ResponseItem::CompactionSummary { .. } => true,
         ResponseItem::GhostSnapshot { .. } => false,
         ResponseItem::Other => false,
     }

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

@@ -1,8 +1,8 @@
 use super::*;
-use crate::context_manager::truncate;
+use crate::truncate;
+use crate::truncate::TruncationPolicy;
 use codex_git::GhostCommit;
 use codex_protocol::models::ContentItem;
-use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::LocalShellAction;
 use codex_protocol::models::LocalShellExecAction;
@@ -12,6 +12,9 @@ use codex_protocol::models::ReasoningItemReasoningSummary;
 use pretty_assertions::assert_eq;
 use regex_lite::Regex;
 
+const EXEC_FORMAT_MAX_BYTES: usize = 10_000;
+const EXEC_FORMAT_MAX_TOKENS: usize = 2_500;
+
 fn assistant_msg(text: &str) -> ResponseItem {
     ResponseItem::Message {
         id: None,
@@ -24,7 +27,9 @@ fn assistant_msg(text: &str) -> ResponseItem {
 
 fn create_history_with_items(items: Vec<ResponseItem>) -> ContextManager {
     let mut h = ContextManager::new();
-    h.record_items(items.iter());
+    // Use a generous but fixed token budget; tests only rely on truncation
+    // behavior, not on a specific model's token limit.
+    h.record_items(items.iter(), TruncationPolicy::Tokens(10_000));
     h
 }
 
@@ -51,9 +56,14 @@ fn reasoning_msg(text: &str) -> ResponseItem {
     }
 }
 
+fn truncate_exec_output(content: &str) -> String {
+    truncate::truncate_text(content, TruncationPolicy::Tokens(EXEC_FORMAT_MAX_TOKENS))
+}
+
 #[test]
 fn filters_non_api_messages() {
     let mut h = ContextManager::default();
+    let policy = TruncationPolicy::Tokens(10_000);
     // System message is not API messages; Other is ignored.
     let system = ResponseItem::Message {
         id: None,
@@ -63,12 +73,12 @@ fn filters_non_api_messages() {
         }],
     };
     let reasoning = reasoning_msg("thinking...");
-    h.record_items([&system, &reasoning, &ResponseItem::Other]);
+    h.record_items([&system, &reasoning, &ResponseItem::Other], policy);
 
     // User and assistant should be retained.
     let u = user_msg("hi");
     let a = assistant_msg("hello");
-    h.record_items([&u, &a]);
+    h.record_items([&u, &a], policy);
 
     let items = h.contents();
     assert_eq!(
@@ -222,7 +232,7 @@ fn normalization_retains_local_shell_outputs() {
         ResponseItem::FunctionCallOutput {
             call_id: "shell-1".to_string(),
             output: FunctionCallOutputPayload {
-                content: "ok".to_string(),
+                content: "Total output lines: 1\n\nok".to_string(),
                 ..Default::default()
             },
         },
@@ -236,6 +246,9 @@ fn normalization_retains_local_shell_outputs() {
 #[test]
 fn record_items_truncates_function_call_output_content() {
     let mut history = ContextManager::new();
+    // Any reasonably small token budget works; the test only cares that
+    // truncation happens and the marker is present.
+    let policy = TruncationPolicy::Tokens(1_000);
     let long_line = "a very long line to trigger truncation\n";
     let long_output = long_line.repeat(2_500);
     let item = ResponseItem::FunctionCallOutput {
@@ -247,15 +260,20 @@ fn record_items_truncates_function_call_output_content() {
         },
     };
 
-    history.record_items([&item]);
+    history.record_items([&item], policy);
 
     assert_eq!(history.items.len(), 1);
     match &history.items[0] {
         ResponseItem::FunctionCallOutput { output, .. } => {
             assert_ne!(output.content, long_output);
             assert!(
-                output.content.starts_with("Total output lines:"),
-                "expected truncated summary, got {}",
+                output.content.contains("tokens truncated"),
+                "expected token-based truncation marker, got {}",
+                output.content
+            );
+            assert!(
+                output.content.contains("tokens truncated"),
+                "expected truncation marker, got {}",
                 output.content
             );
         }
@@ -266,6 +284,7 @@ fn record_items_truncates_function_call_output_content() {
 #[test]
 fn record_items_truncates_custom_tool_call_output_content() {
     let mut history = ContextManager::new();
+    let policy = TruncationPolicy::Tokens(1_000);
     let line = "custom output that is very long\n";
     let long_output = line.repeat(2_500);
     let item = ResponseItem::CustomToolCallOutput {
@@ -273,23 +292,50 @@ fn record_items_truncates_custom_tool_call_output_content() {
         output: long_output.clone(),
     };
 
-    history.record_items([&item]);
+    history.record_items([&item], policy);
 
     assert_eq!(history.items.len(), 1);
     match &history.items[0] {
         ResponseItem::CustomToolCallOutput { output, .. } => {
             assert_ne!(output, &long_output);
             assert!(
-                output.starts_with("Total output lines:"),
-                "expected truncated summary, got {output}"
+                output.contains("tokens truncated"),
+                "expected token-based truncation marker, got {output}"
+            );
+            assert!(
+                output.contains("tokens truncated") || output.contains("bytes truncated"),
+                "expected truncation marker, got {output}"
             );
         }
         other => panic!("unexpected history item: {other:?}"),
     }
 }
 
-fn assert_truncated_message_matches(message: &str, line: &str, total_lines: usize) {
-    let pattern = truncated_message_pattern(line, total_lines);
+#[test]
+fn record_items_respects_custom_token_limit() {
+    let mut history = ContextManager::new();
+    let policy = TruncationPolicy::Tokens(10);
+    let long_output = "tokenized content repeated many times ".repeat(200);
+    let item = ResponseItem::FunctionCallOutput {
+        call_id: "call-custom-limit".to_string(),
+        output: FunctionCallOutputPayload {
+            content: long_output,
+            success: Some(true),
+            ..Default::default()
+        },
+    };
+
+    history.record_items([&item], policy);
+
+    let stored = match &history.items[0] {
+        ResponseItem::FunctionCallOutput { output, .. } => output,
+        other => panic!("unexpected history item: {other:?}"),
+    };
+    assert!(stored.content.contains("tokens truncated"));
+}
+
+fn assert_truncated_message_matches(message: &str, line: &str, expected_removed: usize) {
+    let pattern = truncated_message_pattern(line);
     let regex = Regex::new(&pattern).unwrap_or_else(|err| {
         panic!("failed to compile regex {pattern}: {err}");
     });
@@ -301,26 +347,22 @@ fn assert_truncated_message_matches(message: &str, line: &str, total_lines: usiz
         .expect("missing body capture")
         .as_str();
     assert!(
-        body.len() <= truncate::MODEL_FORMAT_MAX_BYTES,
+        body.len() <= EXEC_FORMAT_MAX_BYTES,
         "body exceeds byte limit: {} bytes",
         body.len()
     );
+    let removed: usize = captures
+        .name("removed")
+        .expect("missing removed capture")
+        .as_str()
+        .parse()
+        .unwrap_or_else(|err| panic!("invalid removed tokens: {err}"));
+    assert_eq!(removed, expected_removed, "mismatched removed token count");
 }
 
-fn truncated_message_pattern(line: &str, total_lines: usize) -> String {
-    let head_take = truncate::MODEL_FORMAT_HEAD_LINES.min(total_lines);
-    let tail_take = truncate::MODEL_FORMAT_TAIL_LINES.min(total_lines.saturating_sub(head_take));
-    let omitted = total_lines.saturating_sub(head_take + tail_take);
+fn truncated_message_pattern(line: &str) -> String {
     let escaped_line = regex_lite::escape(line);
-    if omitted == 0 {
-        return format!(
-            r"(?s)^Total output lines: {total_lines}\n\n(?P<body>{escaped_line}.*\n\[\.{{3}} output truncated to fit {max_bytes} bytes \.{{3}}]\n\n.*)$",
-            max_bytes = truncate::MODEL_FORMAT_MAX_BYTES,
-        );
-    }
-    format!(
-        r"(?s)^Total output lines: {total_lines}\n\n(?P<body>{escaped_line}.*\n\[\.{{3}} omitted {omitted} of {total_lines} lines \.{{3}}]\n\n.*)$",
-    )
+    format!(r"(?s)^(?P<body>{escaped_line}.*?)(?:\r?)?…(?P<removed>\d+) tokens truncated…(?:.*)?$")
 }
 
 #[test]
@@ -328,27 +370,18 @@ fn format_exec_output_truncates_large_error() {
     let line = "very long execution error line that should trigger truncation\n";
     let large_error = line.repeat(2_500); // way beyond both byte and line limits
 
-    let truncated = truncate::format_output_for_model_body(&large_error);
+    let truncated = truncate_exec_output(&large_error);
 
-    let total_lines = large_error.lines().count();
-    assert_truncated_message_matches(&truncated, line, total_lines);
+    assert_truncated_message_matches(&truncated, line, 36250);
     assert_ne!(truncated, large_error);
 }
 
 #[test]
 fn format_exec_output_marks_byte_truncation_without_omitted_lines() {
-    let long_line = "a".repeat(truncate::MODEL_FORMAT_MAX_BYTES + 50);
-    let truncated = truncate::format_output_for_model_body(&long_line);
-
+    let long_line = "a".repeat(EXEC_FORMAT_MAX_BYTES + 10000);
+    let truncated = truncate_exec_output(&long_line);
     assert_ne!(truncated, long_line);
-    let marker_line = format!(
-        "[... output truncated to fit {} bytes ...]",
-        truncate::MODEL_FORMAT_MAX_BYTES
-    );
-    assert!(
-        truncated.contains(&marker_line),
-        "missing byte truncation marker: {truncated}"
-    );
+    assert_truncated_message_matches(&truncated, "a", 2500);
     assert!(
         !truncated.contains("omitted"),
         "line omission marker should not appear when no lines were dropped: {truncated}"
@@ -358,31 +391,25 @@ fn format_exec_output_marks_byte_truncation_without_omitted_lines() {
 #[test]
 fn format_exec_output_returns_original_when_within_limits() {
     let content = "example output\n".repeat(10);
-
-    assert_eq!(truncate::format_output_for_model_body(&content), content);
+    assert_eq!(truncate_exec_output(&content), content);
 }
 
 #[test]
 fn format_exec_output_reports_omitted_lines_and_keeps_head_and_tail() {
-    let total_lines = truncate::MODEL_FORMAT_MAX_LINES + 100;
+    let total_lines = 2_000;
+    let filler = "x".repeat(64);
     let content: String = (0..total_lines)
-        .map(|idx| format!("line-{idx}\n"))
+        .map(|idx| format!("line-{idx}-{filler}\n"))
         .collect();
 
-    let truncated = truncate::format_output_for_model_body(&content);
-    let omitted = total_lines - truncate::MODEL_FORMAT_MAX_LINES;
-    let expected_marker = format!("[... omitted {omitted} of {total_lines} lines ...]");
-
+    let truncated = truncate_exec_output(&content);
+    assert_truncated_message_matches(&truncated, "line-0-", 34_723);
     assert!(
-        truncated.contains(&expected_marker),
-        "missing omitted marker: {truncated}"
-    );
-    assert!(
-        truncated.contains("line-0\n"),
+        truncated.contains("line-0-"),
         "expected head line to remain: {truncated}"
     );
 
-    let last_line = format!("line-{}\n", total_lines - 1);
+    let last_line = format!("line-{}-", total_lines - 1);
     assert!(
         truncated.contains(&last_line),
         "expected tail line to remain: {truncated}"
@@ -391,97 +418,15 @@ fn format_exec_output_reports_omitted_lines_and_keeps_head_and_tail() {
 
 #[test]
 fn format_exec_output_prefers_line_marker_when_both_limits_exceeded() {
-    let total_lines = truncate::MODEL_FORMAT_MAX_LINES + 42;
+    let total_lines = 300;
     let long_line = "x".repeat(256);
     let content: String = (0..total_lines)
         .map(|idx| format!("line-{idx}-{long_line}\n"))
         .collect();
 
-    let truncated = truncate::format_output_for_model_body(&content);
+    let truncated = truncate_exec_output(&content);
 
-    assert!(
-        truncated.contains("[... omitted 42 of 298 lines ...]"),
-        "expected omitted marker when line count exceeds limit: {truncated}"
-    );
-    assert!(
-        !truncated.contains("output truncated to fit"),
-        "line omission marker should take precedence over byte marker: {truncated}"
-    );
-}
-
-#[test]
-fn truncates_across_multiple_under_limit_texts_and_reports_omitted() {
-    // Arrange: several text items, none exceeding per-item limit, but total exceeds budget.
-    let budget = truncate::MODEL_FORMAT_MAX_BYTES;
-    let t1_len = (budget / 2).saturating_sub(10);
-    let t2_len = (budget / 2).saturating_sub(10);
-    let remaining_after_t1_t2 = budget.saturating_sub(t1_len + t2_len);
-    let t3_len = 50; // gets truncated to remaining_after_t1_t2
-    let t4_len = 5; // omitted
-    let t5_len = 7; // omitted
-
-    let t1 = "a".repeat(t1_len);
-    let t2 = "b".repeat(t2_len);
-    let t3 = "c".repeat(t3_len);
-    let t4 = "d".repeat(t4_len);
-    let t5 = "e".repeat(t5_len);
-
-    let item = ResponseItem::FunctionCallOutput {
-        call_id: "call-omit".to_string(),
-        output: FunctionCallOutputPayload {
-            content: "irrelevant".to_string(),
-            content_items: Some(vec![
-                FunctionCallOutputContentItem::InputText { text: t1 },
-                FunctionCallOutputContentItem::InputText { text: t2 },
-                FunctionCallOutputContentItem::InputImage {
-                    image_url: "img:mid".to_string(),
-                },
-                FunctionCallOutputContentItem::InputText { text: t3 },
-                FunctionCallOutputContentItem::InputText { text: t4 },
-                FunctionCallOutputContentItem::InputText { text: t5 },
-            ]),
-            success: Some(true),
-        },
-    };
-
-    let mut history = ContextManager::new();
-    history.record_items([&item]);
-    assert_eq!(history.items.len(), 1);
-    let json = serde_json::to_value(&history.items[0]).expect("serialize to json");
-
-    let output = json
-        .get("output")
-        .expect("output field")
-        .as_array()
-        .expect("array output");
-
-    // Expect: t1 (full), t2 (full), image, t3 (truncated), summary mentioning 2 omitted.
-    assert_eq!(output.len(), 5);
-
-    let first = output[0].as_object().expect("first obj");
-    assert_eq!(first.get("type").unwrap(), "input_text");
-    let first_text = first.get("text").unwrap().as_str().unwrap();
-    assert_eq!(first_text.len(), t1_len);
-
-    let second = output[1].as_object().expect("second obj");
-    assert_eq!(second.get("type").unwrap(), "input_text");
-    let second_text = second.get("text").unwrap().as_str().unwrap();
-    assert_eq!(second_text.len(), t2_len);
-
-    assert_eq!(
-        output[2],
-        serde_json::json!({"type": "input_image", "image_url": "img:mid"})
-    );
-
-    let fourth = output[3].as_object().expect("fourth obj");
-    assert_eq!(fourth.get("type").unwrap(), "input_text");
-    let fourth_text = fourth.get("text").unwrap().as_str().unwrap();
-    assert_eq!(fourth_text.len(), remaining_after_t1_t2);
-
-    let summary = output[4].as_object().expect("summary obj");
-    assert_eq!(summary.get("type").unwrap(), "input_text");
-    let summary_text = summary.get("text").unwrap().as_str().unwrap();
-    assert!(summary_text.contains("omitted 2 text items"));
+    assert_truncated_message_matches(&truncated, "line-0-", 17_423);
 }
 
 //TODO(aibrahim): run CI in release mode.

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

@@ -1,6 +1,4 @@
 mod history;
 mod normalize;
-mod truncate;
 
 pub(crate) use history::ContextManager;
-pub(crate) use truncate::format_output_for_model_body;

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

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