websearch

Add a file that overrides `AGENTS.md` but is not versioned (for local
devs)

.github/ISSUE_TEMPLATE/2-bug-report.yml

@@ -20,6 +20,14 @@ body:
     attributes:
       label: What version of Codex is running?
       description: Copy the output of `codex --version`
+    validations:
+      required: true
+  - type: input
+    id: plan
+    attributes:
+      label: What subscription do you have?
+    validations:
+      required: true
   - type: input
     id: model
     attributes:
@@ -32,11 +40,18 @@ body:
       description: |
         For MacOS and Linux: copy the output of `uname -mprs`
         For Windows: copy the output of `"$([Environment]::OSVersion | ForEach-Object VersionString) $(if ([Environment]::Is64BitOperatingSystem) { "x64" } else { "x86" })"` in the PowerShell console
+  - type: textarea
+    id: actual
+    attributes:
+      label: What issue are you seeing?
+      description: Please include the full error messages and prompts with PII redacted. If possible, please provide text instead of a screenshot. 
+    validations:
+      required: true
   - type: textarea
     id: steps
     attributes:
       label: What steps can reproduce the bug?
-      description: Explain the bug and provide a code snippet that can reproduce it.
+      description: Explain the bug and provide a code snippet that can reproduce it. Please include session id, token limit usage, context window usage if applicable.
     validations:
       required: true
   - type: textarea
@@ -44,11 +59,6 @@ body:
     attributes:
       label: What is the expected behavior?
       description: If possible, please provide text instead of a screenshot.
-  - type: textarea
-    id: actual
-    attributes:
-      label: What do you see instead?
-      description: If possible, please provide text instead of a screenshot.
   - type: textarea
     id: notes
     attributes:

.github/ISSUE_TEMPLATE/4-feature-request.yml

@@ -2,7 +2,6 @@ name: 🎁 Feature Request
 description: Propose a new feature for Codex
 labels:
   - enhancement
-  - needs triage
 body:
   - type: markdown
     attributes:
@@ -19,11 +18,6 @@ body:
       label: What feature would you like to see?
     validations:
       required: true
-  - type: textarea
-    id: author
-    attributes:
-      label: Are you interested in implementing this feature?
-      description: Please wait for acknowledgement before implementing or opening a PR.
   - type: textarea
     id: notes
     attributes:

.github/ISSUE_TEMPLATE/5-vs-code-extension.yml

@@ -14,11 +14,21 @@ body:
     id: version
     attributes:
       label: What version of the VS Code extension are you using?
+    validations:
+      required: true
+  - type: input
+    id: plan
+    attributes:
+      label: What subscription do you have?
+    validations:
+      required: true
   - type: input
     id: ide
     attributes:
       label: Which IDE are you using?
       description: Like `VS Code`, `Cursor`, `Windsurf`, etc.
+    validations:
+      required: true
   - type: input
     id: platform
     attributes:
@@ -26,11 +36,18 @@ body:
       description: |
         For MacOS and Linux: copy the output of `uname -mprs`
         For Windows: copy the output of `"$([Environment]::OSVersion | ForEach-Object VersionString) $(if ([Environment]::Is64BitOperatingSystem) { "x64" } else { "x86" })"` in the PowerShell console
+  - type: textarea
+    id: actual
+    attributes:
+      label: What issue are you seeing?
+      description: Please include the full error messages and prompts with PII redacted. If possible, please provide text instead of a screenshot. 
+    validations:
+      required: true
   - type: textarea
     id: steps
     attributes:
       label: What steps can reproduce the bug?
-      description: Explain the bug and provide a code snippet that can reproduce it.
+      description: Explain the bug and provide a code snippet that can reproduce it. Please include session id, token limit usage, context window usage if applicable.
     validations:
       required: true
   - type: textarea
@@ -38,11 +55,6 @@ body:
     attributes:
       label: What is the expected behavior?
       description: If possible, please provide text instead of a screenshot.
-  - type: textarea
-    id: actual
-    attributes:
-      label: What do you see instead?
-      description: If possible, please provide text instead of a screenshot.
   - type: textarea
     id: notes
     attributes:

.github/prompts/issue-deduplicator.txt

@@ -0,0 +1,18 @@
+You are an assistant that triages new GitHub issues by identifying potential duplicates.
+
+You will receive the following JSON files located in the current working directory:
+- `codex-current-issue.json`: JSON object describing the newly created issue (fields: number, title, body).
+- `codex-existing-issues.json`: JSON array of recent issues (each element includes number, title, body, createdAt).
+
+Instructions:
+- Load both files as JSON and review their contents carefully. The codex-existing-issues.json file is large, ensure you explore all of it.
+- Compare the current issue against the existing issues to find up to five that appear to describe the same underlying problem or request.
+- Only consider an issue a potential duplicate if there is a clear overlap in symptoms, feature requests, reproduction steps, or error messages.
+- Prioritize newer issues when similarity is comparable.
+- Ignore pull requests and issues whose similarity is tenuous.
+- When unsure, prefer returning fewer matches.
+
+Output requirements:
+- Respond with a JSON array of issue numbers (integers), ordered from most likely duplicate to least.
+- Include at most five numbers.
+- If you find no plausible duplicates, respond with `[]`.

.github/prompts/issue-labeler.txt

@@ -0,0 +1,26 @@
+You are an assistant that reviews GitHub issues for the repository.
+
+Your job is to choose the most appropriate existing labels for the issue described later in this prompt.
+Follow these rules:
+- Only pick labels out of the list below.
+- Prefer a small set of precise labels over many broad ones.
+- If none of the labels fit, respond with an empty JSON array: []
+- Output must be a JSON array of label names (strings) with no additional commentary.
+
+Labels to apply:
+1. bug — Reproducible defects in Codex products (CLI, VS Code extension, web, auth).
+2. enhancement — Feature requests or usability improvements that ask for new capabilities, better ergonomics, or quality-of-life tweaks.
+3. extension — VS Code (or other IDE) extension-specific issues.
+4. windows-os — Bugs or friction specific to Windows environments (PowerShell behavior, path handling, copy/paste, OS-specific auth or tooling failures).
+5. mcp — Topics involving Model Context Protocol servers/clients.
+6. codex-web — Issues targeting the Codex web UI/Cloud experience.
+8. azure — Problems or requests tied to Azure OpenAI deployments.
+9. documentation — Updates or corrections needed in docs/README/config references (broken links, missing examples, outdated keys, clarification requests).
+10. model-behavior — Undesirable LLM behavior: forgetting goals, refusing work, hallucinating environment details, quota misreports, or other reasoning/performance anomalies.
+
+Issue information is available in environment variables:
+
+ISSUE_NUMBER
+ISSUE_TITLE
+ISSUE_BODY
+REPO_FULL_NAME

.github/workflows/ci.yml

@@ -60,3 +60,6 @@ jobs:
         run: ./scripts/asciicheck.py codex-cli/README.md
       - name: Check codex-cli/README ToC
         run: python3 scripts/readme_toc.py codex-cli/README.md
+
+      - name: Prettier (run `pnpm run format:fix` to fix)
+        run: pnpm run format

.github/workflows/issue-deduplicator.yml

@@ -0,0 +1,140 @@
+name: Issue Deduplicator
+
+on:
+  issues:
+    types:
+      - opened
+      - labeled
+
+jobs:
+  gather-duplicates:
+    name: Identify potential duplicates
+    if: ${{ github.event.action == 'opened' || (github.event.action == 'labeled' && github.event.label.name == 'codex-deduplicate') }}
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+    outputs:
+      codex_output: ${{ steps.codex.outputs.final-message }}
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Prepare Codex inputs
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          set -eo pipefail
+
+          CURRENT_ISSUE_FILE=codex-current-issue.json
+          EXISTING_ISSUES_FILE=codex-existing-issues.json
+
+          gh issue list --repo "${{ github.repository }}" \
+            --json number,title,body,createdAt \
+            --limit 1000 \
+            --state all \
+            --search "sort:created-desc" \
+            | jq '.' \
+            > "$EXISTING_ISSUES_FILE"
+
+          gh issue view "${{ github.event.issue.number }}" \
+            --repo "${{ github.repository }}" \
+            --json number,title,body \
+            | jq '.' \
+            > "$CURRENT_ISSUE_FILE"
+
+      - id: codex
+        uses: openai/codex-action@main
+        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.
+
+            You will receive the following JSON files located in the current working directory:
+            - `codex-current-issue.json`: JSON object describing the newly created issue (fields: number, title, body).
+            - `codex-existing-issues.json`: JSON array of recent issues (each element includes number, title, body, createdAt).
+
+            Instructions:
+            - Compare the current issue against the existing issues to find up to five that appear to describe the same underlying problem or request.
+            - Focus on the underlying intent and context of each issue—such as reported symptoms, feature requests, reproduction steps, or error messages—rather than relying solely on string similarity or synthetic metrics.
+            - After your analysis, validate your results in 1-2 lines explaining your decision to return the selected matches.
+            - When unsure, prefer returning fewer matches.
+            - Include at most five numbers.
+
+          output-schema: |
+            {
+              "type": "object",
+              "properties": {
+                "issues": {
+                  "type": "array",
+                  "items": {
+                    "type": "string"
+                  }
+                },
+                "reason": { "type": "string" }
+              },
+              "required": ["issues", "reason"],
+              "additionalProperties": false
+            }
+
+  comment-on-issue:
+    name: Comment with potential duplicates
+    needs: gather-duplicates
+    if: ${{ needs.gather-duplicates.result != 'skipped' }}
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      issues: write
+    steps:
+      - name: Comment on issue
+        uses: actions/github-script@v7
+        env:
+          CODEX_OUTPUT: ${{ needs.gather-duplicates.outputs.codex_output }}
+        with:
+          github-token: ${{ github.token }}
+          script: |
+            const raw = process.env.CODEX_OUTPUT ?? '';
+            let parsed;
+            try {
+              parsed = JSON.parse(raw);
+            } catch (error) {
+              core.info(`Codex output was not valid JSON. Raw output: ${raw}`);
+              core.info(`Parse error: ${error.message}`);
+              return;
+            }
+
+            const issues = Array.isArray(parsed?.issues) ? parsed.issues : [];
+            const currentIssueNumber = String(context.payload.issue.number);
+
+            console.log(`Current issue number: ${currentIssueNumber}`);
+            console.log(issues);
+
+            const filteredIssues = issues.filter((value) => String(value) !== currentIssueNumber);
+
+            if (filteredIssues.length === 0) {
+              core.info('Codex reported no potential duplicates.');
+              return;
+            }
+
+            const lines = [
+              'Potential duplicates detected. Please review them and close your issue if it is a duplicate.',
+              '',
+              ...filteredIssues.map((value) => `- #${String(value)}`),
+              '',
+              '*Powered by [Codex Action](https://github.com/openai/codex-action)*'];
+
+            await github.rest.issues.createComment({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              issue_number: context.payload.issue.number,
+              body: lines.join("\n"),
+            });
+
+      - name: Remove codex-deduplicate label
+        if: ${{ always() && github.event.action == 'labeled' && github.event.label.name == 'codex-deduplicate' }}
+        env:
+          GH_TOKEN: ${{ github.token }}
+          GH_REPO: ${{ github.repository }}
+        run: |
+          gh issue edit "${{ github.event.issue.number }}" --remove-label codex-deduplicate || true
+          echo "Attempted to remove label: codex-deduplicate"

.github/workflows/issue-labeler.yml

@@ -0,0 +1,115 @@
+name: Issue Labeler
+
+on:
+  issues:
+    types:
+      - opened
+      - labeled
+
+jobs:
+  gather-labels:
+    name: Generate label suggestions
+    if: ${{ github.event.action == 'opened' || (github.event.action == 'labeled' && github.event.label.name == 'codex-label') }}
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+    outputs:
+      codex_output: ${{ steps.codex.outputs.final-message }}
+    steps:
+      - uses: actions/checkout@v4
+
+      - id: codex
+        uses: openai/codex-action@main
+        with:
+          openai-api-key: ${{ secrets.CODEX_OPENAI_API_KEY }}
+          allow-users: "*"
+          prompt: |
+            You are an assistant that reviews GitHub issues for the repository.
+
+            Your job is to choose the most appropriate existing labels for the issue described later in this prompt.
+            Follow these rules:
+            - Only pick labels out of the list below.
+            - Prefer a small set of precise labels over many broad ones.
+
+            Labels to apply:
+            1. bug — Reproducible defects in Codex products (CLI, VS Code extension, web, auth).
+            2. enhancement — Feature requests or usability improvements that ask for new capabilities, better ergonomics, or quality-of-life tweaks.
+            3. extension — VS Code (or other IDE) extension-specific issues.
+            4. windows-os — Bugs or friction specific to Windows environments (always when PowerShell is mentioned, path handling, copy/paste, OS-specific auth or tooling failures).
+            5. mcp — Topics involving Model Context Protocol servers/clients.
+            6. codex-web — Issues targeting the Codex web UI/Cloud experience.
+            8. azure — Problems or requests tied to Azure OpenAI deployments.
+            9. documentation — Updates or corrections needed in docs/README/config references (broken links, missing examples, outdated keys, clarification requests).
+            10. model-behavior — Undesirable LLM behavior: forgetting goals, refusing work, hallucinating environment details, quota misreports, or other reasoning/performance anomalies.
+
+            Issue number: ${{ github.event.issue.number }}
+
+            Issue title:
+            ${{ github.event.issue.title }}
+
+            Issue body:
+            ${{ github.event.issue.body }}
+
+            Repository full name:
+            ${{ github.repository }}
+
+          output-schema: |
+            {
+              "type": "object",
+              "properties": {
+                "labels": {
+                  "type": "array",
+                  "items": {
+                    "type": "string"
+                  }
+                }
+              },
+              "required": ["labels"],
+              "additionalProperties": false
+            }
+
+  apply-labels:
+    name: Apply labels from Codex output
+    needs: gather-labels
+    if: ${{ needs.gather-labels.result != 'skipped' }}
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      issues: write
+    env:
+      GH_TOKEN: ${{ github.token }}
+      GH_REPO: ${{ github.repository }}
+      ISSUE_NUMBER: ${{ github.event.issue.number }}
+      CODEX_OUTPUT: ${{ needs.gather-labels.outputs.codex_output }}
+    steps:
+      - name: Apply labels
+        run: |
+          json=${CODEX_OUTPUT//$'\r'/}
+          if [ -z "$json" ]; then
+            echo "Codex produced no output. Skipping label application."
+            exit 0
+          fi
+
+          if ! printf '%s' "$json" | jq -e 'type == "object" and (.labels | type == "array")' >/dev/null 2>&1; then
+            echo "Codex output did not include a labels array. Raw output: $json"
+            exit 0
+          fi
+
+          labels=$(printf '%s' "$json" | jq -r '.labels[] | tostring')
+          if [ -z "$labels" ]; then
+            echo "Codex returned an empty array. Nothing to do."
+            exit 0
+          fi
+
+          cmd=(gh issue edit "$ISSUE_NUMBER")
+          while IFS= read -r label; do
+            cmd+=(--add-label "$label")
+          done <<< "$labels"
+
+          "${cmd[@]}" || true
+
+      - name: Remove codex-label trigger
+        if: ${{ always() && github.event.action == 'labeled' && github.event.label.name == 'codex-label' }}
+        run: |
+          gh issue edit "$ISSUE_NUMBER" --remove-label codex-label || true
+          echo "Attempted to remove label: codex-label"

.github/workflows/rust-ci.yml

@@ -148,15 +148,26 @@ jobs:
           targets: ${{ matrix.target }}
           components: clippy
 
-      - uses: actions/cache@v4
+      # Explicit cache restore: split cargo home vs target, so we can
+      # avoid caching the large target dir on the gnu-dev job.
+      - name: Restore cargo home cache
+        id: cache_cargo_home_restore
+        uses: actions/cache/restore@v4
         with:
           path: |
             ~/.cargo/bin/
             ~/.cargo/registry/index/
             ~/.cargo/registry/cache/
             ~/.cargo/git/db/
-            ${{ github.workspace }}/codex-rs/target/
-          key: cargo-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}
+          key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}
+
+      - name: Restore target cache (except gnu-dev)
+        id: cache_target_restore
+        if: ${{ !(matrix.target == 'x86_64-unknown-linux-gnu' && matrix.profile != 'release') }}
+        uses: actions/cache/restore@v4
+        with:
+          path: ${{ github.workspace }}/codex-rs/target/
+          key: cargo-target-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}
 
       - if: ${{ matrix.target == 'x86_64-unknown-linux-musl' || matrix.target == 'aarch64-unknown-linux-musl'}}
         name: Install musl build tools
@@ -194,6 +205,31 @@ jobs:
         env:
           RUST_BACKTRACE: 1
 
+      # Save caches explicitly; make non-fatal so cache packaging
+      # never fails the overall job. Only save when key wasn't hit.
+      - name: Save cargo home cache
+        if: always() && !cancelled() && steps.cache_cargo_home_restore.outputs.cache-hit != 'true'
+        continue-on-error: true
+        uses: actions/cache/save@v4
+        with:
+          path: |
+            ~/.cargo/bin/
+            ~/.cargo/registry/index/
+            ~/.cargo/registry/cache/
+            ~/.cargo/git/db/
+          key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}
+
+      - name: Save target cache (except gnu-dev)
+        if: >-
+          always() && !cancelled() &&
+          (steps.cache_target_restore.outputs.cache-hit != 'true') &&
+          !(matrix.target == 'x86_64-unknown-linux-gnu' && matrix.profile != 'release')
+        continue-on-error: true
+        uses: actions/cache/save@v4
+        with:
+          path: ${{ github.workspace }}/codex-rs/target/
+          key: cargo-target-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}
+
       # Fail the job if any of the previous steps failed.
       - name: verify all steps passed
         if: |

.github/workflows/rust-release.yml

@@ -47,7 +47,7 @@ jobs:
 
   build:
     needs: tag-check
-    name: ${{ matrix.runner }} - ${{ matrix.target }}
+    name: Build - ${{ matrix.runner }} - ${{ matrix.target }}
     runs-on: ${{ matrix.runner }}
     timeout-minutes: 30
     defaults:
@@ -94,11 +94,118 @@ jobs:
       - if: ${{ matrix.target == 'x86_64-unknown-linux-musl' || matrix.target == 'aarch64-unknown-linux-musl'}}
         name: Install musl build tools
         run: |
-          sudo apt install -y musl-tools pkg-config
+          sudo apt-get update
+          sudo apt-get install -y musl-tools pkg-config
 
       - name: Cargo build
         run: cargo build --target ${{ matrix.target }} --release --bin codex --bin codex-responses-api-proxy
 
+      - if: ${{ matrix.runner == 'macos-14' }}
+        name: Configure Apple code signing
+        shell: bash
+        env:
+          KEYCHAIN_PASSWORD: actions
+          APPLE_CERTIFICATE: ${{ secrets.APPLE_CERTIFICATE_P12 }}
+          APPLE_CERTIFICATE_PASSWORD: ${{ secrets.APPLE_CERTIFICATE_PASSWORD }}
+        run: |
+          set -euo pipefail
+
+          if [[ -z "${APPLE_CERTIFICATE:-}" ]]; then
+            echo "APPLE_CERTIFICATE is required for macOS signing"
+            exit 1
+          fi
+
+          if [[ -z "${APPLE_CERTIFICATE_PASSWORD:-}" ]]; then
+            echo "APPLE_CERTIFICATE_PASSWORD is required for macOS signing"
+            exit 1
+          fi
+
+          cert_path="${RUNNER_TEMP}/apple_signing_certificate.p12"
+          echo "$APPLE_CERTIFICATE" | base64 -d > "$cert_path"
+
+          keychain_path="${RUNNER_TEMP}/codex-signing.keychain-db"
+          security create-keychain -p "$KEYCHAIN_PASSWORD" "$keychain_path"
+          security set-keychain-settings -lut 21600 "$keychain_path"
+          security unlock-keychain -p "$KEYCHAIN_PASSWORD" "$keychain_path"
+
+          keychain_args=()
+          cleanup_keychain() {
+            if ((${#keychain_args[@]} > 0)); then
+              security list-keychains -s "${keychain_args[@]}" || true
+              security default-keychain -s "${keychain_args[0]}" || true
+            else
+              security list-keychains -s || true
+            fi
+            if [[ -f "$keychain_path" ]]; then
+              security delete-keychain "$keychain_path" || true
+            fi
+          }
+
+          while IFS= read -r keychain; do
+            [[ -n "$keychain" ]] && keychain_args+=("$keychain")
+          done < <(security list-keychains | sed 's/^[[:space:]]*//;s/[[:space:]]*$//;s/"//g')
+
+          if ((${#keychain_args[@]} > 0)); then
+            security list-keychains -s "$keychain_path" "${keychain_args[@]}"
+          else
+            security list-keychains -s "$keychain_path"
+          fi
+
+          security default-keychain -s "$keychain_path"
+          security import "$cert_path" -k "$keychain_path" -P "$APPLE_CERTIFICATE_PASSWORD" -T /usr/bin/codesign -T /usr/bin/security
+          security set-key-partition-list -S apple-tool:,apple: -s -k "$KEYCHAIN_PASSWORD" "$keychain_path" > /dev/null
+
+          codesign_hashes=()
+          while IFS= read -r hash; do
+            [[ -n "$hash" ]] && codesign_hashes+=("$hash")
+          done < <(security find-identity -v -p codesigning "$keychain_path" \
+            | sed -n 's/.*\([0-9A-F]\{40\}\).*/\1/p' \
+            | sort -u)
+
+          if ((${#codesign_hashes[@]} == 0)); then
+            echo "No signing identities found in $keychain_path"
+            cleanup_keychain
+            rm -f "$cert_path"
+            exit 1
+          fi
+
+          if ((${#codesign_hashes[@]} > 1)); then
+            echo "Multiple signing identities found in $keychain_path:"
+            printf '  %s\n' "${codesign_hashes[@]}"
+            cleanup_keychain
+            rm -f "$cert_path"
+            exit 1
+          fi
+
+          APPLE_CODESIGN_IDENTITY="${codesign_hashes[0]}"
+
+          rm -f "$cert_path"
+
+          echo "APPLE_CODESIGN_IDENTITY=$APPLE_CODESIGN_IDENTITY" >> "$GITHUB_ENV"
+          echo "APPLE_CODESIGN_KEYCHAIN=$keychain_path" >> "$GITHUB_ENV"
+          echo "::add-mask::$APPLE_CODESIGN_IDENTITY"
+
+      - if: ${{ matrix.runner == 'macos-14' }}
+        name: Sign macOS binaries
+        shell: bash
+        run: |
+          set -euo pipefail
+
+          if [[ -z "${APPLE_CODESIGN_IDENTITY:-}" ]]; then
+            echo "APPLE_CODESIGN_IDENTITY is required for macOS signing"
+            exit 1
+          fi
+
+          keychain_args=()
+          if [[ -n "${APPLE_CODESIGN_KEYCHAIN:-}" && -f "${APPLE_CODESIGN_KEYCHAIN}" ]]; then
+            keychain_args+=(--keychain "${APPLE_CODESIGN_KEYCHAIN}")
+          fi
+
+          for binary in codex codex-responses-api-proxy; do
+            path="target/${{ matrix.target }}/release/${binary}"
+            codesign --force --options runtime --timestamp --sign "$APPLE_CODESIGN_IDENTITY" "${keychain_args[@]}" "$path"
+          done
+
       - name: Stage artifacts
         shell: bash
         run: |
@@ -157,6 +264,29 @@ jobs:
             zstd -T0 -19 --rm "$dest/$base"
           done
 
+      - name: Remove signing keychain
+        if: ${{ always() && matrix.runner == 'macos-14' }}
+        shell: bash
+        env:
+          APPLE_CODESIGN_KEYCHAIN: ${{ env.APPLE_CODESIGN_KEYCHAIN }}
+        run: |
+          set -euo pipefail
+          if [[ -n "${APPLE_CODESIGN_KEYCHAIN:-}" ]]; then
+            keychain_args=()
+            while IFS= read -r keychain; do
+              [[ "$keychain" == "$APPLE_CODESIGN_KEYCHAIN" ]] && continue
+              [[ -n "$keychain" ]] && keychain_args+=("$keychain")
+            done < <(security list-keychains | sed 's/^[[:space:]]*//;s/[[:space:]]*$//;s/"//g')
+            if ((${#keychain_args[@]} > 0)); then
+              security list-keychains -s "${keychain_args[@]}"
+              security default-keychain -s "${keychain_args[0]}"
+            fi
+
+            if [[ -f "$APPLE_CODESIGN_KEYCHAIN" ]]; then
+              security delete-keychain "$APPLE_CODESIGN_KEYCHAIN"
+            fi
+          fi
+
       - uses: actions/upload-artifact@v4
         with:
           name: ${{ matrix.target }}

.gitignore

@@ -30,6 +30,7 @@ result
 # cli tools
 CLAUDE.md
 .claude/
+AGENTS.override.md
 
 # caches
 .cache/

AGENTS.md

@@ -8,11 +8,16 @@ In the codex-rs folder where the rust code lives:
 - Never add or modify any code related to `CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR` or `CODEX_SANDBOX_ENV_VAR`.
   - You operate in a sandbox where `CODEX_SANDBOX_NETWORK_DISABLED=1` will be set whenever you use the `shell` tool. Any existing code that uses `CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR` was authored with this fact in mind. It is often used to early exit out of tests that the author knew you would not be able to run given your sandbox limitations.
   - Similarly, when you spawn a process using Seatbelt (`/usr/bin/sandbox-exec`), `CODEX_SANDBOX=seatbelt` will be set on the child process. Integration tests that want to run Seatbelt themselves cannot be run under Seatbelt, so checks for `CODEX_SANDBOX=seatbelt` are also often used to early exit out of tests, as appropriate.
+- Always collapse if statements per https://rust-lang.github.io/rust-clippy/master/index.html#collapsible_if
+- Always inline format! args when possible per https://rust-lang.github.io/rust-clippy/master/index.html#uninlined_format_args
+- Use method references over closures when possible per https://rust-lang.github.io/rust-clippy/master/index.html#redundant_closure_for_method_calls
+- When writing tests, prefer comparing the equality of entire objects over fields one by one.
 
 Run `just fmt` (in `codex-rs` directory) automatically after making Rust code changes; do not ask for approval to run it. Before finalizing a change to `codex-rs`, run `just fix -p <project>` (in `codex-rs` directory) to fix any linter issues in the code. Prefer scoping with `-p` to avoid slow workspace‑wide Clippy builds; only run `just fix` without `-p` if you changed shared crates. Additionally, run the tests:
+
 1. Run the test for the specific project that was changed. For example, if changes were made in `codex-rs/tui`, run `cargo test -p codex-tui`.
 2. Once those pass, if any changes were made in common, core, or protocol, run the complete test suite with `cargo test --all-features`.
-When running interactively, ask the user before running `just fix` to finalize. `just fmt` does not require approval. project-specific or individual tests can be run without asking the user, but do ask the user before running the complete test suite.
+   When running interactively, ask the user before running `just fix` to finalize. `just fmt` does not require approval. project-specific or individual tests can be run without asking the user, but do ask the user before running the complete test suite.
 
 ## TUI style conventions
 
@@ -28,6 +33,7 @@ See `codex-rs/tui/styles.md`.
     - Desired: vec!["  └ ".into(), "M".red(), " ".dim(), "tui/src/app.rs".dim()]
 
 ### TUI Styling (ratatui)
+
 - Prefer Stylize helpers: use "text".dim(), .bold(), .cyan(), .italic(), .underlined() instead of manual Style where possible.
 - Prefer simple conversions: use "text".into() for spans and vec![…].into() for lines; when inference is ambiguous (e.g., Paragraph::new/Cell::from), use Line::from(spans) or Span::from(text).
 - Computed styles: if the Style is computed at runtime, using `Span::styled` is OK (`Span::from(text).set_style(style)` is also acceptable).
@@ -39,6 +45,7 @@ See `codex-rs/tui/styles.md`.
 - Compactness: prefer the form that stays on one line after rustfmt; if only one of Line::from(vec![…]) or vec![…].into() avoids wrapping, choose that. If both wrap, pick the one with fewer wrapped lines.
 
 ### Text wrapping
+
 - Always use textwrap::wrap to wrap plain strings.
 - If you have a ratatui Line and you want to wrap it, use the helpers in tui/src/wrapping.rs, e.g. word_wrap_lines / word_wrap_line.
 - If you need to indent wrapped lines, use the initial_indent / subsequent_indent options from RtOptions if you can, rather than writing custom logic.
@@ -60,8 +67,34 @@ This repo uses snapshot tests (via `insta`), especially in `codex-rs/tui`, to va
   - `cargo insta accept -p codex-tui`
 
 If you don’t have the tool:
+
 - `cargo install cargo-insta`
 
 ### Test assertions
 
 - Tests should use pretty_assertions::assert_eq for clearer diffs. Import this at the top of the test module if it isn't already.
+
+### Integration tests (core)
+
+- Prefer the utilities in `core_test_support::responses` when writing end-to-end Codex tests.
+
+- All `mount_sse*` helpers return a `ResponseMock`; hold onto it so you can assert against outbound `/responses` POST bodies.
+- Use `ResponseMock::single_request()` when a test should only issue one POST, or `ResponseMock::requests()` to inspect every captured `ResponsesRequest`.
+- `ResponsesRequest` exposes helpers (`body_json`, `input`, `function_call_output`, `custom_tool_call_output`, `call_output`, `header`, `path`, `query_param`) so assertions can target structured payloads instead of manual JSON digging.
+- Build SSE payloads with the provided `ev_*` constructors and the `sse(...)`.
+
+- Typical pattern:
+
+  ```rust
+  let mock = responses::mount_sse_once(&server, responses::sse(vec![
+      responses::ev_response_created("resp-1"),
+      responses::ev_function_call(call_id, "shell", &serde_json::to_string(&args)?),
+      responses::ev_completed("resp-1"),
+  ])).await;
+
+  codex.submit(Op::UserTurn { ... }).await?;
+
+  // Assert request body if needed.
+  let request = mock.single_request();
+  // assert using request.function_call_output(call_id) or request.json_body() or other helpers.
+  ```

README.md

@@ -1,4 +1,3 @@
-
 <p align="center"><code>npm i -g @openai/codex</code><br />or <code>brew install codex</code></p>
 
 <p align="center"><strong>Codex CLI</strong> is a coding agent from OpenAI that runs locally on your computer.
@@ -62,8 +61,7 @@ You can also use Codex with an API key, but this requires [additional setup](./d
 
 ### Model Context Protocol (MCP)
 
-Codex CLI supports [MCP servers](./docs/advanced.md#model-context-protocol-mcp). Enable by adding an `mcp_servers` section to your `~/.codex/config.toml`.
-
+Codex can access MCP servers. To configure them, refer to the [config docs](./docs/config.md#mcp_servers).
 
 ### Configuration
 
@@ -83,8 +81,11 @@ Codex CLI supports a rich set of configuration options, with preferences stored
 - [**Authentication**](./docs/authentication.md)
   - [Auth methods](./docs/authentication.md#forcing-a-specific-auth-method-advanced)
   - [Login on a "Headless" machine](./docs/authentication.md#connecting-on-a-headless-machine)
+- **Automating Codex**
+  - [GitHub Action](https://github.com/openai/codex-action)
+  - [TypeScript SDK](./sdk/typescript/README.md)
+  - [Non-interactive mode (`codex exec`)](./docs/exec.md)
 - [**Advanced**](./docs/advanced.md)
-  - [Non-interactive / CI mode](./docs/advanced.md#non-interactive--ci-mode)
   - [Tracing / verbose logging](./docs/advanced.md#tracing--verbose-logging)
   - [Model Context Protocol (MCP)](./docs/advanced.md#model-context-protocol-mcp)
 - [**Zero data retention (ZDR)**](./docs/zdr.md)

codex-cli/bin/codex.js

@@ -80,6 +80,32 @@ function getUpdatedPath(newDirs) {
   return updatedPath;
 }
 
+/**
+ * Use heuristics to detect the package manager that was used to install Codex
+ * in order to give the user a hint about how to update it.
+ */
+function detectPackageManager() {
+  const userAgent = process.env.npm_config_user_agent || "";
+  if (/\bbun\//.test(userAgent)) {
+    return "bun";
+  }
+
+  const execPath = process.env.npm_execpath || "";
+  if (execPath.includes("bun")) {
+    return "bun";
+  }
+
+  if (
+    process.env.BUN_INSTALL ||
+    process.env.BUN_INSTALL_GLOBAL_DIR ||
+    process.env.BUN_INSTALL_BIN_DIR
+  ) {
+    return "bun";
+  }
+
+  return userAgent ? "npm" : null;
+}
+
 const additionalDirs = [];
 const pathDir = path.join(archRoot, "path");
 if (existsSync(pathDir)) {
@@ -87,9 +113,16 @@ if (existsSync(pathDir)) {
 }
 const updatedPath = getUpdatedPath(additionalDirs);
 
+const env = { ...process.env, PATH: updatedPath };
+const packageManagerEnvVar =
+  detectPackageManager() === "bun"
+    ? "CODEX_MANAGED_BY_BUN"
+    : "CODEX_MANAGED_BY_NPM";
+env[packageManagerEnvVar] = "1";
+
 const child = spawn(binaryPath, process.argv.slice(2), {
   stdio: "inherit",
-  env: { ...process.env, PATH: updatedPath, CODEX_MANAGED_BY_NPM: "1" },
+  env,
 });
 
 child.on("error", (err) => {

codex-rs/Cargo.toml

@@ -32,6 +32,7 @@ members = [
     "git-apply",
     "utils/json-to-toml",
     "utils/readiness",
+    "utils/string",
 ]
 resolver = "2"
 
@@ -71,6 +72,7 @@ codex-rmcp-client = { path = "rmcp-client" }
 codex-tui = { path = "tui" }
 codex-utils-json-to-toml = { path = "utils/json-to-toml" }
 codex-utils-readiness = { path = "utils/readiness" }
+codex-utils-string = { path = "utils/string" }
 core_test_support = { path = "core/tests/common" }
 mcp-types = { path = "mcp-types" }
 mcp_test_support = { path = "mcp-server/tests/common" }
@@ -81,10 +83,12 @@ ansi-to-tui = "7.0.0"
 anyhow = "1"
 arboard = "3"
 askama = "0.12"
+assert_matches = "1.5.0"
 assert_cmd = "2"
 async-channel = "2.3.1"
 async-stream = "0.3.6"
 async-trait = "0.1.89"
+axum = { version = "0.8", default-features = false }
 base64 = "0.22.1"
 bytes = "1.10.1"
 chrono = "0.4.42"
@@ -102,7 +106,7 @@ env-flags = "0.1.1"
 env_logger = "0.11.5"
 escargot = "0.5"
 eventsource-stream = "0.2.3"
-futures = "0.3"
+futures = { version = "0.3", default-features = false }
 icu_decimal = "2.0.0"
 icu_locale_core = "2.0.0"
 ignore = "0.4.23"
@@ -110,6 +114,7 @@ image = { version = "^0.25.8", default-features = false }
 indexmap = "2.6.0"
 insta = "1.43.2"
 itertools = "0.14.0"
+keyring = "3.6"
 landlock = "0.4.1"
 lazy_static = "1"
 libc = "0.2.175"
@@ -117,6 +122,7 @@ log = "0.4"
 maplit = "1.0.2"
 mime_guess = "2.0.5"
 multimap = "0.10.0"
+notify = "8.2.0"
 nucleo-matcher = "0.3.1"
 openssl-sys = "*"
 opentelemetry = "0.30.0"
@@ -138,11 +144,13 @@ rand = "0.9"
 ratatui = "0.29.0"
 regex-lite = "0.1.7"
 reqwest = "0.12"
+rmcp = { version = "0.8.0", default-features = false }
 schemars = "0.8.22"
 seccompiler = "0.5.0"
 serde = "1"
 serde_json = "1"
 serde_with = "3.14"
+serial_test = "3.2.0"
 sha1 = "0.10.6"
 sha2 = "0.10"
 shlex = "1.3.0"
@@ -168,8 +176,9 @@ tracing = "0.1.41"
 tracing-appender = "0.2.3"
 tracing-subscriber = "0.3.20"
 tracing-test = "0.2.5"
-tree-sitter = "0.25.9"
-tree-sitter-bash = "0.25.0"
+tree-sitter = "0.25.10"
+tree-sitter-bash = "0.25"
+tree-sitter-highlight = "0.25.10"
 ts-rs = "11"
 unicode-segmentation = "1.12.0"
 unicode-width = "0.2"
@@ -237,5 +246,9 @@ strip = "symbols"
 codegen-units = 1
 
 [patch.crates-io]
+# Uncomment to debug local changes.
 # ratatui = { path = "../../ratatui" }
 ratatui = { git = "https://github.com/nornagon/ratatui", branch = "nornagon-v0.29.0-patch" }
+
+# Uncomment to debug local changes.
+# rmcp = { path = "../../rust-sdk/crates/rmcp" }

codex-rs/README.md

@@ -23,9 +23,15 @@ Codex supports a rich set of configuration options. Note that the Rust CLI uses
 
 ### Model Context Protocol Support
 
-Codex CLI functions as an MCP client that can connect to MCP servers on startup. See the [`mcp_servers`](../docs/config.md#mcp_servers) section in the configuration documentation for details.
+#### MCP client
 
-It is still experimental, but you can also launch Codex as an MCP _server_ by running `codex mcp-server`. Use the [`@modelcontextprotocol/inspector`](https://github.com/modelcontextprotocol/inspector) to try it out:
+Codex CLI functions as an MCP client that allows the Codex CLI and IDE extension to connect to MCP servers on startup. See the [`configuration documentation`](../docs/config.md#mcp_servers) for details.
+
+#### MCP server (experimental)
+
+Codex can be launched as an MCP _server_ by running `codex mcp-server`. This allows _other_ MCP clients to use Codex as a tool for another agent.
+
+Use the [`@modelcontextprotocol/inspector`](https://github.com/modelcontextprotocol/inspector) to try it out:
 
 ```shell
 npx @modelcontextprotocol/inspector codex mcp-server
@@ -71,9 +77,13 @@ To test to see what happens when a command is run under the sandbox provided by
 
 ```
 # macOS
-codex debug seatbelt [--full-auto] [COMMAND]...
+codex sandbox macos [--full-auto] [COMMAND]...
 
 # Linux
+codex sandbox linux [--full-auto] [COMMAND]...
+
+# Legacy aliases
+codex debug seatbelt [--full-auto] [COMMAND]...
 codex debug landlock [--full-auto] [COMMAND]...
 ```
 

codex-rs/ansi-escape/src/lib.rs

@@ -3,11 +3,30 @@ use ansi_to_tui::IntoText;
 use ratatui::text::Line;
 use ratatui::text::Text;
 
+// Expand tabs in a best-effort way for transcript rendering.
+// Tabs can interact poorly with left-gutter prefixes in our TUI and CLI
+// transcript views (e.g., `nl` separates line numbers from content with a tab).
+// Replacing tabs with spaces avoids odd visual artifacts without changing
+// semantics for our use cases.
+fn expand_tabs(s: &str) -> std::borrow::Cow<'_, str> {
+    if s.contains('\t') {
+        // Keep it simple: replace each tab with 4 spaces.
+        // We do not try to align to tab stops since most usages (like `nl`)
+        // look acceptable with a fixed substitution and this avoids stateful math
+        // across spans.
+        std::borrow::Cow::Owned(s.replace('\t', "    "))
+    } else {
+        std::borrow::Cow::Borrowed(s)
+    }
+}
+
 /// This function should be used when the contents of `s` are expected to match
 /// a single line. If multiple lines are found, a warning is logged and only the
 /// first line is returned.
 pub fn ansi_escape_line(s: &str) -> Line<'static> {
-    let text = ansi_escape(s);
+    // Normalize tabs to spaces to avoid odd gutter collisions in transcript mode.
+    let s = expand_tabs(s);
+    let text = ansi_escape(&s);
     match text.lines.as_slice() {
         [] => "".into(),
         [only] => only.clone(),

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

@@ -725,6 +725,7 @@ pub struct FuzzyFileSearchParams {
 pub struct FuzzyFileSearchResult {
     pub root: String,
     pub path: String,
+    pub file_name: String,
     pub score: u32,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub indices: Option<Vec<u32>>,

codex-rs/app-server/README.md

@@ -0,0 +1,15 @@
+# codex-app-server
+
+`codex app-server` is the harness Codex uses to power rich interfaces such as the [Codex VS Code extension](https://marketplace.visualstudio.com/items?itemName=openai.chatgpt). The message schema is currently unstable, but those who wish to build experimental UIs on top of Codex may find it valuable.
+
+## 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.
+
+## Message Schema
+
+Currently, you can dump a TypeScript version of the schema using `codex generate-ts`. It is specific to the version of Codex you used to run `generate-ts`, so the two are guaranteed to be compatible.
+
+```
+codex generate-ts --out DIR
+```

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

@@ -53,6 +53,7 @@ use codex_core::AuthManager;
 use codex_core::CodexConversation;
 use codex_core::ConversationManager;
 use codex_core::Cursor as RolloutCursor;
+use codex_core::INTERACTIVE_SESSION_SOURCES;
 use codex_core::NewConversation;
 use codex_core::RolloutRecorder;
 use codex_core::SessionMeta;
@@ -499,7 +500,7 @@ impl CodexMessageProcessor {
     }
 
     async fn get_user_saved_config(&self, request_id: RequestId) {
-        let toml_value = match load_config_as_toml(&self.config.codex_home) {
+        let toml_value = match load_config_as_toml(&self.config.codex_home).await {
             Ok(val) => val,
             Err(err) => {
                 let error = JSONRPCErrorError {
@@ -652,18 +653,19 @@ impl CodexMessageProcessor {
     }
 
     async fn process_new_conversation(&self, request_id: RequestId, params: NewConversationParams) {
-        let config = match derive_config_from_params(params, self.codex_linux_sandbox_exe.clone()) {
-            Ok(config) => config,
-            Err(err) => {
-                let error = JSONRPCErrorError {
-                    code: INVALID_REQUEST_ERROR_CODE,
-                    message: format!("error deriving config: {err}"),
-                    data: None,
-                };
-                self.outgoing.send_error(request_id, error).await;
-                return;
-            }
-        };
+        let config =
+            match derive_config_from_params(params, self.codex_linux_sandbox_exe.clone()).await {
+                Ok(config) => config,
+                Err(err) => {
+                    let error = JSONRPCErrorError {
+                        code: INVALID_REQUEST_ERROR_CODE,
+                        message: format!("error deriving config: {err}"),
+                        data: None,
+                    };
+                    self.outgoing.send_error(request_id, error).await;
+                    return;
+                }
+            };
 
         match self.conversation_manager.new_conversation(config).await {
             Ok(conversation_id) => {
@@ -708,6 +710,7 @@ impl CodexMessageProcessor {
             &self.config.codex_home,
             page_size,
             cursor_ref,
+            INTERACTIVE_SESSION_SOURCES,
         )
         .await
         {
@@ -750,7 +753,7 @@ impl CodexMessageProcessor {
         // Derive a Config using the same logic as new conversation, honoring overrides if provided.
         let config = match params.overrides {
             Some(overrides) => {
-                derive_config_from_params(overrides, self.codex_linux_sandbox_exe.clone())
+                derive_config_from_params(overrides, self.codex_linux_sandbox_exe.clone()).await
             }
             None => Ok(self.config.as_ref().clone()),
         };
@@ -1047,6 +1050,7 @@ impl CodexMessageProcessor {
                 effort,
                 summary,
                 final_output_json_schema: None,
+                disabled_tools: None,
             })
             .await;
 
@@ -1318,7 +1322,7 @@ async fn apply_bespoke_event_handling(
     }
 }
 
-fn derive_config_from_params(
+async fn derive_config_from_params(
     params: NewConversationParams,
     codex_linux_sandbox_exe: Option<PathBuf>,
 ) -> std::io::Result<Config> {
@@ -1356,7 +1360,7 @@ fn derive_config_from_params(
         .map(|(k, v)| (k, json_to_toml(v)))
         .collect();
 
-    Config::load_with_cli_overrides(cli_overrides, overrides)
+    Config::load_with_cli_overrides(cli_overrides, overrides).await
 }
 
 async fn on_patch_approval_response(

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

@@ -1,5 +1,6 @@
 use std::num::NonZero;
 use std::num::NonZeroUsize;
+use std::path::Path;
 use std::path::PathBuf;
 use std::sync::Arc;
 use std::sync::atomic::AtomicBool;
@@ -56,9 +57,16 @@ pub(crate) async fn run_fuzzy_file_search(
         match res {
             Ok(Ok((root, res))) => {
                 for m in res.matches {
+                    let path = m.path;
+                    //TODO(shijie): Move file name generation to file_search lib.
+                    let file_name = Path::new(&path)
+                        .file_name()
+                        .map(|name| name.to_string_lossy().into_owned())
+                        .unwrap_or_else(|| path.clone());
                     let result = FuzzyFileSearchResult {
                         root: root.clone(),
-                        path: m.path,
+                        path,
+                        file_name,
                         score: m.score,
                         indices: m.indices,
                     };

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

@@ -81,6 +81,7 @@ pub async fn run_main(
         )
     })?;
     let config = Config::load_with_cli_overrides(cli_kv_overrides, ConfigOverrides::default())
+        .await
         .map_err(|e| {
             std::io::Error::new(ErrorKind::InvalidData, format!("error loading config: {e}"))
         })?;

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

@@ -17,6 +17,7 @@ use codex_core::ConversationManager;
 use codex_core::config::Config;
 use codex_core::default_client::USER_AGENT_SUFFIX;
 use codex_core::default_client::get_codex_user_agent;
+use codex_protocol::protocol::SessionSource;
 use std::sync::Arc;
 
 pub(crate) struct MessageProcessor {
@@ -34,8 +35,11 @@ impl MessageProcessor {
         config: Arc<Config>,
     ) -> Self {
         let outgoing = Arc::new(outgoing);
-        let auth_manager = AuthManager::shared(config.codex_home.clone());
-        let conversation_manager = Arc::new(ConversationManager::new(auth_manager.clone()));
+        let auth_manager = AuthManager::shared(config.codex_home.clone(), false);
+        let conversation_manager = Arc::new(ConversationManager::new(
+            auth_manager.clone(),
+            SessionSource::VSCode,
+        ));
         let codex_message_processor = CodexMessageProcessor::new(
             auth_manager,
             conversation_manager,

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

@@ -1,3 +1,5 @@
+use anyhow::Context;
+use anyhow::Result;
 use app_test_support::McpProcess;
 use codex_app_server_protocol::JSONRPCResponse;
 use codex_app_server_protocol::RequestId;
@@ -9,30 +11,41 @@ use tokio::time::timeout;
 const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn test_fuzzy_file_search_sorts_and_includes_indices() {
+async fn test_fuzzy_file_search_sorts_and_includes_indices() -> Result<()> {
     // Prepare a temporary Codex home and a separate root with test files.
-    let codex_home = TempDir::new().expect("create temp codex home");
-    let root = TempDir::new().expect("create temp search root");
+    let codex_home = TempDir::new().context("create temp codex home")?;
+    let root = TempDir::new().context("create temp search root")?;
 
-    // Create files designed to have deterministic ordering for query "abc".
-    std::fs::write(root.path().join("abc"), "x").expect("write file abc");
-    std::fs::write(root.path().join("abcde"), "x").expect("write file abcx");
-    std::fs::write(root.path().join("abexy"), "x").expect("write file abcx");
-    std::fs::write(root.path().join("zzz.txt"), "x").expect("write file zzz");
+    // Create files designed to have deterministic ordering for query "abe".
+    std::fs::write(root.path().join("abc"), "x").context("write file abc")?;
+    std::fs::write(root.path().join("abcde"), "x").context("write file abcde")?;
+    std::fs::write(root.path().join("abexy"), "x").context("write file abexy")?;
+    std::fs::write(root.path().join("zzz.txt"), "x").context("write file zzz")?;
+    let sub_dir = root.path().join("sub");
+    std::fs::create_dir_all(&sub_dir).context("create sub dir")?;
+    let sub_abce_path = sub_dir.join("abce");
+    std::fs::write(&sub_abce_path, "x").context("write file sub/abce")?;
+    let sub_abce_rel = sub_abce_path
+        .strip_prefix(root.path())
+        .context("strip root prefix from sub/abce")?
+        .to_string_lossy()
+        .to_string();
 
     // Start MCP server and initialize.
-    let mut mcp = McpProcess::new(codex_home.path()).await.expect("spawn mcp");
+    let mut mcp = McpProcess::new(codex_home.path())
+        .await
+        .context("spawn mcp")?;
     timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
         .await
-        .expect("init timeout")
-        .expect("init failed");
+        .context("init timeout")?
+        .context("init failed")?;
 
     let root_path = root.path().to_string_lossy().to_string();
     // Send fuzzyFileSearch request.
     let request_id = mcp
         .send_fuzzy_file_search_request("abe", vec![root_path.clone()], None)
         .await
-        .expect("send fuzzyFileSearch");
+        .context("send fuzzyFileSearch")?;
 
     // Read response and verify shape and ordering.
     let resp: JSONRPCResponse = timeout(
@@ -40,39 +53,65 @@ async fn test_fuzzy_file_search_sorts_and_includes_indices() {
         mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
     )
     .await
-    .expect("fuzzyFileSearch timeout")
-    .expect("fuzzyFileSearch resp");
+    .context("fuzzyFileSearch timeout")?
+    .context("fuzzyFileSearch resp")?;
 
     let value = resp.result;
+    // The path separator on Windows affects the score.
+    let expected_score = if cfg!(windows) { 69 } else { 72 };
+
     assert_eq!(
         value,
         json!({
             "files": [
-                { "root": root_path.clone(), "path": "abexy", "score": 88, "indices": [0, 1, 2] },
-                { "root": root_path.clone(), "path": "abcde", "score": 74, "indices": [0, 1, 4] },
+                {
+                    "root": root_path.clone(),
+                    "path": "abexy",
+                    "file_name": "abexy",
+                    "score": 88,
+                    "indices": [0, 1, 2],
+                },
+                {
+                    "root": root_path.clone(),
+                    "path": "abcde",
+                    "file_name": "abcde",
+                    "score": 74,
+                    "indices": [0, 1, 4],
+                },
+                {
+                    "root": root_path.clone(),
+                    "path": sub_abce_rel,
+                    "file_name": "abce",
+                    "score": expected_score,
+                    "indices": [4, 5, 7],
+                },
             ]
         })
     );
+
+    Ok(())
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn test_fuzzy_file_search_accepts_cancellation_token() {
-    let codex_home = TempDir::new().expect("create temp codex home");
-    let root = TempDir::new().expect("create temp search root");
+async fn test_fuzzy_file_search_accepts_cancellation_token() -> Result<()> {
+    let codex_home = TempDir::new().context("create temp codex home")?;
+    let root = TempDir::new().context("create temp search root")?;
 
-    std::fs::write(root.path().join("alpha.txt"), "contents").expect("write alpha");
+    std::fs::write(root.path().join("alpha.txt"), "contents").context("write alpha")?;
 
-    let mut mcp = McpProcess::new(codex_home.path()).await.expect("spawn mcp");
+    let mut mcp = McpProcess::new(codex_home.path())
+        .await
+        .context("spawn mcp")?;
     timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
         .await
-        .expect("init timeout")
-        .expect("init failed");
+        .context("init timeout")?
+        .context("init failed")?;
 
     let root_path = root.path().to_string_lossy().to_string();
     let request_id = mcp
         .send_fuzzy_file_search_request("alp", vec![root_path.clone()], None)
         .await
-        .expect("send fuzzyFileSearch");
+        .context("send fuzzyFileSearch")?;
 
     let request_id_2 = mcp
         .send_fuzzy_file_search_request(
@@ -81,24 +120,27 @@ async fn test_fuzzy_file_search_accepts_cancellation_token() {
             Some(request_id.to_string()),
         )
         .await
-        .expect("send fuzzyFileSearch");
+        .context("send fuzzyFileSearch")?;
 
     let resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(request_id_2)),
     )
     .await
-    .expect("fuzzyFileSearch timeout")
-    .expect("fuzzyFileSearch resp");
+    .context("fuzzyFileSearch timeout")?
+    .context("fuzzyFileSearch resp")?;
 
     let files = resp
         .result
         .get("files")
-        .and_then(|value| value.as_array())
-        .cloned()
-        .expect("files array");
+        .context("files key missing")?
+        .as_array()
+        .context("files not array")?
+        .clone();
 
     assert_eq!(files.len(), 1);
     assert_eq!(files[0]["root"], root_path);
     assert_eq!(files[0]["path"], "alpha.txt");
+
+    Ok(())
 }

codex-rs/apply-patch/Cargo.toml

@@ -23,5 +23,6 @@ tree-sitter-bash = { workspace = true }
 
 [dev-dependencies]
 assert_cmd = { workspace = true }
+assert_matches = { workspace = true }
 pretty_assertions = { workspace = true }
 tempfile = { workspace = true }

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

@@ -843,6 +843,7 @@ pub fn print_summary(
 #[cfg(test)]
 mod tests {
     use super::*;
+    use assert_matches::assert_matches;
     use pretty_assertions::assert_eq;
     use std::fs;
     use std::string::ToString;
@@ -894,10 +895,10 @@ mod tests {
 
     fn assert_not_match(script: &str) {
         let args = args_bash(script);
-        assert!(matches!(
+        assert_matches!(
             maybe_parse_apply_patch(&args),
             MaybeApplyPatch::NotApplyPatch
-        ));
+        );
     }
 
     #[test]
@@ -905,10 +906,10 @@ mod tests {
         let patch = "*** Begin Patch\n*** Add File: foo\n+hi\n*** End Patch".to_string();
         let args = vec![patch];
         let dir = tempdir().unwrap();
-        assert!(matches!(
+        assert_matches!(
             maybe_parse_apply_patch_verified(&args, dir.path()),
             MaybeApplyPatchVerified::CorrectnessError(ApplyPatchError::ImplicitInvocation)
-        ));
+        );
     }
 
     #[test]
@@ -916,10 +917,10 @@ mod tests {
         let script = "*** Begin Patch\n*** Add File: foo\n+hi\n*** End Patch";
         let args = args_bash(script);
         let dir = tempdir().unwrap();
-        assert!(matches!(
+        assert_matches!(
             maybe_parse_apply_patch_verified(&args, dir.path()),
             MaybeApplyPatchVerified::CorrectnessError(ApplyPatchError::ImplicitInvocation)
-        ));
+        );
     }
 
     #[test]

codex-rs/chatgpt/src/apply_command.rs

@@ -29,7 +29,8 @@ pub async fn run_apply_command(
             .parse_overrides()
             .map_err(anyhow::Error::msg)?,
         ConfigOverrides::default(),
-    )?;
+    )
+    .await?;
 
     init_chatgpt_token_from_auth(&config.codex_home).await?;
 

codex-rs/cli/Cargo.toml

@@ -32,6 +32,7 @@ codex-app-server-protocol = { workspace = true }
 codex-protocol-ts = { workspace = true }
 codex-responses-api-proxy = { workspace = true }
 codex-tui = { workspace = true }
+codex-rmcp-client = { workspace = true }
 codex-cloud-tasks = { path = "../cloud-tasks" }
 ctor = { workspace = true }
 owo-colors = { workspace = true }
@@ -46,6 +47,7 @@ tokio = { workspace = true, features = [
 ] }
 
 [dev-dependencies]
+assert_matches = { workspace = true }
 assert_cmd = { workspace = true }
 predicates = { workspace = true }
 pretty_assertions = { workspace = true }

codex-rs/cli/src/debug_sandbox.rs

@@ -73,7 +73,8 @@ async fn run_command_under_sandbox(
             codex_linux_sandbox_exe,
             ..Default::default()
         },
-    )?;
+    )
+    .await?;
 
     // In practice, this should be `std::env::current_dir()` because this CLI
     // does not support `--cwd`, but let's use the config value for consistency.

codex-rs/cli/src/login.rs

@@ -9,6 +9,8 @@ use codex_core::config::ConfigOverrides;
 use codex_login::ServerOptions;
 use codex_login::run_device_code_login;
 use codex_login::run_login_server;
+use std::io::IsTerminal;
+use std::io::Read;
 use std::path::PathBuf;
 
 pub async fn login_with_chatgpt(codex_home: PathBuf) -> std::io::Result<()> {
@@ -24,7 +26,7 @@ pub async fn login_with_chatgpt(codex_home: PathBuf) -> std::io::Result<()> {
 }
 
 pub async fn run_login_with_chatgpt(cli_config_overrides: CliConfigOverrides) -> ! {
-    let config = load_config_or_exit(cli_config_overrides);
+    let config = load_config_or_exit(cli_config_overrides).await;
 
     match login_with_chatgpt(config.codex_home).await {
         Ok(_) => {
@@ -42,7 +44,7 @@ pub async fn run_login_with_api_key(
     cli_config_overrides: CliConfigOverrides,
     api_key: String,
 ) -> ! {
-    let config = load_config_or_exit(cli_config_overrides);
+    let config = load_config_or_exit(cli_config_overrides).await;
 
     match login_with_api_key(&config.codex_home, &api_key) {
         Ok(_) => {
@@ -56,13 +58,40 @@ pub async fn run_login_with_api_key(
     }
 }
 
+pub fn read_api_key_from_stdin() -> String {
+    let mut stdin = std::io::stdin();
+
+    if stdin.is_terminal() {
+        eprintln!(
+            "--with-api-key expects the API key on stdin. Try piping it, e.g. `printenv OPENAI_API_KEY | codex login --with-api-key`."
+        );
+        std::process::exit(1);
+    }
+
+    eprintln!("Reading API key from stdin...");
+
+    let mut buffer = String::new();
+    if let Err(err) = stdin.read_to_string(&mut buffer) {
+        eprintln!("Failed to read API key from stdin: {err}");
+        std::process::exit(1);
+    }
+
+    let api_key = buffer.trim().to_string();
+    if api_key.is_empty() {
+        eprintln!("No API key provided via stdin.");
+        std::process::exit(1);
+    }
+
+    api_key
+}
+
 /// Login using the OAuth device code flow.
 pub async fn run_login_with_device_code(
     cli_config_overrides: CliConfigOverrides,
     issuer_base_url: Option<String>,
     client_id: Option<String>,
 ) -> ! {
-    let config = load_config_or_exit(cli_config_overrides);
+    let config = load_config_or_exit(cli_config_overrides).await;
     let mut opts = ServerOptions::new(
         config.codex_home,
         client_id.unwrap_or(CLIENT_ID.to_string()),
@@ -83,7 +112,7 @@ pub async fn run_login_with_device_code(
 }
 
 pub async fn run_login_status(cli_config_overrides: CliConfigOverrides) -> ! {
-    let config = load_config_or_exit(cli_config_overrides);
+    let config = load_config_or_exit(cli_config_overrides).await;
 
     match CodexAuth::from_codex_home(&config.codex_home) {
         Ok(Some(auth)) => match auth.mode {
@@ -114,7 +143,7 @@ pub async fn run_login_status(cli_config_overrides: CliConfigOverrides) -> ! {
 }
 
 pub async fn run_logout(cli_config_overrides: CliConfigOverrides) -> ! {
-    let config = load_config_or_exit(cli_config_overrides);
+    let config = load_config_or_exit(cli_config_overrides).await;
 
     match logout(&config.codex_home) {
         Ok(true) => {
@@ -132,7 +161,7 @@ pub async fn run_logout(cli_config_overrides: CliConfigOverrides) -> ! {
     }
 }
 
-fn load_config_or_exit(cli_config_overrides: CliConfigOverrides) -> Config {
+async fn load_config_or_exit(cli_config_overrides: CliConfigOverrides) -> Config {
     let cli_overrides = match cli_config_overrides.parse_overrides() {
         Ok(v) => v,
         Err(e) => {
@@ -142,7 +171,7 @@ fn load_config_or_exit(cli_config_overrides: CliConfigOverrides) -> Config {
     };
 
     let config_overrides = ConfigOverrides::default();
-    match Config::load_with_cli_overrides(cli_overrides, config_overrides) {
+    match Config::load_with_cli_overrides(cli_overrides, config_overrides).await {
         Ok(config) => config,
         Err(e) => {
             eprintln!("Error loading configuration: {e}");

codex-rs/cli/src/main.rs

@@ -7,6 +7,7 @@ use codex_chatgpt::apply_command::ApplyCommand;
 use codex_chatgpt::apply_command::run_apply_command;
 use codex_cli::LandlockCommand;
 use codex_cli::SeatbeltCommand;
+use codex_cli::login::read_api_key_from_stdin;
 use codex_cli::login::run_login_status;
 use codex_cli::login::run_login_with_api_key;
 use codex_cli::login::run_login_with_chatgpt;
@@ -25,6 +26,8 @@ use supports_color::Stream;
 mod mcp_cmd;
 
 use crate::mcp_cmd::McpCli;
+use codex_core::config::Config;
+use codex_core::config::ConfigOverrides;
 
 /// Codex CLI
 ///
@@ -44,6 +47,9 @@ struct MultitoolCli {
     #[clap(flatten)]
     pub config_overrides: CliConfigOverrides,
 
+    #[clap(flatten)]
+    pub feature_toggles: FeatureToggles,
+
     #[clap(flatten)]
     interactive: TuiCli,
 
@@ -75,8 +81,9 @@ enum Subcommand {
     /// Generate shell completion scripts.
     Completion(CompletionCommand),
 
-    /// Internal debugging commands.
-    Debug(DebugArgs),
+    /// Run commands within a Codex-provided sandbox.
+    #[clap(visible_alias = "debug")]
+    Sandbox(SandboxArgs),
 
     /// Apply the latest diff produced by Codex agent as a `git apply` to your local working tree.
     #[clap(visible_alias = "a")]
@@ -95,6 +102,9 @@ enum Subcommand {
     /// Internal: run the responses API proxy.
     #[clap(hide = true)]
     ResponsesApiProxy(ResponsesApiProxyArgs),
+
+    /// Inspect feature flags.
+    Features(FeaturesCli),
 }
 
 #[derive(Debug, Parser)]
@@ -120,18 +130,20 @@ struct ResumeCommand {
 }
 
 #[derive(Debug, Parser)]
-struct DebugArgs {
+struct SandboxArgs {
     #[command(subcommand)]
-    cmd: DebugCommand,
+    cmd: SandboxCommand,
 }
 
 #[derive(Debug, clap::Subcommand)]
-enum DebugCommand {
+enum SandboxCommand {
     /// Run a command under Seatbelt (macOS only).
-    Seatbelt(SeatbeltCommand),
+    #[clap(visible_alias = "seatbelt")]
+    Macos(SeatbeltCommand),
 
     /// Run a command under Landlock+seccomp (Linux only).
-    Landlock(LandlockCommand),
+    #[clap(visible_alias = "landlock")]
+    Linux(LandlockCommand),
 }
 
 #[derive(Debug, Parser)]
@@ -139,12 +151,21 @@ struct LoginCommand {
     #[clap(skip)]
     config_overrides: CliConfigOverrides,
 
-    #[arg(long = "api-key", value_name = "API_KEY")]
+    #[arg(
+        long = "with-api-key",
+        help = "Read the API key from stdin (e.g. `printenv OPENAI_API_KEY | codex login --with-api-key`)"
+    )]
+    with_api_key: bool,
+
+    #[arg(
+        long = "api-key",
+        value_name = "API_KEY",
+        help = "(deprecated) Previously accepted the API key directly; now exits with guidance to use --with-api-key",
+        hide = true
+    )]
     api_key: Option<String>,
 
-    /// EXPERIMENTAL: Use device code flow (not yet supported)
-    /// This feature is experimental and may changed in future releases.
-    #[arg(long = "experimental_use-device-code", hide = true)]
+    #[arg(long = "device-auth")]
     use_device_code: bool,
 
     /// EXPERIMENTAL: Use custom OAuth issuer base URL (advanced)
@@ -218,6 +239,53 @@ fn print_exit_messages(exit_info: AppExitInfo) {
     }
 }
 
+#[derive(Debug, Default, Parser, Clone)]
+struct FeatureToggles {
+    /// Enable a feature (repeatable). Equivalent to `-c features.<name>=true`.
+    #[arg(long = "enable", value_name = "FEATURE", action = clap::ArgAction::Append, global = true)]
+    enable: Vec<String>,
+
+    /// Disable a feature (repeatable). Equivalent to `-c features.<name>=false`.
+    #[arg(long = "disable", value_name = "FEATURE", action = clap::ArgAction::Append, global = true)]
+    disable: Vec<String>,
+}
+
+impl FeatureToggles {
+    fn to_overrides(&self) -> Vec<String> {
+        let mut v = Vec::new();
+        for k in &self.enable {
+            v.push(format!("features.{k}=true"));
+        }
+        for k in &self.disable {
+            v.push(format!("features.{k}=false"));
+        }
+        v
+    }
+}
+
+#[derive(Debug, Parser)]
+struct FeaturesCli {
+    #[command(subcommand)]
+    sub: FeaturesSubcommand,
+}
+
+#[derive(Debug, Parser)]
+enum FeaturesSubcommand {
+    /// List known features with their stage and effective state.
+    List,
+}
+
+fn stage_str(stage: codex_core::features::Stage) -> &'static str {
+    use codex_core::features::Stage;
+    match stage {
+        Stage::Experimental => "experimental",
+        Stage::Beta => "beta",
+        Stage::Stable => "stable",
+        Stage::Deprecated => "deprecated",
+        Stage::Removed => "removed",
+    }
+}
+
 /// As early as possible in the process lifecycle, apply hardening measures. We
 /// skip this in debug builds to avoid interfering with debugging.
 #[ctor::ctor]
@@ -235,11 +303,17 @@ fn main() -> anyhow::Result<()> {
 
 async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()> {
     let MultitoolCli {
-        config_overrides: root_config_overrides,
+        config_overrides: mut root_config_overrides,
+        feature_toggles,
         mut interactive,
         subcommand,
     } = MultitoolCli::parse();
 
+    // Fold --enable/--disable into config overrides so they flow to all subcommands.
+    root_config_overrides
+        .raw_overrides
+        .extend(feature_toggles.to_overrides());
+
     match subcommand {
         None => {
             prepend_config_flags(
@@ -279,7 +353,8 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
                 last,
                 config_overrides,
             );
-            codex_tui::run_main(interactive, codex_linux_sandbox_exe).await?;
+            let exit_info = codex_tui::run_main(interactive, codex_linux_sandbox_exe).await?;
+            print_exit_messages(exit_info);
         }
         Some(Subcommand::Login(mut login_cli)) => {
             prepend_config_flags(
@@ -298,7 +373,13 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
                             login_cli.client_id,
                         )
                         .await;
-                    } else if let Some(api_key) = login_cli.api_key {
+                    } else if login_cli.api_key.is_some() {
+                        eprintln!(
+                            "The --api-key flag is no longer supported. Pipe the key instead, e.g. `printenv OPENAI_API_KEY | codex login --with-api-key`."
+                        );
+                        std::process::exit(1);
+                    } else if login_cli.with_api_key {
+                        let api_key = read_api_key_from_stdin();
                         run_login_with_api_key(login_cli.config_overrides, api_key).await;
                     } else {
                         run_login_with_chatgpt(login_cli.config_overrides).await;
@@ -323,8 +404,8 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
             );
             codex_cloud_tasks::run_main(cloud_cli, codex_linux_sandbox_exe).await?;
         }
-        Some(Subcommand::Debug(debug_args)) => match debug_args.cmd {
-            DebugCommand::Seatbelt(mut seatbelt_cli) => {
+        Some(Subcommand::Sandbox(sandbox_args)) => match sandbox_args.cmd {
+            SandboxCommand::Macos(mut seatbelt_cli) => {
                 prepend_config_flags(
                     &mut seatbelt_cli.config_overrides,
                     root_config_overrides.clone(),
@@ -335,7 +416,7 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
                 )
                 .await?;
             }
-            DebugCommand::Landlock(mut landlock_cli) => {
+            SandboxCommand::Linux(mut landlock_cli) => {
                 prepend_config_flags(
                     &mut landlock_cli.config_overrides,
                     root_config_overrides.clone(),
@@ -361,6 +442,30 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
         Some(Subcommand::GenerateTs(gen_cli)) => {
             codex_protocol_ts::generate_ts(&gen_cli.out_dir, gen_cli.prettier.as_deref())?;
         }
+        Some(Subcommand::Features(FeaturesCli { sub })) => match sub {
+            FeaturesSubcommand::List => {
+                // Respect root-level `-c` overrides plus top-level flags like `--profile`.
+                let cli_kv_overrides = root_config_overrides
+                    .parse_overrides()
+                    .map_err(|e| anyhow::anyhow!(e))?;
+
+                // Thread through relevant top-level flags (at minimum, `--profile`).
+                // Also honor `--search` since it maps to a feature toggle.
+                let overrides = ConfigOverrides {
+                    config_profile: interactive.config_profile.clone(),
+                    tools_web_search_request: interactive.web_search.then_some(true),
+                    ..Default::default()
+                };
+
+                let config = Config::load_with_cli_overrides(cli_kv_overrides, overrides).await?;
+                for def in codex_core::features::FEATURES.iter() {
+                    let name = def.key;
+                    let stage = stage_str(def.stage);
+                    let enabled = config.features.enabled(def.id);
+                    println!("{name}\t{stage}\t{enabled}");
+                }
+            }
+        },
     }
 
     Ok(())
@@ -454,6 +559,7 @@ fn print_completion(cmd: CompletionCommand) {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use assert_matches::assert_matches;
     use codex_core::protocol::TokenUsage;
     use codex_protocol::ConversationId;
 
@@ -463,6 +569,7 @@ mod tests {
             interactive,
             config_overrides: root_overrides,
             subcommand,
+            feature_toggles: _,
         } = cli;
 
         let Subcommand::Resume(ResumeCommand {
@@ -586,14 +693,14 @@ mod tests {
         assert_eq!(interactive.model.as_deref(), Some("gpt-5-test"));
         assert!(interactive.oss);
         assert_eq!(interactive.config_profile.as_deref(), Some("my-profile"));
-        assert!(matches!(
+        assert_matches!(
             interactive.sandbox_mode,
             Some(codex_common::SandboxModeCliArg::WorkspaceWrite)
-        ));
-        assert!(matches!(
+        );
+        assert_matches!(
             interactive.approval_policy,
             Some(codex_common::ApprovalModeCliArg::OnRequest)
-        ));
+        );
         assert!(interactive.full_auto);
         assert_eq!(
             interactive.cwd.as_deref(),

codex-rs/cli/src/mcp_cmd.rs

@@ -4,6 +4,7 @@ use anyhow::Context;
 use anyhow::Result;
 use anyhow::anyhow;
 use anyhow::bail;
+use clap::ArgGroup;
 use codex_common::CliConfigOverrides;
 use codex_core::config::Config;
 use codex_core::config::ConfigOverrides;
@@ -12,6 +13,12 @@ use codex_core::config::load_global_mcp_servers;
 use codex_core::config::write_global_mcp_servers;
 use codex_core::config_types::McpServerConfig;
 use codex_core::config_types::McpServerTransportConfig;
+use codex_core::features::Feature;
+use codex_core::mcp::auth::compute_auth_statuses;
+use codex_core::protocol::McpAuthStatus;
+use codex_rmcp_client::delete_oauth_tokens;
+use codex_rmcp_client::perform_oauth_login;
+use codex_rmcp_client::supports_oauth_login;
 
 /// [experimental] Launch Codex as an MCP server or manage configured MCP servers.
 ///
@@ -43,6 +50,14 @@ pub enum McpSubcommand {
 
     /// [experimental] Remove a global MCP server entry.
     Remove(RemoveArgs),
+
+    /// [experimental] Authenticate with a configured MCP server via OAuth.
+    /// Requires experimental_use_rmcp_client = true in config.toml.
+    Login(LoginArgs),
+
+    /// [experimental] Remove stored OAuth credentials for a server.
+    /// Requires experimental_use_rmcp_client = true in config.toml.
+    Logout(LogoutArgs),
 }
 
 #[derive(Debug, clap::Parser)]
@@ -67,13 +82,61 @@ pub struct AddArgs {
     /// Name for the MCP server configuration.
     pub name: String,
 
-    /// Environment variables to set when launching the server.
-    #[arg(long, value_parser = parse_env_pair, value_name = "KEY=VALUE")]
-    pub env: Vec<(String, String)>,
+    #[command(flatten)]
+    pub transport_args: AddMcpTransportArgs,
+}
 
+#[derive(Debug, clap::Args)]
+#[command(
+    group(
+        ArgGroup::new("transport")
+            .args(["command", "url"])
+            .required(true)
+            .multiple(false)
+    )
+)]
+pub struct AddMcpTransportArgs {
+    #[command(flatten)]
+    pub stdio: Option<AddMcpStdioArgs>,
+
+    #[command(flatten)]
+    pub streamable_http: Option<AddMcpStreamableHttpArgs>,
+}
+
+#[derive(Debug, clap::Args)]
+pub struct AddMcpStdioArgs {
     /// Command to launch the MCP server.
-    #[arg(trailing_var_arg = true, num_args = 1..)]
+    /// Use --url for a streamable HTTP server.
+    #[arg(
+            trailing_var_arg = true,
+            num_args = 0..,
+        )]
     pub command: Vec<String>,
+
+    /// Environment variables to set when launching the server.
+    /// Only valid with stdio servers.
+    #[arg(
+        long,
+        value_parser = parse_env_pair,
+        value_name = "KEY=VALUE",
+    )]
+    pub env: Vec<(String, String)>,
+}
+
+#[derive(Debug, clap::Args)]
+pub struct AddMcpStreamableHttpArgs {
+    /// URL for a streamable HTTP MCP server.
+    #[arg(long)]
+    pub url: String,
+
+    /// Optional environment variable to read for a bearer token.
+    /// Only valid with streamable HTTP servers.
+    #[arg(
+        long = "bearer-token-env-var",
+        value_name = "ENV_VAR",
+        requires = "url"
+    )]
+    pub bearer_token_env_var: Option<String>,
 }
 
 #[derive(Debug, clap::Parser)]
@@ -82,6 +145,18 @@ pub struct RemoveArgs {
     pub name: String,
 }
 
+#[derive(Debug, clap::Parser)]
+pub struct LoginArgs {
+    /// Name of the MCP server to authenticate with oauth.
+    pub name: String,
+}
+
+#[derive(Debug, clap::Parser)]
+pub struct LogoutArgs {
+    /// Name of the MCP server to deauthenticate.
+    pub name: String,
+}
+
 impl McpCli {
     pub async fn run(self) -> Result<()> {
         let McpCli {
@@ -91,16 +166,22 @@ impl McpCli {
 
         match subcommand {
             McpSubcommand::List(args) => {
-                run_list(&config_overrides, args)?;
+                run_list(&config_overrides, args).await?;
             }
             McpSubcommand::Get(args) => {
-                run_get(&config_overrides, args)?;
+                run_get(&config_overrides, args).await?;
             }
             McpSubcommand::Add(args) => {
-                run_add(&config_overrides, args)?;
+                run_add(&config_overrides, args).await?;
             }
             McpSubcommand::Remove(args) => {
-                run_remove(&config_overrides, args)?;
+                run_remove(&config_overrides, args).await?;
+            }
+            McpSubcommand::Login(args) => {
+                run_login(&config_overrides, args).await?;
+            }
+            McpSubcommand::Logout(args) => {
+                run_logout(&config_overrides, args).await?;
             }
         }
 
@@ -108,40 +189,63 @@ impl McpCli {
     }
 }
 
-fn run_add(config_overrides: &CliConfigOverrides, add_args: AddArgs) -> Result<()> {
+async fn run_add(config_overrides: &CliConfigOverrides, add_args: AddArgs) -> Result<()> {
     // Validate any provided overrides even though they are not currently applied.
-    config_overrides.parse_overrides().map_err(|e| anyhow!(e))?;
+    let overrides = config_overrides.parse_overrides().map_err(|e| anyhow!(e))?;
+    let config = Config::load_with_cli_overrides(overrides, ConfigOverrides::default())
+        .await
+        .context("failed to load configuration")?;
 
-    let AddArgs { name, env, command } = add_args;
+    let AddArgs {
+        name,
+        transport_args,
+    } = add_args;
 
     validate_server_name(&name)?;
 
-    let mut command_parts = command.into_iter();
-    let command_bin = command_parts
-        .next()
-        .ok_or_else(|| anyhow!("command is required"))?;
-    let command_args: Vec<String> = command_parts.collect();
-
-    let env_map = if env.is_empty() {
-        None
-    } else {
-        let mut map = HashMap::new();
-        for (key, value) in env {
-            map.insert(key, value);
-        }
-        Some(map)
-    };
-
     let codex_home = find_codex_home().context("failed to resolve CODEX_HOME")?;
     let mut servers = load_global_mcp_servers(&codex_home)
+        .await
         .with_context(|| format!("failed to load MCP servers from {}", codex_home.display()))?;
 
-    let new_entry = McpServerConfig {
-        transport: McpServerTransportConfig::Stdio {
-            command: command_bin,
-            args: command_args,
-            env: env_map,
+    let transport = match transport_args {
+        AddMcpTransportArgs {
+            stdio: Some(stdio), ..
+        } => {
+            let mut command_parts = stdio.command.into_iter();
+            let command_bin = command_parts
+                .next()
+                .ok_or_else(|| anyhow!("command is required"))?;
+            let command_args: Vec<String> = command_parts.collect();
+
+            let env_map = if stdio.env.is_empty() {
+                None
+            } else {
+                Some(stdio.env.into_iter().collect::<HashMap<_, _>>())
+            };
+            McpServerTransportConfig::Stdio {
+                command: command_bin,
+                args: command_args,
+                env: env_map,
+            }
+        }
+        AddMcpTransportArgs {
+            streamable_http:
+                Some(AddMcpStreamableHttpArgs {
+                    url,
+                    bearer_token_env_var,
+                }),
+            ..
+        } => McpServerTransportConfig::StreamableHttp {
+            url,
+            bearer_token_env_var,
         },
+        AddMcpTransportArgs { .. } => bail!("exactly one of --command or --url must be provided"),
+    };
+
+    let new_entry = McpServerConfig {
+        transport: transport.clone(),
+        enabled: true,
         startup_timeout_sec: None,
         tool_timeout_sec: None,
     };
@@ -153,10 +257,21 @@ fn run_add(config_overrides: &CliConfigOverrides, add_args: AddArgs) -> Result<(
 
     println!("Added global MCP server '{name}'.");
 
+    if let McpServerTransportConfig::StreamableHttp {
+        url,
+        bearer_token_env_var: None,
+    } = transport
+        && matches!(supports_oauth_login(&url).await, Ok(true))
+    {
+        println!("Detected OAuth support. Starting OAuth flow…");
+        perform_oauth_login(&name, &url, config.mcp_oauth_credentials_store_mode).await?;
+        println!("Successfully logged in.");
+    }
+
     Ok(())
 }
 
-fn run_remove(config_overrides: &CliConfigOverrides, remove_args: RemoveArgs) -> Result<()> {
+async fn run_remove(config_overrides: &CliConfigOverrides, remove_args: RemoveArgs) -> Result<()> {
     config_overrides.parse_overrides().map_err(|e| anyhow!(e))?;
 
     let RemoveArgs { name } = remove_args;
@@ -165,6 +280,7 @@ fn run_remove(config_overrides: &CliConfigOverrides, remove_args: RemoveArgs) ->
 
     let codex_home = find_codex_home().context("failed to resolve CODEX_HOME")?;
     let mut servers = load_global_mcp_servers(&codex_home)
+        .await
         .with_context(|| format!("failed to load MCP servers from {}", codex_home.display()))?;
 
     let removed = servers.remove(&name).is_some();
@@ -183,18 +299,83 @@ fn run_remove(config_overrides: &CliConfigOverrides, remove_args: RemoveArgs) ->
     Ok(())
 }
 
-fn run_list(config_overrides: &CliConfigOverrides, list_args: ListArgs) -> Result<()> {
+async fn run_login(config_overrides: &CliConfigOverrides, login_args: LoginArgs) -> Result<()> {
     let overrides = config_overrides.parse_overrides().map_err(|e| anyhow!(e))?;
     let config = Config::load_with_cli_overrides(overrides, ConfigOverrides::default())
+        .await
+        .context("failed to load configuration")?;
+
+    if !config.features.enabled(Feature::RmcpClient) {
+        bail!(
+            "OAuth login is only supported when experimental_use_rmcp_client is true in config.toml."
+        );
+    }
+
+    let LoginArgs { name } = login_args;
+
+    let Some(server) = config.mcp_servers.get(&name) else {
+        bail!("No MCP server named '{name}' found.");
+    };
+
+    let url = match &server.transport {
+        McpServerTransportConfig::StreamableHttp { url, .. } => url.clone(),
+        _ => bail!("OAuth login is only supported for streamable HTTP servers."),
+    };
+
+    perform_oauth_login(&name, &url, config.mcp_oauth_credentials_store_mode).await?;
+    println!("Successfully logged in to MCP server '{name}'.");
+    Ok(())
+}
+
+async fn run_logout(config_overrides: &CliConfigOverrides, logout_args: LogoutArgs) -> Result<()> {
+    let overrides = config_overrides.parse_overrides().map_err(|e| anyhow!(e))?;
+    let config = Config::load_with_cli_overrides(overrides, ConfigOverrides::default())
+        .await
+        .context("failed to load configuration")?;
+
+    let LogoutArgs { name } = logout_args;
+
+    let server = config
+        .mcp_servers
+        .get(&name)
+        .ok_or_else(|| anyhow!("No MCP server named '{name}' found in configuration."))?;
+
+    let url = match &server.transport {
+        McpServerTransportConfig::StreamableHttp { url, .. } => url.clone(),
+        _ => bail!("OAuth logout is only supported for streamable_http transports."),
+    };
+
+    match delete_oauth_tokens(&name, &url, config.mcp_oauth_credentials_store_mode) {
+        Ok(true) => println!("Removed OAuth credentials for '{name}'."),
+        Ok(false) => println!("No OAuth credentials stored for '{name}'."),
+        Err(err) => return Err(anyhow!("failed to delete OAuth credentials: {err}")),
+    }
+
+    Ok(())
+}
+
+async fn run_list(config_overrides: &CliConfigOverrides, list_args: ListArgs) -> Result<()> {
+    let overrides = config_overrides.parse_overrides().map_err(|e| anyhow!(e))?;
+    let config = Config::load_with_cli_overrides(overrides, ConfigOverrides::default())
+        .await
         .context("failed to load configuration")?;
 
     let mut entries: Vec<_> = config.mcp_servers.iter().collect();
     entries.sort_by(|(a, _), (b, _)| a.cmp(b));
+    let auth_statuses = compute_auth_statuses(
+        config.mcp_servers.iter(),
+        config.mcp_oauth_credentials_store_mode,
+    )
+    .await;
 
     if list_args.json {
         let json_entries: Vec<_> = entries
             .into_iter()
             .map(|(name, cfg)| {
+                let auth_status = auth_statuses
+                    .get(name.as_str())
+                    .copied()
+                    .unwrap_or(McpAuthStatus::Unsupported);
                 let transport = match &cfg.transport {
                     McpServerTransportConfig::Stdio { command, args, env } => serde_json::json!({
                         "type": "stdio",
@@ -202,17 +383,21 @@ fn run_list(config_overrides: &CliConfigOverrides, list_args: ListArgs) -> Resul
                         "args": args,
                         "env": env,
                     }),
-                    McpServerTransportConfig::StreamableHttp { url, bearer_token } => {
+                    McpServerTransportConfig::StreamableHttp {
+                        url,
+                        bearer_token_env_var,
+                    } => {
                         serde_json::json!({
                             "type": "streamable_http",
                             "url": url,
-                            "bearer_token": bearer_token,
+                            "bearer_token_env_var": bearer_token_env_var,
                         })
                     }
                 };
 
                 serde_json::json!({
                     "name": name,
+                    "enabled": cfg.enabled,
                     "transport": transport,
                     "startup_timeout_sec": cfg
                         .startup_timeout_sec
@@ -220,6 +405,7 @@ fn run_list(config_overrides: &CliConfigOverrides, list_args: ListArgs) -> Resul
                     "tool_timeout_sec": cfg
                         .tool_timeout_sec
                         .map(|timeout| timeout.as_secs_f64()),
+                    "auth_status": auth_status,
                 })
             })
             .collect();
@@ -233,8 +419,8 @@ fn run_list(config_overrides: &CliConfigOverrides, list_args: ListArgs) -> Resul
         return Ok(());
     }
 
-    let mut stdio_rows: Vec<[String; 4]> = Vec::new();
-    let mut http_rows: Vec<[String; 3]> = Vec::new();
+    let mut stdio_rows: Vec<[String; 6]> = Vec::new();
+    let mut http_rows: Vec<[String; 5]> = Vec::new();
 
     for (name, cfg) in entries {
         match &cfg.transport {
@@ -257,21 +443,59 @@ fn run_list(config_overrides: &CliConfigOverrides, list_args: ListArgs) -> Resul
                             .join(", ")
                     }
                 };
-                stdio_rows.push([name.clone(), command.clone(), args_display, env_display]);
-            }
-            McpServerTransportConfig::StreamableHttp { url, bearer_token } => {
-                let has_bearer = if bearer_token.is_some() {
-                    "True"
+                let status = if cfg.enabled {
+                    "enabled".to_string()
                 } else {
-                    "False"
+                    "disabled".to_string()
                 };
-                http_rows.push([name.clone(), url.clone(), has_bearer.into()]);
+                let auth_status = auth_statuses
+                    .get(name.as_str())
+                    .copied()
+                    .unwrap_or(McpAuthStatus::Unsupported)
+                    .to_string();
+                stdio_rows.push([
+                    name.clone(),
+                    command.clone(),
+                    args_display,
+                    env_display,
+                    status,
+                    auth_status,
+                ]);
+            }
+            McpServerTransportConfig::StreamableHttp {
+                url,
+                bearer_token_env_var,
+            } => {
+                let status = if cfg.enabled {
+                    "enabled".to_string()
+                } else {
+                    "disabled".to_string()
+                };
+                let auth_status = auth_statuses
+                    .get(name.as_str())
+                    .copied()
+                    .unwrap_or(McpAuthStatus::Unsupported)
+                    .to_string();
+                http_rows.push([
+                    name.clone(),
+                    url.clone(),
+                    bearer_token_env_var.clone().unwrap_or("-".to_string()),
+                    status,
+                    auth_status,
+                ]);
             }
         }
     }
 
     if !stdio_rows.is_empty() {
-        let mut widths = ["Name".len(), "Command".len(), "Args".len(), "Env".len()];
+        let mut widths = [
+            "Name".len(),
+            "Command".len(),
+            "Args".len(),
+            "Env".len(),
+            "Status".len(),
+            "Auth".len(),
+        ];
         for row in &stdio_rows {
             for (i, cell) in row.iter().enumerate() {
                 widths[i] = widths[i].max(cell.len());
@@ -279,28 +503,36 @@ fn run_list(config_overrides: &CliConfigOverrides, list_args: ListArgs) -> Resul
         }
 
         println!(
-            "{:<name_w$}  {:<cmd_w$}  {:<args_w$}  {:<env_w$}",
-            "Name",
-            "Command",
-            "Args",
-            "Env",
+            "{name:<name_w$}  {command:<cmd_w$}  {args:<args_w$}  {env:<env_w$}  {status:<status_w$}  {auth:<auth_w$}",
+            name = "Name",
+            command = "Command",
+            args = "Args",
+            env = "Env",
+            status = "Status",
+            auth = "Auth",
             name_w = widths[0],
             cmd_w = widths[1],
             args_w = widths[2],
             env_w = widths[3],
+            status_w = widths[4],
+            auth_w = widths[5],
         );
 
         for row in &stdio_rows {
             println!(
-                "{:<name_w$}  {:<cmd_w$}  {:<args_w$}  {:<env_w$}",
-                row[0],
-                row[1],
-                row[2],
-                row[3],
+                "{name:<name_w$}  {command:<cmd_w$}  {args:<args_w$}  {env:<env_w$}  {status:<status_w$}  {auth:<auth_w$}",
+                name = row[0].as_str(),
+                command = row[1].as_str(),
+                args = row[2].as_str(),
+                env = row[3].as_str(),
+                status = row[4].as_str(),
+                auth = row[5].as_str(),
                 name_w = widths[0],
                 cmd_w = widths[1],
                 args_w = widths[2],
                 env_w = widths[3],
+                status_w = widths[4],
+                auth_w = widths[5],
             );
         }
     }
@@ -310,7 +542,13 @@ fn run_list(config_overrides: &CliConfigOverrides, list_args: ListArgs) -> Resul
     }
 
     if !http_rows.is_empty() {
-        let mut widths = ["Name".len(), "Url".len(), "Has Bearer Token".len()];
+        let mut widths = [
+            "Name".len(),
+            "Url".len(),
+            "Bearer Token Env Var".len(),
+            "Status".len(),
+            "Auth".len(),
+        ];
         for row in &http_rows {
             for (i, cell) in row.iter().enumerate() {
                 widths[i] = widths[i].max(cell.len());
@@ -318,24 +556,32 @@ fn run_list(config_overrides: &CliConfigOverrides, list_args: ListArgs) -> Resul
         }
 
         println!(
-            "{:<name_w$}  {:<url_w$}  {:<token_w$}",
-            "Name",
-            "Url",
-            "Has Bearer Token",
+            "{name:<name_w$}  {url:<url_w$}  {token:<token_w$}  {status:<status_w$}  {auth:<auth_w$}",
+            name = "Name",
+            url = "Url",
+            token = "Bearer Token Env Var",
+            status = "Status",
+            auth = "Auth",
             name_w = widths[0],
             url_w = widths[1],
             token_w = widths[2],
+            status_w = widths[3],
+            auth_w = widths[4],
         );
 
         for row in &http_rows {
             println!(
-                "{:<name_w$}  {:<url_w$}  {:<token_w$}",
-                row[0],
-                row[1],
-                row[2],
+                "{name:<name_w$}  {url:<url_w$}  {token:<token_w$}  {status:<status_w$}  {auth:<auth_w$}",
+                name = row[0].as_str(),
+                url = row[1].as_str(),
+                token = row[2].as_str(),
+                status = row[3].as_str(),
+                auth = row[4].as_str(),
                 name_w = widths[0],
                 url_w = widths[1],
                 token_w = widths[2],
+                status_w = widths[3],
+                auth_w = widths[4],
             );
         }
     }
@@ -343,9 +589,10 @@ fn run_list(config_overrides: &CliConfigOverrides, list_args: ListArgs) -> Resul
     Ok(())
 }
 
-fn run_get(config_overrides: &CliConfigOverrides, get_args: GetArgs) -> Result<()> {
+async fn run_get(config_overrides: &CliConfigOverrides, get_args: GetArgs) -> Result<()> {
     let overrides = config_overrides.parse_overrides().map_err(|e| anyhow!(e))?;
     let config = Config::load_with_cli_overrides(overrides, ConfigOverrides::default())
+        .await
         .context("failed to load configuration")?;
 
     let Some(server) = config.mcp_servers.get(&get_args.name) else {
@@ -360,14 +607,18 @@ fn run_get(config_overrides: &CliConfigOverrides, get_args: GetArgs) -> Result<(
                 "args": args,
                 "env": env,
             }),
-            McpServerTransportConfig::StreamableHttp { url, bearer_token } => serde_json::json!({
+            McpServerTransportConfig::StreamableHttp {
+                url,
+                bearer_token_env_var,
+            } => serde_json::json!({
                 "type": "streamable_http",
                 "url": url,
-                "bearer_token": bearer_token,
+                "bearer_token_env_var": bearer_token_env_var,
             }),
         };
         let output = serde_json::to_string_pretty(&serde_json::json!({
             "name": get_args.name,
+            "enabled": server.enabled,
             "transport": transport,
             "startup_timeout_sec": server
                 .startup_timeout_sec
@@ -381,6 +632,7 @@ fn run_get(config_overrides: &CliConfigOverrides, get_args: GetArgs) -> Result<(
     }
 
     println!("{}", get_args.name);
+    println!("  enabled: {}", server.enabled);
     match &server.transport {
         McpServerTransportConfig::Stdio { command, args, env } => {
             println!("  transport: stdio");
@@ -406,11 +658,14 @@ fn run_get(config_overrides: &CliConfigOverrides, get_args: GetArgs) -> Result<(
             };
             println!("  env: {env_display}");
         }
-        McpServerTransportConfig::StreamableHttp { url, bearer_token } => {
+        McpServerTransportConfig::StreamableHttp {
+            url,
+            bearer_token_env_var,
+        } => {
             println!("  transport: streamable_http");
             println!("  url: {url}");
-            let bearer = bearer_token.as_deref().unwrap_or("-");
-            println!("  bearer_token: {bearer}");
+            let env_var = bearer_token_env_var.as_deref().unwrap_or("-");
+            println!("  bearer_token_env_var: {env_var}");
         }
     }
     if let Some(timeout) = server.startup_timeout_sec {

codex-rs/cli/tests/mcp_add_remove.rs

@@ -13,8 +13,8 @@ fn codex_command(codex_home: &Path) -> Result<assert_cmd::Command> {
     Ok(cmd)
 }
 
-#[test]
-fn add_and_remove_server_updates_global_config() -> Result<()> {
+#[tokio::test]
+async fn add_and_remove_server_updates_global_config() -> Result<()> {
     let codex_home = TempDir::new()?;
 
     let mut add_cmd = codex_command(codex_home.path())?;
@@ -24,7 +24,7 @@ fn add_and_remove_server_updates_global_config() -> Result<()> {
         .success()
         .stdout(contains("Added global MCP server 'docs'."));
 
-    let servers = load_global_mcp_servers(codex_home.path())?;
+    let servers = load_global_mcp_servers(codex_home.path()).await?;
     assert_eq!(servers.len(), 1);
     let docs = servers.get("docs").expect("server should exist");
     match &docs.transport {
@@ -35,6 +35,7 @@ fn add_and_remove_server_updates_global_config() -> Result<()> {
         }
         other => panic!("unexpected transport: {other:?}"),
     }
+    assert!(docs.enabled);
 
     let mut remove_cmd = codex_command(codex_home.path())?;
     remove_cmd
@@ -43,7 +44,7 @@ fn add_and_remove_server_updates_global_config() -> Result<()> {
         .success()
         .stdout(contains("Removed global MCP server 'docs'."));
 
-    let servers = load_global_mcp_servers(codex_home.path())?;
+    let servers = load_global_mcp_servers(codex_home.path()).await?;
     assert!(servers.is_empty());
 
     let mut remove_again_cmd = codex_command(codex_home.path())?;
@@ -53,14 +54,14 @@ fn add_and_remove_server_updates_global_config() -> Result<()> {
         .success()
         .stdout(contains("No MCP server named 'docs' found."));
 
-    let servers = load_global_mcp_servers(codex_home.path())?;
+    let servers = load_global_mcp_servers(codex_home.path()).await?;
     assert!(servers.is_empty());
 
     Ok(())
 }
 
-#[test]
-fn add_with_env_preserves_key_order_and_values() -> Result<()> {
+#[tokio::test]
+async fn add_with_env_preserves_key_order_and_values() -> Result<()> {
     let codex_home = TempDir::new()?;
 
     let mut add_cmd = codex_command(codex_home.path())?;
@@ -80,7 +81,7 @@ fn add_with_env_preserves_key_order_and_values() -> Result<()> {
         .assert()
         .success();
 
-    let servers = load_global_mcp_servers(codex_home.path())?;
+    let servers = load_global_mcp_servers(codex_home.path()).await?;
     let envy = servers.get("envy").expect("server should exist");
     let env = match &envy.transport {
         McpServerTransportConfig::Stdio { env: Some(env), .. } => env,
@@ -90,6 +91,122 @@ fn add_with_env_preserves_key_order_and_values() -> Result<()> {
     assert_eq!(env.len(), 2);
     assert_eq!(env.get("FOO"), Some(&"bar".to_string()));
     assert_eq!(env.get("ALPHA"), Some(&"beta".to_string()));
+    assert!(envy.enabled);
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn add_streamable_http_without_manual_token() -> Result<()> {
+    let codex_home = TempDir::new()?;
+
+    let mut add_cmd = codex_command(codex_home.path())?;
+    add_cmd
+        .args(["mcp", "add", "github", "--url", "https://example.com/mcp"])
+        .assert()
+        .success();
+
+    let servers = load_global_mcp_servers(codex_home.path()).await?;
+    let github = servers.get("github").expect("github server should exist");
+    match &github.transport {
+        McpServerTransportConfig::StreamableHttp {
+            url,
+            bearer_token_env_var,
+        } => {
+            assert_eq!(url, "https://example.com/mcp");
+            assert!(bearer_token_env_var.is_none());
+        }
+        other => panic!("unexpected transport: {other:?}"),
+    }
+    assert!(github.enabled);
+
+    assert!(!codex_home.path().join(".credentials.json").exists());
+    assert!(!codex_home.path().join(".env").exists());
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn add_streamable_http_with_custom_env_var() -> Result<()> {
+    let codex_home = TempDir::new()?;
+
+    let mut add_cmd = codex_command(codex_home.path())?;
+    add_cmd
+        .args([
+            "mcp",
+            "add",
+            "issues",
+            "--url",
+            "https://example.com/issues",
+            "--bearer-token-env-var",
+            "GITHUB_TOKEN",
+        ])
+        .assert()
+        .success();
+
+    let servers = load_global_mcp_servers(codex_home.path()).await?;
+    let issues = servers.get("issues").expect("issues server should exist");
+    match &issues.transport {
+        McpServerTransportConfig::StreamableHttp {
+            url,
+            bearer_token_env_var,
+        } => {
+            assert_eq!(url, "https://example.com/issues");
+            assert_eq!(bearer_token_env_var.as_deref(), Some("GITHUB_TOKEN"));
+        }
+        other => panic!("unexpected transport: {other:?}"),
+    }
+    assert!(issues.enabled);
+    Ok(())
+}
+
+#[tokio::test]
+async fn add_streamable_http_rejects_removed_flag() -> Result<()> {
+    let codex_home = TempDir::new()?;
+
+    let mut add_cmd = codex_command(codex_home.path())?;
+    add_cmd
+        .args([
+            "mcp",
+            "add",
+            "github",
+            "--url",
+            "https://example.com/mcp",
+            "--with-bearer-token",
+        ])
+        .assert()
+        .failure()
+        .stderr(contains("--with-bearer-token"));
+
+    let servers = load_global_mcp_servers(codex_home.path()).await?;
+    assert!(servers.is_empty());
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn add_cant_add_command_and_url() -> Result<()> {
+    let codex_home = TempDir::new()?;
+
+    let mut add_cmd = codex_command(codex_home.path())?;
+    add_cmd
+        .args([
+            "mcp",
+            "add",
+            "github",
+            "--url",
+            "https://example.com/mcp",
+            "--command",
+            "--",
+            "echo",
+            "hello",
+        ])
+        .assert()
+        .failure()
+        .stderr(contains("unexpected argument '--command' found"));
+
+    let servers = load_global_mcp_servers(codex_home.path()).await?;
+    assert!(servers.is_empty());
 
     Ok(())
 }

codex-rs/cli/tests/mcp_list.rs

@@ -1,6 +1,7 @@
 use std::path::Path;
 
 use anyhow::Result;
+use predicates::prelude::PredicateBooleanExt;
 use predicates::str::contains;
 use pretty_assertions::assert_eq;
 use serde_json::Value as JsonValue;
@@ -53,6 +54,10 @@ fn list_and_get_render_expected_output() -> Result<()> {
     assert!(stdout.contains("docs"));
     assert!(stdout.contains("docs-server"));
     assert!(stdout.contains("TOKEN=secret"));
+    assert!(stdout.contains("Status"));
+    assert!(stdout.contains("Auth"));
+    assert!(stdout.contains("enabled"));
+    assert!(stdout.contains("Unsupported"));
 
     let mut list_json_cmd = codex_command(codex_home.path())?;
     let json_output = list_json_cmd.args(["mcp", "list", "--json"]).output()?;
@@ -64,6 +69,7 @@ fn list_and_get_render_expected_output() -> Result<()> {
         json!([
           {
             "name": "docs",
+            "enabled": true,
             "transport": {
               "type": "stdio",
               "command": "docs-server",
@@ -76,7 +82,8 @@ fn list_and_get_render_expected_output() -> Result<()> {
               }
             },
             "startup_timeout_sec": null,
-            "tool_timeout_sec": null
+            "tool_timeout_sec": null,
+            "auth_status": "unsupported"
           }
         ]
         )
@@ -91,6 +98,7 @@ fn list_and_get_render_expected_output() -> Result<()> {
     assert!(stdout.contains("command: docs-server"));
     assert!(stdout.contains("args: --port 4000"));
     assert!(stdout.contains("env: TOKEN=secret"));
+    assert!(stdout.contains("enabled: true"));
     assert!(stdout.contains("remove: codex mcp remove docs"));
 
     let mut get_json_cmd = codex_command(codex_home.path())?;
@@ -98,7 +106,7 @@ fn list_and_get_render_expected_output() -> Result<()> {
         .args(["mcp", "get", "docs", "--json"])
         .assert()
         .success()
-        .stdout(contains("\"name\": \"docs\""));
+        .stdout(contains("\"name\": \"docs\"").and(contains("\"enabled\": true")));
 
     Ok(())
 }

codex-rs/cloud-tasks/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
+edition = "2024"
 name = "codex-cloud-tasks"
 version = { workspace = true }
-edition = "2024"
 
 [lib]
 name = "codex_cloud_tasks"
@@ -11,26 +11,28 @@ path = "src/lib.rs"
 workspace = true
 
 [dependencies]
-anyhow = "1"
-clap = { version = "4", features = ["derive"] }
+anyhow = { workspace = true }
+base64 = { workspace = true }
+chrono = { workspace = true, features = ["serde"] }
+clap = { workspace = true, features = ["derive"] }
+codex-cloud-tasks-client = { path = "../cloud-tasks-client", features = [
+    "mock",
+    "online",
+] }
 codex-common = { path = "../common", features = ["cli"] }
-tokio = { version = "1", features = ["macros", "rt-multi-thread"] }
-tracing = { version = "0.1.41", features = ["log"] }
-tracing-subscriber = { version = "0.3.19", features = ["env-filter"] }
-codex-cloud-tasks-client = { path = "../cloud-tasks-client", features = ["mock", "online"] }
-ratatui = { version = "0.29.0" }
-crossterm = { version = "0.28.1", features = ["event-stream"] }
-tokio-stream = "0.1.17"
-chrono = { version = "0.4", features = ["serde"] }
-codex-login = { path = "../login" }
 codex-core = { path = "../core" }
-throbber-widgets-tui = "0.8.0"
-base64 = "0.22"
-serde_json = "1"
-reqwest = { version = "0.12", features = ["json"] }
-serde = { version = "1", features = ["derive"] }
-unicode-width = "0.1"
+codex-login = { path = "../login" }
 codex-tui = { path = "../tui" }
+crossterm = { workspace = true, features = ["event-stream"] }
+ratatui = { workspace = true }
+reqwest = { workspace = true, features = ["json"] }
+serde = { workspace = true, features = ["derive"] }
+serde_json = { workspace = true }
+tokio = { workspace = true, features = ["macros", "rt-multi-thread"] }
+tokio-stream = { workspace = true }
+tracing = { workspace = true, features = ["log"] }
+tracing-subscriber = { workspace = true, features = ["env-filter"] }
+unicode-width = { workspace = true }
 
 [dev-dependencies]
-async-trait = "0.1"
+async-trait = { workspace = true }

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

@@ -1,4 +1,5 @@
 use std::time::Duration;
+use std::time::Instant;
 
 // Environment filter data models for the TUI
 #[derive(Clone, Debug, Default)]
@@ -42,15 +43,13 @@ use crate::scrollable_diff::ScrollableDiff;
 use codex_cloud_tasks_client::CloudBackend;
 use codex_cloud_tasks_client::TaskId;
 use codex_cloud_tasks_client::TaskSummary;
-use throbber_widgets_tui::ThrobberState;
-
 #[derive(Default)]
 pub struct App {
     pub tasks: Vec<TaskSummary>,
     pub selected: usize,
     pub status: String,
     pub diff_overlay: Option<DiffOverlay>,
-    pub throbber: ThrobberState,
+    pub spinner_start: Option<Instant>,
     pub refresh_inflight: bool,
     pub details_inflight: bool,
     // Environment filter state
@@ -82,7 +81,7 @@ impl App {
             selected: 0,
             status: "Press r to refresh".to_string(),
             diff_overlay: None,
-            throbber: ThrobberState::default(),
+            spinner_start: None,
             refresh_inflight: false,
             details_inflight: false,
             env_filter: None,

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

@@ -1,3 +1,4 @@
+use clap::Args;
 use clap::Parser;
 use codex_common::CliConfigOverrides;
 
@@ -6,4 +7,43 @@ use codex_common::CliConfigOverrides;
 pub struct Cli {
     #[clap(skip)]
     pub config_overrides: CliConfigOverrides,
+
+    #[command(subcommand)]
+    pub command: Option<Command>,
+}
+
+#[derive(Debug, clap::Subcommand)]
+pub enum Command {
+    /// Submit a new Codex Cloud task without launching the TUI.
+    Exec(ExecCommand),
+}
+
+#[derive(Debug, Args)]
+pub struct ExecCommand {
+    /// Task prompt to run in Codex Cloud.
+    #[arg(value_name = "QUERY")]
+    pub query: Option<String>,
+
+    /// Target environment identifier (see `codex cloud` to browse).
+    #[arg(long = "env", value_name = "ENV_ID")]
+    pub environment: String,
+
+    /// Number of assistant attempts (best-of-N).
+    #[arg(
+        long = "attempts",
+        default_value_t = 1usize,
+        value_parser = parse_attempts
+    )]
+    pub attempts: usize,
+}
+
+fn parse_attempts(input: &str) -> Result<usize, String> {
+    let value: usize = input
+        .parse()
+        .map_err(|_| "attempts must be an integer between 1 and 4".to_string())?;
+    if (1..=4).contains(&value) {
+        Ok(value)
+    } else {
+        Err("attempts must be between 1 and 4".to_string())
+    }
 }

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

@@ -7,7 +7,9 @@ mod ui;
 pub mod util;
 pub use cli::Cli;
 
+use anyhow::anyhow;
 use std::io::IsTerminal;
+use std::io::Read;
 use std::path::PathBuf;
 use std::sync::Arc;
 use std::time::Duration;
@@ -23,6 +25,175 @@ struct ApplyJob {
     diff_override: Option<String>,
 }
 
+struct BackendContext {
+    backend: Arc<dyn codex_cloud_tasks_client::CloudBackend>,
+    base_url: String,
+}
+
+async fn init_backend(user_agent_suffix: &str) -> anyhow::Result<BackendContext> {
+    let use_mock = matches!(
+        std::env::var("CODEX_CLOUD_TASKS_MODE").ok().as_deref(),
+        Some("mock") | Some("MOCK")
+    );
+    let base_url = std::env::var("CODEX_CLOUD_TASKS_BASE_URL")
+        .unwrap_or_else(|_| "https://chatgpt.com/backend-api".to_string());
+
+    set_user_agent_suffix(user_agent_suffix);
+
+    if use_mock {
+        return Ok(BackendContext {
+            backend: Arc::new(codex_cloud_tasks_client::MockClient),
+            base_url,
+        });
+    }
+
+    let ua = codex_core::default_client::get_codex_user_agent();
+    let mut http = codex_cloud_tasks_client::HttpClient::new(base_url.clone())?.with_user_agent(ua);
+    let style = if base_url.contains("/backend-api") {
+        "wham"
+    } else {
+        "codex-api"
+    };
+    append_error_log(format!("startup: base_url={base_url} path_style={style}"));
+
+    let auth = match codex_core::config::find_codex_home()
+        .ok()
+        .map(|home| codex_login::AuthManager::new(home, false))
+        .and_then(|am| am.auth())
+    {
+        Some(auth) => auth,
+        None => {
+            eprintln!(
+                "Not signed in. Please run 'codex login' to sign in with ChatGPT, then re-run 'codex cloud'."
+            );
+            std::process::exit(1);
+        }
+    };
+
+    if let Some(acc) = auth.get_account_id() {
+        append_error_log(format!("auth: mode=ChatGPT account_id={acc}"));
+    }
+
+    let token = match auth.get_token().await {
+        Ok(t) if !t.is_empty() => t,
+        _ => {
+            eprintln!(
+                "Not signed in. Please run 'codex login' to sign in with ChatGPT, then re-run 'codex cloud'."
+            );
+            std::process::exit(1);
+        }
+    };
+
+    http = http.with_bearer_token(token.clone());
+    if let Some(acc) = auth
+        .get_account_id()
+        .or_else(|| util::extract_chatgpt_account_id(&token))
+    {
+        append_error_log(format!("auth: set ChatGPT-Account-Id header: {acc}"));
+        http = http.with_chatgpt_account_id(acc);
+    }
+
+    Ok(BackendContext {
+        backend: Arc::new(http),
+        base_url,
+    })
+}
+
+async fn run_exec_command(args: crate::cli::ExecCommand) -> anyhow::Result<()> {
+    let crate::cli::ExecCommand {
+        query,
+        environment,
+        attempts,
+    } = args;
+    let ctx = init_backend("codex_cloud_tasks_exec").await?;
+    let prompt = resolve_query_input(query)?;
+    let env_id = resolve_environment_id(&ctx, &environment).await?;
+    let created = codex_cloud_tasks_client::CloudBackend::create_task(
+        &*ctx.backend,
+        &env_id,
+        &prompt,
+        "main",
+        false,
+        attempts,
+    )
+    .await?;
+    let url = util::task_url(&ctx.base_url, &created.id.0);
+    println!("{url}");
+    Ok(())
+}
+
+async fn resolve_environment_id(ctx: &BackendContext, requested: &str) -> anyhow::Result<String> {
+    let trimmed = requested.trim();
+    if trimmed.is_empty() {
+        return Err(anyhow!("environment id must not be empty"));
+    }
+    let normalized = util::normalize_base_url(&ctx.base_url);
+    let headers = util::build_chatgpt_headers().await;
+    let environments = crate::env_detect::list_environments(&normalized, &headers).await?;
+    if environments.is_empty() {
+        return Err(anyhow!(
+            "no cloud environments are available for this workspace"
+        ));
+    }
+
+    if let Some(row) = environments.iter().find(|row| row.id == trimmed) {
+        return Ok(row.id.clone());
+    }
+
+    let label_matches = environments
+        .iter()
+        .filter(|row| {
+            row.label
+                .as_deref()
+                .map(|label| label.eq_ignore_ascii_case(trimmed))
+                .unwrap_or(false)
+        })
+        .collect::<Vec<_>>();
+    match label_matches.as_slice() {
+        [] => Err(anyhow!(
+            "environment '{trimmed}' not found; run `codex cloud` to list available environments"
+        )),
+        [single] => Ok(single.id.clone()),
+        [first, rest @ ..] => {
+            let first_id = &first.id;
+            if rest.iter().all(|row| row.id == *first_id) {
+                Ok(first_id.clone())
+            } else {
+                Err(anyhow!(
+                    "environment label '{trimmed}' is ambiguous; run `codex cloud` to pick the desired environment id"
+                ))
+            }
+        }
+    }
+}
+
+fn resolve_query_input(query_arg: Option<String>) -> anyhow::Result<String> {
+    match query_arg {
+        Some(q) if q != "-" => Ok(q),
+        maybe_dash => {
+            let force_stdin = matches!(maybe_dash.as_deref(), Some("-"));
+            if std::io::stdin().is_terminal() && !force_stdin {
+                return Err(anyhow!(
+                    "no query provided. Pass one as an argument or pipe it via stdin."
+                ));
+            }
+            if !force_stdin {
+                eprintln!("Reading query from stdin...");
+            }
+            let mut buffer = String::new();
+            std::io::stdin()
+                .read_to_string(&mut buffer)
+                .map_err(|e| anyhow!("failed to read query from stdin: {e}"))?;
+            if buffer.trim().is_empty() {
+                return Err(anyhow!(
+                    "no query provided via stdin (received empty input)."
+                ));
+            }
+            Ok(buffer)
+        }
+    }
+}
+
 fn level_from_status(status: codex_cloud_tasks_client::ApplyStatus) -> app::ApplyResultLevel {
     match status {
         codex_cloud_tasks_client::ApplyStatus::Success => app::ApplyResultLevel::Success,
@@ -148,7 +319,14 @@ fn spawn_apply(
 // (no standalone patch summarizer needed – UI displays raw diffs)
 
 /// Entry point for the `codex cloud` subcommand.
-pub async fn run_main(_cli: Cli, _codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()> {
+pub async fn run_main(cli: Cli, _codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()> {
+    if let Some(command) = cli.command {
+        return match command {
+            crate::cli::Command::Exec(args) => run_exec_command(args).await,
+        };
+    }
+    let Cli { .. } = cli;
+
     // Very minimal logging setup; mirrors other crates' pattern.
     let default_level = "error";
     let _ = tracing_subscriber::fmt()
@@ -162,72 +340,8 @@ pub async fn run_main(_cli: Cli, _codex_linux_sandbox_exe: Option<PathBuf>) -> a
         .try_init();
 
     info!("Launching Cloud Tasks list UI");
-    set_user_agent_suffix("codex_cloud_tasks_tui");
-
-    // Default to online unless explicitly configured to use mock.
-    let use_mock = matches!(
-        std::env::var("CODEX_CLOUD_TASKS_MODE").ok().as_deref(),
-        Some("mock") | Some("MOCK")
-    );
-
-    let backend: Arc<dyn codex_cloud_tasks_client::CloudBackend> = if use_mock {
-        Arc::new(codex_cloud_tasks_client::MockClient)
-    } else {
-        // Build an HTTP client against the configured (or default) base URL.
-        let base_url = std::env::var("CODEX_CLOUD_TASKS_BASE_URL")
-            .unwrap_or_else(|_| "https://chatgpt.com/backend-api".to_string());
-        let ua = codex_core::default_client::get_codex_user_agent();
-        let mut http =
-            codex_cloud_tasks_client::HttpClient::new(base_url.clone())?.with_user_agent(ua);
-        // Log which base URL and path style we're going to use.
-        let style = if base_url.contains("/backend-api") {
-            "wham"
-        } else {
-            "codex-api"
-        };
-        append_error_log(format!("startup: base_url={base_url} path_style={style}"));
-
-        // Require ChatGPT login (SWIC). Exit with a clear message if missing.
-        let _token = match codex_core::config::find_codex_home()
-            .ok()
-            .map(codex_login::AuthManager::new)
-            .and_then(|am| am.auth())
-        {
-            Some(auth) => {
-                // Log account context for debugging workspace selection.
-                if let Some(acc) = auth.get_account_id() {
-                    append_error_log(format!("auth: mode=ChatGPT account_id={acc}"));
-                }
-                match auth.get_token().await {
-                    Ok(t) if !t.is_empty() => {
-                        // Attach token and ChatGPT-Account-Id header if available
-                        http = http.with_bearer_token(t.clone());
-                        if let Some(acc) = auth
-                            .get_account_id()
-                            .or_else(|| util::extract_chatgpt_account_id(&t))
-                        {
-                            append_error_log(format!("auth: set ChatGPT-Account-Id header: {acc}"));
-                            http = http.with_chatgpt_account_id(acc);
-                        }
-                        t
-                    }
-                    _ => {
-                        eprintln!(
-                            "Not signed in. Please run 'codex login' to sign in with ChatGPT, then re-run 'codex cloud'."
-                        );
-                        std::process::exit(1);
-                    }
-                }
-            }
-            None => {
-                eprintln!(
-                    "Not signed in. Please run 'codex login' to sign in with ChatGPT, then re-run 'codex cloud'."
-                );
-                std::process::exit(1);
-            }
-        };
-        Arc::new(http)
-    };
+    let BackendContext { backend, .. } = init_backend("codex_cloud_tasks_tui").await?;
+    let backend = backend;
 
     // Terminal setup
     use crossterm::ExecutableCommand;
@@ -400,16 +514,20 @@ pub async fn run_main(_cli: Cli, _codex_linux_sandbox_exe: Option<PathBuf>) -> a
                         let _ = frame_tx.send(Instant::now() + codex_tui::ComposerInput::recommended_flush_delay());
                     }
                 }
-                // Advance throbber only while loading.
+                // Keep spinner pulsing only while loading.
                 if app.refresh_inflight
                     || app.details_inflight
                     || app.env_loading
                     || app.apply_preflight_inflight
                     || app.apply_inflight
                 {
-                    app.throbber.calc_next();
+                    if app.spinner_start.is_none() {
+                        app.spinner_start = Some(Instant::now());
+                    }
                     needs_redraw = true;
-                    let _ = frame_tx.send(Instant::now() + Duration::from_millis(100));
+                    let _ = frame_tx.send(Instant::now() + Duration::from_millis(600));
+                } else {
+                    app.spinner_start = None;
                 }
                 render_if_needed(&mut terminal, &mut app, &mut needs_redraw)?;
             }

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

@@ -16,6 +16,7 @@ use ratatui::widgets::ListState;
 use ratatui::widgets::Padding;
 use ratatui::widgets::Paragraph;
 use std::sync::OnceLock;
+use std::time::Instant;
 
 use crate::app::App;
 use crate::app::AttemptView;
@@ -229,7 +230,7 @@ fn draw_list(frame: &mut Frame, area: Rect, app: &mut App) {
 
     // In-box spinner during initial/refresh loads
     if app.refresh_inflight {
-        draw_centered_spinner(frame, inner, &mut app.throbber, "Loading tasks…");
+        draw_centered_spinner(frame, inner, &mut app.spinner_start, "Loading tasks…");
     }
 }
 
@@ -291,7 +292,7 @@ fn draw_footer(frame: &mut Frame, area: Rect, app: &mut App) {
         || app.apply_preflight_inflight
         || app.apply_inflight
     {
-        draw_inline_spinner(frame, top[1], &mut app.throbber, "Loading…");
+        draw_inline_spinner(frame, top[1], &mut app.spinner_start, "Loading…");
     } else {
         frame.render_widget(Clear, top[1]);
     }
@@ -449,7 +450,12 @@ fn draw_diff_overlay(frame: &mut Frame, area: Rect, app: &mut App) {
         .map(|o| o.sd.wrapped_lines().is_empty())
         .unwrap_or(true);
     if app.details_inflight && raw_empty {
-        draw_centered_spinner(frame, content_area, &mut app.throbber, "Loading details…");
+        draw_centered_spinner(
+            frame,
+            content_area,
+            &mut app.spinner_start,
+            "Loading details…",
+        );
     } else {
         let scroll = app
             .diff_overlay
@@ -494,11 +500,11 @@ pub fn draw_apply_modal(frame: &mut Frame, area: Rect, app: &mut App) {
         frame.render_widget(header, rows[0]);
         // Body: spinner while preflight/apply runs; otherwise show result message and path lists
         if app.apply_preflight_inflight {
-            draw_centered_spinner(frame, rows[1], &mut app.throbber, "Checking…");
+            draw_centered_spinner(frame, rows[1], &mut app.spinner_start, "Checking…");
         } else if app.apply_inflight {
-            draw_centered_spinner(frame, rows[1], &mut app.throbber, "Applying…");
+            draw_centered_spinner(frame, rows[1], &mut app.spinner_start, "Applying…");
         } else if m.result_message.is_none() {
-            draw_centered_spinner(frame, rows[1], &mut app.throbber, "Loading…");
+            draw_centered_spinner(frame, rows[1], &mut app.spinner_start, "Loading…");
         } else if let Some(msg) = &m.result_message {
             let mut body_lines: Vec<Line> = Vec::new();
             let first = match m.result_level {
@@ -859,29 +865,29 @@ fn format_relative_time(ts: chrono::DateTime<Utc>) -> String {
 fn draw_inline_spinner(
     frame: &mut Frame,
     area: Rect,
-    state: &mut throbber_widgets_tui::ThrobberState,
+    spinner_start: &mut Option<Instant>,
     label: &str,
 ) {
-    use ratatui::style::Style;
-    use throbber_widgets_tui::BRAILLE_EIGHT;
-    use throbber_widgets_tui::Throbber;
-    use throbber_widgets_tui::WhichUse;
-    let w = Throbber::default()
-        .label(label)
-        .style(Style::default().cyan())
-        .throbber_style(Style::default().magenta().bold())
-        .throbber_set(BRAILLE_EIGHT)
-        .use_type(WhichUse::Spin);
-    frame.render_stateful_widget(w, area, state);
+    use ratatui::widgets::Paragraph;
+    let start = spinner_start.get_or_insert_with(Instant::now);
+    let blink_on = (start.elapsed().as_millis() / 600).is_multiple_of(2);
+    let dot = if blink_on {
+        "• ".into()
+    } else {
+        "◦ ".dim()
+    };
+    let label = label.cyan();
+    let line = Line::from(vec![dot, label]);
+    frame.render_widget(Paragraph::new(line), area);
 }
 
 fn draw_centered_spinner(
     frame: &mut Frame,
     area: Rect,
-    state: &mut throbber_widgets_tui::ThrobberState,
+    spinner_start: &mut Option<Instant>,
     label: &str,
 ) {
-    // Center a 1xN throbber within the given rect
+    // Center a 1xN spinner within the given rect
     let rows = Layout::default()
         .direction(Direction::Vertical)
         .constraints([
@@ -898,7 +904,7 @@ fn draw_centered_spinner(
             Constraint::Percentage(50),
         ])
         .split(rows[1]);
-    draw_inline_spinner(frame, cols[1], state, label);
+    draw_inline_spinner(frame, cols[1], spinner_start, label);
 }
 
 // Styling helpers for diff rendering live inline where used.
@@ -918,7 +924,12 @@ pub fn draw_env_modal(frame: &mut Frame, area: Rect, app: &mut App) {
     let content = overlay_content(inner);
 
     if app.env_loading {
-        draw_centered_spinner(frame, content, &mut app.throbber, "Loading environments…");
+        draw_centered_spinner(
+            frame,
+            content,
+            &mut app.spinner_start,
+            "Loading environments…",
+        );
         return;
     }
 

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

@@ -70,7 +70,7 @@ pub async fn build_chatgpt_headers() -> HeaderMap {
         HeaderValue::from_str(&ua).unwrap_or(HeaderValue::from_static("codex-cli")),
     );
     if let Ok(home) = codex_core::config::find_codex_home() {
-        let am = codex_login::AuthManager::new(home);
+        let am = codex_login::AuthManager::new(home, false);
         if let Some(auth) = am.auth()
             && let Ok(tok) = auth.get_token().await
             && !tok.is_empty()
@@ -91,3 +91,18 @@ pub async fn build_chatgpt_headers() -> HeaderMap {
     }
     headers
 }
+
+/// Construct a browser-friendly task URL for the given backend base URL.
+pub fn task_url(base_url: &str, task_id: &str) -> String {
+    let normalized = normalize_base_url(base_url);
+    if let Some(root) = normalized.strip_suffix("/backend-api") {
+        return format!("{root}/codex/tasks/{task_id}");
+    }
+    if let Some(root) = normalized.strip_suffix("/api/codex") {
+        return format!("{root}/codex/tasks/{task_id}");
+    }
+    if normalized.ends_with("/codex") {
+        return format!("{normalized}/tasks/{task_id}");
+    }
+    format!("{normalized}/codex/tasks/{task_id}")
+}

codex-rs/common/src/model_presets.rs

@@ -20,49 +20,49 @@ const PRESETS: &[ModelPreset] = &[
     ModelPreset {
         id: "gpt-5-codex-low",
         label: "gpt-5-codex low",
-        description: "",
+        description: "Fastest responses with limited reasoning",
         model: "gpt-5-codex",
         effort: Some(ReasoningEffort::Low),
     },
     ModelPreset {
         id: "gpt-5-codex-medium",
         label: "gpt-5-codex medium",
-        description: "",
+        description: "Dynamically adjusts reasoning based on the task",
         model: "gpt-5-codex",
         effort: Some(ReasoningEffort::Medium),
     },
     ModelPreset {
         id: "gpt-5-codex-high",
         label: "gpt-5-codex high",
-        description: "",
+        description: "Maximizes reasoning depth for complex or ambiguous problems",
         model: "gpt-5-codex",
         effort: Some(ReasoningEffort::High),
     },
     ModelPreset {
         id: "gpt-5-minimal",
         label: "gpt-5 minimal",
-        description: "— fastest responses with limited reasoning; ideal for coding, instructions, or lightweight tasks",
+        description: "Fastest responses with little reasoning",
         model: "gpt-5",
         effort: Some(ReasoningEffort::Minimal),
     },
     ModelPreset {
         id: "gpt-5-low",
         label: "gpt-5 low",
-        description: "— balances speed with some reasoning; useful for straightforward queries and short explanations",
+        description: "Balances speed with some reasoning; useful for straightforward queries and short explanations",
         model: "gpt-5",
         effort: Some(ReasoningEffort::Low),
     },
     ModelPreset {
         id: "gpt-5-medium",
         label: "gpt-5 medium",
-        description: "— default setting; provides a solid balance of reasoning depth and latency for general-purpose tasks",
+        description: "Provides a solid balance of reasoning depth and latency for general-purpose tasks",
         model: "gpt-5",
         effort: Some(ReasoningEffort::Medium),
     },
     ModelPreset {
         id: "gpt-5-high",
         label: "gpt-5 high",
-        description: "— maximizes reasoning depth for complex or ambiguous problems",
+        description: "Maximizes reasoning depth for complex or ambiguous problems",
         model: "gpt-5",
         effort: Some(ReasoningEffort::High),
     },

codex-rs/core/Cargo.toml

@@ -19,13 +19,14 @@ async-trait = { workspace = true }
 base64 = { workspace = true }
 bytes = { workspace = true }
 chrono = { workspace = true, features = ["serde"] }
+codex-app-server-protocol = { workspace = true }
 codex-apply-patch = { workspace = true }
 codex-file-search = { workspace = true }
 codex-mcp-client = { workspace = true }
-codex-rmcp-client = { workspace = true }
-codex-protocol = { workspace = true }
-codex-app-server-protocol = { workspace = true }
 codex-otel = { workspace = true, features = ["otel"] }
+codex-protocol = { workspace = true }
+codex-rmcp-client = { workspace = true }
+codex-utils-string = { workspace = true }
 dirs = { workspace = true }
 dunce = { workspace = true }
 env-flags = { workspace = true }
@@ -60,7 +61,7 @@ tokio = { workspace = true, features = [
     "rt-multi-thread",
     "signal",
 ] }
-tokio-util = { workspace = true }
+tokio-util = { workspace = true, features = ["rt"] }
 toml = { workspace = true }
 toml_edit = { workspace = true }
 tracing = { workspace = true, features = ["log"] }
@@ -75,6 +76,9 @@ wildmatch = { workspace = true }
 landlock = { workspace = true }
 seccompiler = { workspace = true }
 
+[target.'cfg(target_os = "macos")'.dependencies]
+core-foundation = "0.9"
+
 # Build OpenSSL from source for musl builds.
 [target.x86_64-unknown-linux-musl.dependencies]
 openssl-sys = { workspace = true, features = ["vendored"] }
@@ -85,16 +89,18 @@ openssl-sys = { workspace = true, features = ["vendored"] }
 
 [dev-dependencies]
 assert_cmd = { workspace = true }
+assert_matches = { workspace = true }
 core_test_support = { workspace = true }
 escargot = { workspace = true }
 maplit = { workspace = true }
 predicates = { workspace = true }
 pretty_assertions = { workspace = true }
+serial_test = { workspace = true }
 tempfile = { workspace = true }
 tokio-test = { workspace = true }
+tracing-test = { workspace = true, features = ["no-env-filter"] }
 walkdir = { workspace = true }
 wiremock = { workspace = true }
-tracing-test = {  workspace = true, features = ["no-env-filter"] }
 
 [package.metadata.cargo-shear]
 ignored = ["openssl-sys"]

codex-rs/core/README.md

@@ -12,7 +12,7 @@ Expects `/usr/bin/sandbox-exec` to be present.
 
 ### Linux
 
-Expects the binary containing `codex-core` to run the equivalent of `codex debug landlock` when `arg0` is `codex-linux-sandbox`. See the `codex-arg0` crate for details.
+Expects the binary containing `codex-core` to run the equivalent of `codex sandbox linux` (legacy alias: `codex debug landlock`) when `arg0` is `codex-linux-sandbox`. See the `codex-arg0` crate for details.
 
 ### All Platforms
 

codex-rs/core/gpt_5_codex_prompt.md

@@ -10,12 +10,14 @@ You are Codex, based on GPT-5. You are running as a coding agent in the Codex CL
 
 - 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.
 - 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
 
@@ -89,7 +91,7 @@ You are producing plain text that will later be styled by the CLI. Follow these
 - 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; add a language hint whenever obvious.
+- 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.

codex-rs/core/src/auth.rs

@@ -73,7 +73,7 @@ impl CodexAuth {
 
     /// Loads the available auth information from the auth.json.
     pub fn from_codex_home(codex_home: &Path) -> std::io::Result<Option<CodexAuth>> {
-        load_auth(codex_home)
+        load_auth(codex_home, false)
     }
 
     pub async fn get_token_data(&self) -> Result<TokenData, std::io::Error> {
@@ -188,6 +188,7 @@ impl CodexAuth {
 }
 
 pub const OPENAI_API_KEY_ENV_VAR: &str = "OPENAI_API_KEY";
+pub const CODEX_API_KEY_ENV_VAR: &str = "CODEX_API_KEY";
 
 pub fn read_openai_api_key_from_env() -> Option<String> {
     env::var(OPENAI_API_KEY_ENV_VAR)
@@ -196,6 +197,13 @@ pub fn read_openai_api_key_from_env() -> Option<String> {
         .filter(|value| !value.is_empty())
 }
 
+pub fn read_codex_api_key_from_env() -> Option<String> {
+    env::var(CODEX_API_KEY_ENV_VAR)
+        .ok()
+        .map(|value| value.trim().to_string())
+        .filter(|value| !value.is_empty())
+}
+
 pub fn get_auth_file(codex_home: &Path) -> PathBuf {
     codex_home.join("auth.json")
 }
@@ -221,7 +229,18 @@ pub fn login_with_api_key(codex_home: &Path, api_key: &str) -> std::io::Result<(
     write_auth_json(&get_auth_file(codex_home), &auth_dot_json)
 }
 
-fn load_auth(codex_home: &Path) -> std::io::Result<Option<CodexAuth>> {
+fn load_auth(
+    codex_home: &Path,
+    enable_codex_api_key_env: bool,
+) -> std::io::Result<Option<CodexAuth>> {
+    if enable_codex_api_key_env && let Some(api_key) = read_codex_api_key_from_env() {
+        let client = crate::default_client::create_client();
+        return Ok(Some(CodexAuth::from_api_key_with_client(
+            api_key.as_str(),
+            client,
+        )));
+    }
+
     let auth_file = get_auth_file(codex_home);
     let client = crate::default_client::create_client();
     let auth_dot_json = match try_read_auth_json(&auth_file) {
@@ -455,7 +474,7 @@ mod tests {
             auth_dot_json,
             auth_file: _,
             ..
-        } = super::load_auth(codex_home.path()).unwrap().unwrap();
+        } = super::load_auth(codex_home.path(), false).unwrap().unwrap();
         assert_eq!(None, api_key);
         assert_eq!(AuthMode::ChatGPT, mode);
 
@@ -494,7 +513,7 @@ mod tests {
         )
         .unwrap();
 
-        let auth = super::load_auth(dir.path()).unwrap().unwrap();
+        let auth = super::load_auth(dir.path(), false).unwrap().unwrap();
         assert_eq!(auth.mode, AuthMode::ApiKey);
         assert_eq!(auth.api_key, Some("sk-test-key".to_string()));
 
@@ -577,6 +596,7 @@ mod tests {
 pub struct AuthManager {
     codex_home: PathBuf,
     inner: RwLock<CachedAuth>,
+    enable_codex_api_key_env: bool,
 }
 
 impl AuthManager {
@@ -584,11 +604,14 @@ impl AuthManager {
     /// preferred auth method. Errors loading auth are swallowed; `auth()` will
     /// simply return `None` in that case so callers can treat it as an
     /// unauthenticated state.
-    pub fn new(codex_home: PathBuf) -> Self {
-        let auth = CodexAuth::from_codex_home(&codex_home).ok().flatten();
+    pub fn new(codex_home: PathBuf, enable_codex_api_key_env: bool) -> Self {
+        let auth = load_auth(&codex_home, enable_codex_api_key_env)
+            .ok()
+            .flatten();
         Self {
             codex_home,
             inner: RwLock::new(CachedAuth { auth }),
+            enable_codex_api_key_env,
         }
     }
 
@@ -598,6 +621,7 @@ impl AuthManager {
         Arc::new(Self {
             codex_home: PathBuf::new(),
             inner: RwLock::new(cached),
+            enable_codex_api_key_env: false,
         })
     }
 
@@ -609,7 +633,9 @@ impl AuthManager {
     /// Force a reload of the auth information from auth.json. Returns
     /// whether the auth value changed.
     pub fn reload(&self) -> bool {
-        let new_auth = CodexAuth::from_codex_home(&self.codex_home).ok().flatten();
+        let new_auth = load_auth(&self.codex_home, self.enable_codex_api_key_env)
+            .ok()
+            .flatten();
         if let Ok(mut guard) = self.inner.write() {
             let changed = !AuthManager::auths_equal(&guard.auth, &new_auth);
             guard.auth = new_auth;
@@ -628,8 +654,8 @@ impl AuthManager {
     }
 
     /// Convenience constructor returning an `Arc` wrapper.
-    pub fn shared(codex_home: PathBuf) -> Arc<Self> {
-        Arc::new(Self::new(codex_home))
+    pub fn shared(codex_home: PathBuf, enable_codex_api_key_env: bool) -> Arc<Self> {
+        Arc::new(Self::new(codex_home, enable_codex_api_key_env))
     }
 
     /// Attempt to refresh the current auth token (if any). On success, reload

codex-rs/core/src/chat_completions.rs

@@ -389,10 +389,12 @@ async fn process_chat_sse<S>(
     let mut reasoning_text = String::new();
 
     loop {
-        let sse = match otel_event_manager
-            .log_sse_event(|| timeout(idle_timeout, stream.next()))
-            .await
-        {
+        let start = std::time::Instant::now();
+        let response = timeout(idle_timeout, stream.next()).await;
+        let duration = start.elapsed();
+        otel_event_manager.log_sse_event(&response, duration);
+
+        let sse = match response {
             Ok(Some(Ok(ev))) => ev,
             Ok(Some(Err(e))) => {
                 let _ = tx_event

codex-rs/core/src/client.rs

@@ -43,16 +43,19 @@ 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::openai_tools::create_allowed_tools_json_for_responses_api;
 use crate::openai_tools::create_tools_json_for_responses_api;
 use crate::protocol::RateLimitSnapshot;
 use crate::protocol::RateLimitWindow;
 use crate::protocol::TokenUsage;
+use crate::state::TaskKind;
 use crate::token_data::PlanType;
 use crate::util::backoff;
 use codex_otel::otel_event_manager::OtelEventManager;
 use codex_protocol::config_types::ReasoningEffort as ReasoningEffortConfig;
 use codex_protocol::config_types::ReasoningSummary as ReasoningSummaryConfig;
 use codex_protocol::models::ResponseItem;
+use std::collections::HashSet;
 use std::sync::Arc;
 
 #[derive(Debug, Deserialize)]
@@ -63,7 +66,6 @@ struct ErrorResponse {
 #[derive(Debug, Deserialize)]
 struct Error {
     r#type: Option<String>,
-    #[allow(dead_code)]
     code: Option<String>,
     message: Option<String>,
 
@@ -124,8 +126,16 @@ impl ModelClient {
     /// the provider config.  Public callers always invoke `stream()` – the
     /// specialised helpers are private to avoid accidental misuse.
     pub async fn stream(&self, prompt: &Prompt) -> Result<ResponseStream> {
+        self.stream_with_task_kind(prompt, TaskKind::Regular).await
+    }
+
+    pub(crate) async fn stream_with_task_kind(
+        &self,
+        prompt: &Prompt,
+        task_kind: TaskKind,
+    ) -> Result<ResponseStream> {
         match self.provider.wire_api {
-            WireApi::Responses => self.stream_responses(prompt).await,
+            WireApi::Responses => self.stream_responses(prompt, task_kind).await,
             WireApi::Chat => {
                 // Create the raw streaming connection first.
                 let response_stream = stream_chat_completions(
@@ -166,7 +176,11 @@ impl ModelClient {
     }
 
     /// Implementation for the OpenAI *Responses* experimental API.
-    async fn stream_responses(&self, prompt: &Prompt) -> Result<ResponseStream> {
+    async fn stream_responses(
+        &self,
+        prompt: &Prompt,
+        task_kind: TaskKind,
+    ) -> Result<ResponseStream> {
         if let Some(path) = &*CODEX_RS_SSE_FIXTURE {
             // short circuit for tests
             warn!(path, "Streaming from fixture");
@@ -221,14 +235,40 @@ impl ModelClient {
         //
         // For Azure, we send `store: true` and preserve reasoning item IDs.
         let azure_workaround = self.provider.is_azure_responses_endpoint();
+        let tool_choice = if let Some(disabled) = &prompt.disabled_tools {
+            if disabled.is_empty() {
+                serde_json::json!("auto")
+            } else {
+                let allowed = create_allowed_tools_json_for_responses_api(&prompt.tools, disabled);
+                let total_unique = prompt
+                    .tools
+                    .iter()
+                    .map(super::client_common::tools::ToolSpec::name)
+                    .collect::<HashSet<_>>()
+                    .len();
+                if allowed.is_empty() {
+                    serde_json::json!("none")
+                } else if allowed.len() == total_unique {
+                    serde_json::json!("auto")
+                } else {
+                    serde_json::json!({
+                        "type": "allowed_tools",
+                        "mode": "auto",
+                        "tools": allowed,
+                    })
+                }
+            }
+        } else {
+            serde_json::json!("auto")
+        };
 
         let payload = ResponsesApiRequest {
             model: &self.config.model,
             instructions: &full_instructions,
             input: &input_with_instructions,
             tools: &tools_json,
-            tool_choice: "auto",
-            parallel_tool_calls: false,
+            tool_choice,
+            parallel_tool_calls: prompt.parallel_tool_calls,
             reasoning,
             store: azure_workaround,
             stream: true,
@@ -245,7 +285,7 @@ impl ModelClient {
         let max_attempts = self.provider.request_max_retries();
         for attempt in 0..=max_attempts {
             match self
-                .attempt_stream_responses(attempt, &payload_json, &auth_manager)
+                .attempt_stream_responses(attempt, &payload_json, &auth_manager, task_kind)
                 .await
             {
                 Ok(stream) => {
@@ -273,6 +313,7 @@ impl ModelClient {
         attempt: u64,
         payload_json: &Value,
         auth_manager: &Option<Arc<AuthManager>>,
+        task_kind: TaskKind,
     ) -> std::result::Result<ResponseStream, StreamAttemptError> {
         // Always fetch the latest auth in case a prior attempt refreshed the token.
         let auth = auth_manager.as_ref().and_then(|m| m.auth());
@@ -295,6 +336,7 @@ impl ModelClient {
             .header("conversation_id", self.conversation_id.to_string())
             .header("session_id", self.conversation_id.to_string())
             .header(reqwest::header::ACCEPT, "text/event-stream")
+            .header("Codex-Task-Type", task_kind.header_value())
             .json(payload_json);
 
         if let Some(auth) = auth.as_ref()
@@ -650,10 +692,12 @@ async fn process_sse<S>(
     let mut response_error: Option<CodexErr> = None;
 
     loop {
-        let sse = match otel_event_manager
-            .log_sse_event(|| timeout(idle_timeout, stream.next()))
-            .await
-        {
+        let start = std::time::Instant::now();
+        let response = timeout(idle_timeout, stream.next()).await;
+        let duration = start.elapsed();
+        otel_event_manager.log_sse_event(&response, duration);
+
+        let sse = match response {
             Ok(Some(Ok(sse))) => sse,
             Ok(Some(Err(e))) => {
                 debug!("SSE Error: {e:#}");
@@ -794,9 +838,13 @@ async fn process_sse<S>(
                     if let Some(error) = error {
                         match serde_json::from_value::<Error>(error.clone()) {
                             Ok(error) => {
-                                let delay = try_parse_retry_after(&error);
-                                let message = error.message.unwrap_or_default();
-                                response_error = Some(CodexErr::Stream(message, delay));
+                                if is_context_window_error(&error) {
+                                    response_error = Some(CodexErr::ContextWindowExceeded);
+                                } else {
+                                    let delay = try_parse_retry_after(&error);
+                                    let message = error.message.clone().unwrap_or_default();
+                                    response_error = Some(CodexErr::Stream(message, delay));
+                                }
                             }
                             Err(e) => {
                                 let error = format!("failed to parse ErrorResponse: {e}");
@@ -922,9 +970,14 @@ fn try_parse_retry_after(err: &Error) -> Option<Duration> {
     None
 }
 
+fn is_context_window_error(error: &Error) -> bool {
+    error.code.as_deref() == Some("context_length_exceeded")
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
+    use assert_matches::assert_matches;
     use serde_json::json;
     use tokio::sync::mpsc;
     use tokio_test::io::Builder as IoBuilder;
@@ -1179,6 +1232,74 @@ mod tests {
         }
     }
 
+    #[tokio::test]
+    async fn context_window_error_is_fatal() {
+        let raw_error = r#"{"type":"response.failed","sequence_number":3,"response":{"id":"resp_5c66275b97b9baef1ed95550adb3b7ec13b17aafd1d2f11b","object":"response","created_at":1759510079,"status":"failed","background":false,"error":{"code":"context_length_exceeded","message":"Your input exceeds the context window of this model. Please adjust your input and try again."},"usage":null,"user":null,"metadata":{}}}"#;
+
+        let sse1 = format!("event: response.failed\ndata: {raw_error}\n\n");
+        let provider = ModelProviderInfo {
+            name: "test".to_string(),
+            base_url: Some("https://test.com".to_string()),
+            env_key: Some("TEST_API_KEY".to_string()),
+            env_key_instructions: None,
+            wire_api: WireApi::Responses,
+            query_params: None,
+            http_headers: None,
+            env_http_headers: None,
+            request_max_retries: Some(0),
+            stream_max_retries: Some(0),
+            stream_idle_timeout_ms: Some(1000),
+            requires_openai_auth: false,
+        };
+
+        let otel_event_manager = otel_event_manager();
+
+        let events = collect_events(&[sse1.as_bytes()], provider, otel_event_manager).await;
+
+        assert_eq!(events.len(), 1);
+
+        match &events[0] {
+            Err(err @ CodexErr::ContextWindowExceeded) => {
+                assert_eq!(err.to_string(), CodexErr::ContextWindowExceeded.to_string());
+            }
+            other => panic!("unexpected context window event: {other:?}"),
+        }
+    }
+
+    #[tokio::test]
+    async fn context_window_error_with_newline_is_fatal() {
+        let raw_error = r#"{"type":"response.failed","sequence_number":4,"response":{"id":"resp_fatal_newline","object":"response","created_at":1759510080,"status":"failed","background":false,"error":{"code":"context_length_exceeded","message":"Your input exceeds the context window of this model. Please adjust your input and try\nagain."},"usage":null,"user":null,"metadata":{}}}"#;
+
+        let sse1 = format!("event: response.failed\ndata: {raw_error}\n\n");
+        let provider = ModelProviderInfo {
+            name: "test".to_string(),
+            base_url: Some("https://test.com".to_string()),
+            env_key: Some("TEST_API_KEY".to_string()),
+            env_key_instructions: None,
+            wire_api: WireApi::Responses,
+            query_params: None,
+            http_headers: None,
+            env_http_headers: None,
+            request_max_retries: Some(0),
+            stream_max_retries: Some(0),
+            stream_idle_timeout_ms: Some(1000),
+            requires_openai_auth: false,
+        };
+
+        let otel_event_manager = otel_event_manager();
+
+        let events = collect_events(&[sse1.as_bytes()], provider, otel_event_manager).await;
+
+        assert_eq!(events.len(), 1);
+
+        match &events[0] {
+            Err(err @ CodexErr::ContextWindowExceeded) => {
+                assert_eq!(err.to_string(), CodexErr::ContextWindowExceeded.to_string());
+            }
+            other => panic!("unexpected context window event: {other:?}"),
+        }
+    }
+
     // ────────────────────────────
     // Table-driven test from `main`
     // ────────────────────────────
@@ -1316,10 +1437,7 @@ mod tests {
         let resp: ErrorResponse =
             serde_json::from_str(json).expect("should deserialize old schema");
 
-        assert!(matches!(
-            resp.error.plan_type,
-            Some(PlanType::Known(KnownPlan::Pro))
-        ));
+        assert_matches!(resp.error.plan_type, Some(PlanType::Known(KnownPlan::Pro)));
 
         let plan_json = serde_json::to_string(&resp.error.plan_type).expect("serialize plan_type");
         assert_eq!(plan_json, "\"pro\"");
@@ -1334,7 +1452,7 @@ mod tests {
         let resp: ErrorResponse =
             serde_json::from_str(json).expect("should deserialize old schema");
 
-        assert!(matches!(resp.error.plan_type, Some(PlanType::Unknown(ref s)) if s == "vip"));
+        assert_matches!(resp.error.plan_type, Some(PlanType::Unknown(ref s)) if s == "vip");
 
         let plan_json = serde_json::to_string(&resp.error.plan_type).expect("serialize plan_type");
         assert_eq!(plan_json, "\"vip\"");

codex-rs/core/src/client_common.rs

@@ -1,6 +1,6 @@
+use crate::client_common::tools::ToolSpec;
 use crate::error::Result;
 use crate::model_family::ModelFamily;
-use crate::openai_tools::OpenAiTool;
 use crate::protocol::RateLimitSnapshot;
 use crate::protocol::TokenUsage;
 use codex_apply_patch::APPLY_PATCH_TOOL_INSTRUCTIONS;
@@ -9,9 +9,11 @@ use codex_protocol::config_types::ReasoningSummary as ReasoningSummaryConfig;
 use codex_protocol::config_types::Verbosity as VerbosityConfig;
 use codex_protocol::models::ResponseItem;
 use futures::Stream;
+use serde::Deserialize;
 use serde::Serialize;
 use serde_json::Value;
 use std::borrow::Cow;
+use std::collections::HashSet;
 use std::ops::Deref;
 use std::pin::Pin;
 use std::task::Context;
@@ -29,13 +31,19 @@ pub struct Prompt {
 
     /// Tools available to the model, including additional tools sourced from
     /// external MCP servers.
-    pub(crate) tools: Vec<OpenAiTool>,
+    pub(crate) tools: Vec<ToolSpec>,
+
+    /// Whether parallel tool calls are permitted for this prompt.
+    pub(crate) parallel_tool_calls: bool,
 
     /// Optional override for the built-in BASE_INSTRUCTIONS.
     pub base_instructions_override: Option<String>,
 
     /// Optional the output schema for the model's response.
     pub output_schema: Option<Value>,
+
+    /// Optional list of tool names to disable for this prompt.
+    pub disabled_tools: Option<Vec<String>>,
 }
 
 impl Prompt {
@@ -49,8 +57,8 @@ impl Prompt {
         // AND
         // - there is no apply_patch tool present
         let is_apply_patch_tool_present = self.tools.iter().any(|tool| match tool {
-            OpenAiTool::Function(f) => f.name == "apply_patch",
-            OpenAiTool::Freeform(f) => f.name == "apply_patch",
+            ToolSpec::Function(f) => f.name == "apply_patch",
+            ToolSpec::Freeform(f) => f.name == "apply_patch",
             _ => false,
         });
         if self.base_instructions_override.is_none()
@@ -64,10 +72,125 @@ impl Prompt {
     }
 
     pub(crate) fn get_formatted_input(&self) -> Vec<ResponseItem> {
-        self.input.clone()
+        let mut input = self.input.clone();
+
+        // when using the *Freeform* apply_patch tool specifically, tool outputs
+        // should be structured text, not json. Do NOT reserialize when using
+        // the Function tool - note that this differs from the check above for
+        // instructions. We declare the result as a named variable for clarity.
+        let is_freeform_apply_patch_tool_present = self.tools.iter().any(|tool| match tool {
+            ToolSpec::Freeform(f) => f.name == "apply_patch",
+            _ => false,
+        });
+        if is_freeform_apply_patch_tool_present {
+            reserialize_shell_outputs(&mut input);
+        }
+
+        input
     }
 }
 
+fn reserialize_shell_outputs(items: &mut [ResponseItem]) {
+    let mut shell_call_ids: HashSet<String> = HashSet::new();
+
+    items.iter_mut().for_each(|item| match item {
+        ResponseItem::LocalShellCall { call_id, id, .. } => {
+            if let Some(identifier) = call_id.clone().or_else(|| id.clone()) {
+                shell_call_ids.insert(identifier);
+            }
+        }
+        ResponseItem::CustomToolCall {
+            id: _,
+            status: _,
+            call_id,
+            name,
+            input: _,
+        } => {
+            if name == "apply_patch" {
+                shell_call_ids.insert(call_id.clone());
+            }
+        }
+        ResponseItem::CustomToolCallOutput { call_id, output } => {
+            if shell_call_ids.remove(call_id)
+                && let Some(structured) = parse_structured_shell_output(output)
+            {
+                *output = structured
+            }
+        }
+        ResponseItem::FunctionCall { name, call_id, .. }
+            if is_shell_tool_name(name) || name == "apply_patch" =>
+        {
+            shell_call_ids.insert(call_id.clone());
+        }
+        ResponseItem::FunctionCallOutput { call_id, output } => {
+            if shell_call_ids.remove(call_id)
+                && let Some(structured) = parse_structured_shell_output(&output.content)
+            {
+                output.content = structured
+            }
+        }
+        _ => {}
+    })
+}
+
+fn is_shell_tool_name(name: &str) -> bool {
+    matches!(name, "shell" | "container.exec")
+}
+
+#[derive(Deserialize)]
+struct ExecOutputJson {
+    output: String,
+    metadata: ExecOutputMetadataJson,
+}
+
+#[derive(Deserialize)]
+struct ExecOutputMetadataJson {
+    exit_code: i32,
+    duration_seconds: f32,
+}
+
+fn parse_structured_shell_output(raw: &str) -> Option<String> {
+    let parsed: ExecOutputJson = serde_json::from_str(raw).ok()?;
+    Some(build_structured_output(&parsed))
+}
+
+fn build_structured_output(parsed: &ExecOutputJson) -> String {
+    let mut sections = Vec::new();
+    sections.push(format!("Exit code: {}", parsed.metadata.exit_code));
+    sections.push(format!(
+        "Wall time: {} seconds",
+        parsed.metadata.duration_seconds
+    ));
+
+    let mut output = parsed.output.clone();
+    if let Some(total_lines) = extract_total_output_lines(&parsed.output) {
+        sections.push(format!("Total output lines: {total_lines}"));
+        if let Some(stripped) = strip_total_output_header(&output) {
+            output = stripped.to_string();
+        }
+    }
+
+    sections.push("Output:".to_string());
+    sections.push(output);
+
+    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> {
+    let after_prefix = output.strip_prefix("Total output lines: ")?;
+    let (_, remainder) = after_prefix.split_once('\n')?;
+    let remainder = remainder.strip_prefix('\n').unwrap_or(remainder);
+    Some(remainder)
+}
+
 #[derive(Debug)]
 pub enum ResponseEvent {
     Created,
@@ -148,7 +271,7 @@ pub(crate) struct ResponsesApiRequest<'a> {
     // separate enum for serialization.
     pub(crate) input: &'a Vec<ResponseItem>,
     pub(crate) tools: &'a [serde_json::Value],
-    pub(crate) tool_choice: &'static str,
+    pub(crate) tool_choice: Value,
     pub(crate) parallel_tool_calls: bool,
     pub(crate) reasoning: Option<Reasoning>,
     pub(crate) store: bool,
@@ -160,6 +283,65 @@ pub(crate) struct ResponsesApiRequest<'a> {
     pub(crate) text: Option<TextControls>,
 }
 
+pub(crate) mod tools {
+    use crate::openai_tools::JsonSchema;
+    use serde::Deserialize;
+    use serde::Serialize;
+
+    /// When serialized as JSON, this produces a valid "Tool" in the OpenAI
+    /// Responses API.
+    #[derive(Debug, Clone, Serialize, PartialEq)]
+    #[serde(tag = "type")]
+    pub(crate) enum ToolSpec {
+        #[serde(rename = "function")]
+        Function(ResponsesApiTool),
+        #[serde(rename = "local_shell")]
+        LocalShell {},
+        // TODO: Understand why we get an error on web_search although the API docs say it's supported.
+        // https://platform.openai.com/docs/guides/tools-web-search?api-mode=responses#:~:text=%7B%20type%3A%20%22web_search%22%20%7D%2C
+        #[serde(rename = "web_search")]
+        WebSearch {},
+        #[serde(rename = "custom")]
+        Freeform(FreeformTool),
+    }
+
+    impl ToolSpec {
+        pub(crate) fn name(&self) -> &str {
+            match self {
+                ToolSpec::Function(tool) => tool.name.as_str(),
+                ToolSpec::LocalShell {} => "local_shell",
+                ToolSpec::WebSearch {} => "web_search",
+                ToolSpec::Freeform(tool) => tool.name.as_str(),
+            }
+        }
+    }
+
+    #[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
+    pub struct FreeformTool {
+        pub(crate) name: String,
+        pub(crate) description: String,
+        pub(crate) format: FreeformToolFormat,
+    }
+
+    #[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
+    pub struct FreeformToolFormat {
+        pub(crate) r#type: String,
+        pub(crate) syntax: String,
+        pub(crate) definition: String,
+    }
+
+    #[derive(Debug, Clone, Serialize, PartialEq)]
+    pub struct ResponsesApiTool {
+        pub(crate) name: String,
+        pub(crate) description: String,
+        /// TODO: Validation. When strict is set to true, the JSON schema,
+        /// `required` and `additional_properties` must be present. All fields in
+        /// `properties` must be present in `required`.
+        pub(crate) strict: bool,
+        pub(crate) parameters: JsonSchema,
+    }
+}
+
 pub(crate) fn create_reasoning_param_for_request(
     model_family: &ModelFamily,
     effort: Option<ReasoningEffortConfig>,
@@ -278,8 +460,8 @@ mod tests {
             instructions: "i",
             input: &input,
             tools: &tools,
-            tool_choice: "auto",
-            parallel_tool_calls: false,
+            tool_choice: serde_json::json!("auto"),
+            parallel_tool_calls: true,
             reasoning: None,
             store: false,
             stream: true,
@@ -319,8 +501,8 @@ mod tests {
             instructions: "i",
             input: &input,
             tools: &tools,
-            tool_choice: "auto",
-            parallel_tool_calls: false,
+            tool_choice: serde_json::json!("auto"),
+            parallel_tool_calls: true,
             reasoning: None,
             store: false,
             stream: true,
@@ -355,8 +537,8 @@ mod tests {
             instructions: "i",
             input: &input,
             tools: &tools,
-            tool_choice: "auto",
-            parallel_tool_calls: false,
+            tool_choice: serde_json::json!("auto"),
+            parallel_tool_calls: true,
             reasoning: None,
             store: false,
             stream: true,

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

@@ -16,6 +16,7 @@ use crate::protocol::InputItem;
 use crate::protocol::InputMessageKind;
 use crate::protocol::TaskStartedEvent;
 use crate::protocol::TurnContextItem;
+use crate::state::TaskKind;
 use crate::truncate::truncate_middle;
 use crate::util::backoff;
 use askama::Template;
@@ -70,14 +71,10 @@ async fn run_compact_task_inner(
     input: Vec<InputItem>,
 ) {
     let initial_input_for_turn: ResponseInputItem = ResponseInputItem::from(input);
-    let turn_input = sess
+    let mut turn_input = sess
         .turn_input_with_history(vec![initial_input_for_turn.clone().into()])
         .await;
-
-    let prompt = Prompt {
-        input: turn_input,
-        ..Default::default()
-    };
+    let mut truncated_count = 0usize;
 
     let max_retries = turn_context.client.get_provider().stream_max_retries();
     let mut retries = 0;
@@ -93,25 +90,55 @@ async fn run_compact_task_inner(
     sess.persist_rollout_items(&[rollout_item]).await;
 
     loop {
+        let prompt = Prompt {
+            input: turn_input.clone(),
+            disabled_tools: turn_context.disabled_tools.clone(),
+            ..Default::default()
+        };
         let attempt_result =
             drain_to_completed(&sess, turn_context.as_ref(), &sub_id, &prompt).await;
 
         match attempt_result {
             Ok(()) => {
+                if truncated_count > 0 {
+                    sess.notify_background_event(
+                        &sub_id,
+                        format!(
+                            "Trimmed {truncated_count} older conversation item(s) before compacting so the prompt fits the model context window."
+                        ),
+                    )
+                    .await;
+                }
                 break;
             }
             Err(CodexErr::Interrupted) => {
                 return;
             }
+            Err(e @ CodexErr::ContextWindowExceeded) => {
+                if turn_input.len() > 1 {
+                    turn_input.remove(0);
+                    truncated_count += 1;
+                    retries = 0;
+                    continue;
+                }
+                sess.set_total_tokens_full(&sub_id, turn_context.as_ref())
+                    .await;
+                let event = Event {
+                    id: sub_id.clone(),
+                    msg: EventMsg::Error(ErrorEvent {
+                        message: e.to_string(),
+                    }),
+                };
+                sess.send_event(event).await;
+                return;
+            }
             Err(e) => {
                 if retries < max_retries {
                     retries += 1;
                     let delay = backoff(retries);
                     sess.notify_stream_error(
                         &sub_id,
-                        format!(
-                            "stream error: {e}; retrying {retries}/{max_retries} in {delay:?}…"
-                        ),
+                        format!("Re-connecting... {retries}/{max_retries}"),
                     )
                     .await;
                     tokio::time::sleep(delay).await;
@@ -233,7 +260,11 @@ async fn drain_to_completed(
     sub_id: &str,
     prompt: &Prompt,
 ) -> CodexResult<()> {
-    let mut stream = turn_context.client.clone().stream(prompt).await?;
+    let mut stream = turn_context
+        .client
+        .clone()
+        .stream_with_task_kind(prompt, TaskKind::Compact)
+        .await?;
     loop {
         let maybe_event = stream.next().await;
         let Some(event) = maybe_event else {

codex-rs/core/src/config_loader/macos.rs

@@ -0,0 +1,118 @@
+use std::io;
+use toml::Value as TomlValue;
+
+#[cfg(target_os = "macos")]
+mod native {
+    use super::*;
+    use base64::Engine;
+    use base64::prelude::BASE64_STANDARD;
+    use core_foundation::base::TCFType;
+    use core_foundation::string::CFString;
+    use core_foundation::string::CFStringRef;
+    use std::ffi::c_void;
+    use tokio::task;
+
+    pub(crate) async fn load_managed_admin_config_layer(
+        override_base64: Option<&str>,
+    ) -> io::Result<Option<TomlValue>> {
+        if let Some(encoded) = override_base64 {
+            let trimmed = encoded.trim();
+            return if trimmed.is_empty() {
+                Ok(None)
+            } else {
+                parse_managed_preferences_base64(trimmed).map(Some)
+            };
+        }
+
+        const LOAD_ERROR: &str = "Failed to load managed preferences configuration";
+
+        match task::spawn_blocking(load_managed_admin_config).await {
+            Ok(result) => result,
+            Err(join_err) => {
+                if join_err.is_cancelled() {
+                    tracing::error!("Managed preferences load task was cancelled");
+                } else {
+                    tracing::error!("Managed preferences load task failed: {join_err}");
+                }
+                Err(io::Error::other(LOAD_ERROR))
+            }
+        }
+    }
+
+    pub(super) fn load_managed_admin_config() -> io::Result<Option<TomlValue>> {
+        #[link(name = "CoreFoundation", kind = "framework")]
+        unsafe extern "C" {
+            fn CFPreferencesCopyAppValue(
+                key: CFStringRef,
+                application_id: CFStringRef,
+            ) -> *mut c_void;
+        }
+
+        const MANAGED_PREFERENCES_APPLICATION_ID: &str = "com.openai.codex";
+        const MANAGED_PREFERENCES_CONFIG_KEY: &str = "config_toml_base64";
+
+        let application_id = CFString::new(MANAGED_PREFERENCES_APPLICATION_ID);
+        let key = CFString::new(MANAGED_PREFERENCES_CONFIG_KEY);
+
+        let value_ref = unsafe {
+            CFPreferencesCopyAppValue(
+                key.as_concrete_TypeRef(),
+                application_id.as_concrete_TypeRef(),
+            )
+        };
+
+        if value_ref.is_null() {
+            tracing::debug!(
+                "Managed preferences for {} key {} not found",
+                MANAGED_PREFERENCES_APPLICATION_ID,
+                MANAGED_PREFERENCES_CONFIG_KEY
+            );
+            return Ok(None);
+        }
+
+        let value = unsafe { CFString::wrap_under_create_rule(value_ref as _) };
+        let contents = value.to_string();
+        let trimmed = contents.trim();
+
+        parse_managed_preferences_base64(trimmed).map(Some)
+    }
+
+    pub(super) fn parse_managed_preferences_base64(encoded: &str) -> io::Result<TomlValue> {
+        let decoded = BASE64_STANDARD.decode(encoded.as_bytes()).map_err(|err| {
+            tracing::error!("Failed to decode managed preferences as base64: {err}");
+            io::Error::new(io::ErrorKind::InvalidData, err)
+        })?;
+
+        let decoded_str = String::from_utf8(decoded).map_err(|err| {
+            tracing::error!("Managed preferences base64 contents were not valid UTF-8: {err}");
+            io::Error::new(io::ErrorKind::InvalidData, err)
+        })?;
+
+        match toml::from_str::<TomlValue>(&decoded_str) {
+            Ok(TomlValue::Table(parsed)) => Ok(TomlValue::Table(parsed)),
+            Ok(other) => {
+                tracing::error!(
+                    "Managed preferences TOML must have a table at the root, found {other:?}",
+                );
+                Err(io::Error::new(
+                    io::ErrorKind::InvalidData,
+                    "managed preferences root must be a table",
+                ))
+            }
+            Err(err) => {
+                tracing::error!("Failed to parse managed preferences TOML: {err}");
+                Err(io::Error::new(io::ErrorKind::InvalidData, err))
+            }
+        }
+    }
+}
+
+#[cfg(target_os = "macos")]
+pub(crate) use native::load_managed_admin_config_layer;
+
+#[cfg(not(target_os = "macos"))]
+pub(crate) async fn load_managed_admin_config_layer(
+    _override_base64: Option<&str>,
+) -> io::Result<Option<TomlValue>> {
+    Ok(None)
+}

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

@@ -0,0 +1,311 @@
+mod macos;
+
+use crate::config::CONFIG_TOML_FILE;
+use macos::load_managed_admin_config_layer;
+use std::io;
+use std::path::Path;
+use std::path::PathBuf;
+use tokio::fs;
+use toml::Value as TomlValue;
+
+#[cfg(unix)]
+const CODEX_MANAGED_CONFIG_SYSTEM_PATH: &str = "/etc/codex/managed_config.toml";
+
+#[derive(Debug)]
+pub(crate) struct LoadedConfigLayers {
+    pub base: TomlValue,
+    pub managed_config: Option<TomlValue>,
+    pub managed_preferences: Option<TomlValue>,
+}
+
+#[derive(Debug, Default)]
+pub(crate) struct LoaderOverrides {
+    pub managed_config_path: Option<PathBuf>,
+    #[cfg(target_os = "macos")]
+    pub managed_preferences_base64: Option<String>,
+}
+
+// Configuration layering pipeline (top overrides bottom):
+//
+//        +-------------------------+
+//        | Managed preferences (*) |
+//        +-------------------------+
+//                    ^
+//                    |
+//        +-------------------------+
+//        |  managed_config.toml   |
+//        +-------------------------+
+//                    ^
+//                    |
+//        +-------------------------+
+//        |    config.toml (base)   |
+//        +-------------------------+
+//
+// (*) Only available on macOS via managed device profiles.
+
+pub async fn load_config_as_toml(codex_home: &Path) -> io::Result<TomlValue> {
+    load_config_as_toml_with_overrides(codex_home, LoaderOverrides::default()).await
+}
+
+fn default_empty_table() -> TomlValue {
+    TomlValue::Table(Default::default())
+}
+
+pub(crate) async fn load_config_layers_with_overrides(
+    codex_home: &Path,
+    overrides: LoaderOverrides,
+) -> io::Result<LoadedConfigLayers> {
+    load_config_layers_internal(codex_home, overrides).await
+}
+
+async fn load_config_as_toml_with_overrides(
+    codex_home: &Path,
+    overrides: LoaderOverrides,
+) -> io::Result<TomlValue> {
+    let layers = load_config_layers_internal(codex_home, overrides).await?;
+    Ok(apply_managed_layers(layers))
+}
+
+async fn load_config_layers_internal(
+    codex_home: &Path,
+    overrides: LoaderOverrides,
+) -> io::Result<LoadedConfigLayers> {
+    #[cfg(target_os = "macos")]
+    let LoaderOverrides {
+        managed_config_path,
+        managed_preferences_base64,
+    } = overrides;
+
+    #[cfg(not(target_os = "macos"))]
+    let LoaderOverrides {
+        managed_config_path,
+    } = overrides;
+
+    let managed_config_path =
+        managed_config_path.unwrap_or_else(|| managed_config_default_path(codex_home));
+
+    let user_config_path = codex_home.join(CONFIG_TOML_FILE);
+    let user_config = read_config_from_path(&user_config_path, true).await?;
+    let managed_config = read_config_from_path(&managed_config_path, false).await?;
+
+    #[cfg(target_os = "macos")]
+    let managed_preferences =
+        load_managed_admin_config_layer(managed_preferences_base64.as_deref()).await?;
+
+    #[cfg(not(target_os = "macos"))]
+    let managed_preferences = load_managed_admin_config_layer(None).await?;
+
+    Ok(LoadedConfigLayers {
+        base: user_config.unwrap_or_else(default_empty_table),
+        managed_config,
+        managed_preferences,
+    })
+}
+
+async fn read_config_from_path(
+    path: &Path,
+    log_missing_as_info: bool,
+) -> io::Result<Option<TomlValue>> {
+    match fs::read_to_string(path).await {
+        Ok(contents) => match toml::from_str::<TomlValue>(&contents) {
+            Ok(value) => Ok(Some(value)),
+            Err(err) => {
+                tracing::error!("Failed to parse {}: {err}", path.display());
+                Err(io::Error::new(io::ErrorKind::InvalidData, err))
+            }
+        },
+        Err(err) if err.kind() == io::ErrorKind::NotFound => {
+            if log_missing_as_info {
+                tracing::info!("{} not found, using defaults", path.display());
+            } else {
+                tracing::debug!("{} not found", path.display());
+            }
+            Ok(None)
+        }
+        Err(err) => {
+            tracing::error!("Failed to read {}: {err}", path.display());
+            Err(err)
+        }
+    }
+}
+
+/// Merge config `overlay` into `base`, giving `overlay` precedence.
+pub(crate) fn merge_toml_values(base: &mut TomlValue, overlay: &TomlValue) {
+    if let TomlValue::Table(overlay_table) = overlay
+        && let TomlValue::Table(base_table) = base
+    {
+        for (key, value) in overlay_table {
+            if let Some(existing) = base_table.get_mut(key) {
+                merge_toml_values(existing, value);
+            } else {
+                base_table.insert(key.clone(), value.clone());
+            }
+        }
+    } else {
+        *base = overlay.clone();
+    }
+}
+
+fn managed_config_default_path(codex_home: &Path) -> PathBuf {
+    #[cfg(unix)]
+    {
+        let _ = codex_home;
+        PathBuf::from(CODEX_MANAGED_CONFIG_SYSTEM_PATH)
+    }
+
+    #[cfg(not(unix))]
+    {
+        codex_home.join("managed_config.toml")
+    }
+}
+
+fn apply_managed_layers(layers: LoadedConfigLayers) -> TomlValue {
+    let LoadedConfigLayers {
+        mut base,
+        managed_config,
+        managed_preferences,
+    } = layers;
+
+    for overlay in [managed_config, managed_preferences].into_iter().flatten() {
+        merge_toml_values(&mut base, &overlay);
+    }
+
+    base
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use tempfile::tempdir;
+
+    #[tokio::test]
+    async fn merges_managed_config_layer_on_top() {
+        let tmp = tempdir().expect("tempdir");
+        let managed_path = tmp.path().join("managed_config.toml");
+
+        std::fs::write(
+            tmp.path().join(CONFIG_TOML_FILE),
+            r#"foo = 1
+
+[nested]
+value = "base"
+"#,
+        )
+        .expect("write base");
+        std::fs::write(
+            &managed_path,
+            r#"foo = 2
+
+[nested]
+value = "managed_config"
+extra = true
+"#,
+        )
+        .expect("write managed config");
+
+        let overrides = LoaderOverrides {
+            managed_config_path: Some(managed_path),
+            #[cfg(target_os = "macos")]
+            managed_preferences_base64: None,
+        };
+
+        let loaded = load_config_as_toml_with_overrides(tmp.path(), overrides)
+            .await
+            .expect("load config");
+        let table = loaded.as_table().expect("top-level table expected");
+
+        assert_eq!(table.get("foo"), Some(&TomlValue::Integer(2)));
+        let nested = table
+            .get("nested")
+            .and_then(|v| v.as_table())
+            .expect("nested");
+        assert_eq!(
+            nested.get("value"),
+            Some(&TomlValue::String("managed_config".to_string()))
+        );
+        assert_eq!(nested.get("extra"), Some(&TomlValue::Boolean(true)));
+    }
+
+    #[tokio::test]
+    async fn returns_empty_when_all_layers_missing() {
+        let tmp = tempdir().expect("tempdir");
+        let managed_path = tmp.path().join("managed_config.toml");
+        let overrides = LoaderOverrides {
+            managed_config_path: Some(managed_path),
+            #[cfg(target_os = "macos")]
+            managed_preferences_base64: None,
+        };
+
+        let layers = load_config_layers_with_overrides(tmp.path(), overrides)
+            .await
+            .expect("load layers");
+        let base_table = layers.base.as_table().expect("base table expected");
+        assert!(
+            base_table.is_empty(),
+            "expected empty base layer when configs missing"
+        );
+        assert!(
+            layers.managed_config.is_none(),
+            "managed config layer should be absent when file missing"
+        );
+
+        #[cfg(not(target_os = "macos"))]
+        {
+            let loaded = load_config_as_toml(tmp.path()).await.expect("load config");
+            let table = loaded.as_table().expect("top-level table expected");
+            assert!(
+                table.is_empty(),
+                "expected empty table when configs missing"
+            );
+        }
+    }
+
+    #[cfg(target_os = "macos")]
+    #[tokio::test]
+    async fn managed_preferences_take_highest_precedence() {
+        use base64::Engine;
+
+        let managed_payload = r#"
+[nested]
+value = "managed"
+flag = false
+"#;
+        let encoded = base64::prelude::BASE64_STANDARD.encode(managed_payload.as_bytes());
+        let tmp = tempdir().expect("tempdir");
+        let managed_path = tmp.path().join("managed_config.toml");
+
+        std::fs::write(
+            tmp.path().join(CONFIG_TOML_FILE),
+            r#"[nested]
+value = "base"
+"#,
+        )
+        .expect("write base");
+        std::fs::write(
+            &managed_path,
+            r#"[nested]
+value = "managed_config"
+flag = true
+"#,
+        )
+        .expect("write managed config");
+
+        let overrides = LoaderOverrides {
+            managed_config_path: Some(managed_path),
+            managed_preferences_base64: Some(encoded),
+        };
+
+        let loaded = load_config_as_toml_with_overrides(tmp.path(), overrides)
+            .await
+            .expect("load config");
+        let nested = loaded
+            .get("nested")
+            .and_then(|v| v.as_table())
+            .expect("nested table");
+        assert_eq!(
+            nested.get("value"),
+            Some(&TomlValue::String("managed".to_string()))
+        );
+        assert_eq!(nested.get("flag"), Some(&TomlValue::Boolean(false)));
+    }
+}

codex-rs/core/src/config_profile.rs

@@ -20,6 +20,18 @@ pub struct ConfigProfile {
     pub model_verbosity: Option<Verbosity>,
     pub chatgpt_base_url: Option<String>,
     pub experimental_instructions_file: Option<PathBuf>,
+    pub include_plan_tool: Option<bool>,
+    pub include_apply_patch_tool: Option<bool>,
+    pub include_view_image_tool: Option<bool>,
+    pub experimental_use_unified_exec_tool: Option<bool>,
+    pub experimental_use_exec_command_tool: Option<bool>,
+    pub experimental_use_rmcp_client: Option<bool>,
+    pub experimental_use_freeform_apply_patch: Option<bool>,
+    pub tools_web_search: Option<bool>,
+    pub tools_view_image: Option<bool>,
+    /// Optional feature toggles scoped to this profile.
+    #[serde(default)]
+    pub features: Option<crate::features::FeaturesToml>,
 }
 
 impl From<ConfigProfile> for codex_app_server_protocol::Profile {

codex-rs/core/src/config_types.rs

@@ -20,6 +20,10 @@ pub struct McpServerConfig {
     #[serde(flatten)]
     pub transport: McpServerTransportConfig,
 
+    /// When `false`, Codex skips initializing this MCP server.
+    #[serde(default = "default_enabled")]
+    pub enabled: bool,
+
     /// Startup timeout in seconds for initializing MCP server & initially listing tools.
     #[serde(
         default,
@@ -48,6 +52,7 @@ impl<'de> Deserialize<'de> for McpServerConfig {
 
             url: Option<String>,
             bearer_token: Option<String>,
+            bearer_token_env_var: Option<String>,
 
             #[serde(default)]
             startup_timeout_sec: Option<f64>,
@@ -55,6 +60,8 @@ impl<'de> Deserialize<'de> for McpServerConfig {
             startup_timeout_ms: Option<u64>,
             #[serde(default, with = "option_duration_secs")]
             tool_timeout_sec: Option<Duration>,
+            #[serde(default)]
+            enabled: Option<bool>,
         }
 
         let raw = RawMcpServerConfig::deserialize(deserializer)?;
@@ -86,11 +93,15 @@ impl<'de> Deserialize<'de> for McpServerConfig {
                 args,
                 env,
                 url,
-                bearer_token,
+                bearer_token_env_var,
                 ..
             } => {
                 throw_if_set("stdio", "url", url.as_ref())?;
-                throw_if_set("stdio", "bearer_token", bearer_token.as_ref())?;
+                throw_if_set(
+                    "stdio",
+                    "bearer_token_env_var",
+                    bearer_token_env_var.as_ref(),
+                )?;
                 McpServerTransportConfig::Stdio {
                     command,
                     args: args.unwrap_or_default(),
@@ -100,6 +111,7 @@ impl<'de> Deserialize<'de> for McpServerConfig {
             RawMcpServerConfig {
                 url: Some(url),
                 bearer_token,
+                bearer_token_env_var,
                 command,
                 args,
                 env,
@@ -108,7 +120,11 @@ impl<'de> Deserialize<'de> for McpServerConfig {
                 throw_if_set("streamable_http", "command", command.as_ref())?;
                 throw_if_set("streamable_http", "args", args.as_ref())?;
                 throw_if_set("streamable_http", "env", env.as_ref())?;
-                McpServerTransportConfig::StreamableHttp { url, bearer_token }
+                throw_if_set("streamable_http", "bearer_token", bearer_token.as_ref())?;
+                McpServerTransportConfig::StreamableHttp {
+                    url,
+                    bearer_token_env_var,
+                }
             }
             _ => return Err(SerdeError::custom("invalid transport")),
         };
@@ -117,10 +133,15 @@ impl<'de> Deserialize<'de> for McpServerConfig {
             transport,
             startup_timeout_sec,
             tool_timeout_sec: raw.tool_timeout_sec,
+            enabled: raw.enabled.unwrap_or_else(default_enabled),
         })
     }
 }
 
+const fn default_enabled() -> bool {
+    true
+}
+
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq)]
 #[serde(untagged, deny_unknown_fields, rename_all = "snake_case")]
 pub enum McpServerTransportConfig {
@@ -135,11 +156,11 @@ pub enum McpServerTransportConfig {
     /// https://modelcontextprotocol.io/specification/2025-06-18/basic/transports#streamable-http
     StreamableHttp {
         url: String,
-        /// A plain text bearer token to use for authentication.
-        /// This bearer token will be included in the HTTP request header as an `Authorization: Bearer <token>` header.
-        /// This should be used with caution because it lives on disk in clear text.
+        /// Name of the environment variable to read for an HTTP bearer token.
+        /// When set, requests will include the token via `Authorization: Bearer <token>`.
+        /// The actual secret value must be provided via the environment.
         #[serde(default, skip_serializing_if = "Option::is_none")]
-        bearer_token: Option<String>,
+        bearer_token_env_var: Option<String>,
     },
 }
 
@@ -450,6 +471,7 @@ mod tests {
                 env: None
             }
         );
+        assert!(cfg.enabled);
     }
 
     #[test]
@@ -470,6 +492,7 @@ mod tests {
                 env: None
             }
         );
+        assert!(cfg.enabled);
     }
 
     #[test]
@@ -491,6 +514,20 @@ mod tests {
                 env: Some(HashMap::from([("FOO".to_string(), "BAR".to_string())]))
             }
         );
+        assert!(cfg.enabled);
+    }
+
+    #[test]
+    fn deserialize_disabled_server_config() {
+        let cfg: McpServerConfig = toml::from_str(
+            r#"
+            command = "echo"
+            enabled = false
+        "#,
+        )
+        .expect("should deserialize disabled server config");
+
+        assert!(!cfg.enabled);
     }
 
     #[test]
@@ -506,17 +543,18 @@ mod tests {
             cfg.transport,
             McpServerTransportConfig::StreamableHttp {
                 url: "https://example.com/mcp".to_string(),
-                bearer_token: None
+                bearer_token_env_var: None
             }
         );
+        assert!(cfg.enabled);
     }
 
     #[test]
-    fn deserialize_streamable_http_server_config_with_bearer_token() {
+    fn deserialize_streamable_http_server_config_with_env_var() {
         let cfg: McpServerConfig = toml::from_str(
             r#"
             url = "https://example.com/mcp"
-            bearer_token = "secret"
+            bearer_token_env_var = "GITHUB_TOKEN"
         "#,
         )
         .expect("should deserialize http config");
@@ -525,9 +563,10 @@ mod tests {
             cfg.transport,
             McpServerTransportConfig::StreamableHttp {
                 url: "https://example.com/mcp".to_string(),
-                bearer_token: Some("secret".to_string())
+                bearer_token_env_var: Some("GITHUB_TOKEN".to_string())
             }
         );
+        assert!(cfg.enabled);
     }
 
     #[test]
@@ -553,13 +592,18 @@ mod tests {
     }
 
     #[test]
-    fn deserialize_rejects_bearer_token_for_stdio_transport() {
-        toml::from_str::<McpServerConfig>(
+    fn deserialize_rejects_inline_bearer_token_field() {
+        let err = toml::from_str::<McpServerConfig>(
             r#"
-            command = "echo"
+            url = "https://example.com"
             bearer_token = "secret"
         "#,
         )
-        .expect_err("should reject bearer token for stdio transport");
+        .expect_err("should reject bearer_token field");
+
+        assert!(
+            err.to_string().contains("bearer_token is not supported"),
+            "unexpected error: {err}"
+        );
     }
 }

codex-rs/core/src/conversation_manager.rs

@@ -17,6 +17,7 @@ use codex_protocol::ConversationId;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::protocol::InitialHistory;
 use codex_protocol::protocol::RolloutItem;
+use codex_protocol::protocol::SessionSource;
 use std::collections::HashMap;
 use std::path::PathBuf;
 use std::sync::Arc;
@@ -35,20 +36,25 @@ pub struct NewConversation {
 pub struct ConversationManager {
     conversations: Arc<RwLock<HashMap<ConversationId, Arc<CodexConversation>>>>,
     auth_manager: Arc<AuthManager>,
+    session_source: SessionSource,
 }
 
 impl ConversationManager {
-    pub fn new(auth_manager: Arc<AuthManager>) -> Self {
+    pub fn new(auth_manager: Arc<AuthManager>, session_source: SessionSource) -> Self {
         Self {
             conversations: Arc::new(RwLock::new(HashMap::new())),
             auth_manager,
+            session_source,
         }
     }
 
     /// Construct with a dummy AuthManager containing the provided CodexAuth.
     /// Used for integration tests: should not be used by ordinary business logic.
     pub fn with_auth(auth: CodexAuth) -> Self {
-        Self::new(crate::AuthManager::from_auth_for_testing(auth))
+        Self::new(
+            crate::AuthManager::from_auth_for_testing(auth),
+            SessionSource::Exec,
+        )
     }
 
     pub async fn new_conversation(&self, config: Config) -> CodexResult<NewConversation> {
@@ -64,7 +70,13 @@ impl ConversationManager {
         let CodexSpawnOk {
             codex,
             conversation_id,
-        } = Codex::spawn(config, auth_manager, InitialHistory::New).await?;
+        } = Codex::spawn(
+            config,
+            auth_manager,
+            InitialHistory::New,
+            self.session_source,
+        )
+        .await?;
         self.finalize_spawn(codex, conversation_id).await
     }
 
@@ -121,7 +133,7 @@ impl ConversationManager {
         let CodexSpawnOk {
             codex,
             conversation_id,
-        } = Codex::spawn(config, auth_manager, initial_history).await?;
+        } = Codex::spawn(config, auth_manager, initial_history, self.session_source).await?;
         self.finalize_spawn(codex, conversation_id).await
     }
 
@@ -155,7 +167,7 @@ impl ConversationManager {
         let CodexSpawnOk {
             codex,
             conversation_id,
-        } = Codex::spawn(config, auth_manager, history).await?;
+        } = Codex::spawn(config, auth_manager, history, self.session_source).await?;
 
         self.finalize_spawn(codex, conversation_id).await
     }
@@ -198,6 +210,7 @@ fn truncate_before_nth_user_message(history: InitialHistory, n: usize) -> Initia
 mod tests {
     use super::*;
     use crate::codex::make_session_and_context;
+    use assert_matches::assert_matches;
     use codex_protocol::models::ContentItem;
     use codex_protocol::models::ReasoningItemReasoningSummary;
     use codex_protocol::models::ResponseItem;
@@ -224,7 +237,7 @@ mod tests {
 
     #[test]
     fn drops_from_last_user_only() {
-        let items = vec![
+        let items = [
             user_msg("u1"),
             assistant_msg("a1"),
             assistant_msg("a2"),
@@ -271,7 +284,7 @@ mod tests {
             .map(RolloutItem::ResponseItem)
             .collect();
         let truncated2 = truncate_before_nth_user_message(InitialHistory::Forked(initial2), 2);
-        assert!(matches!(truncated2, InitialHistory::New));
+        assert_matches!(truncated2, InitialHistory::New);
     }
 
     #[test]

codex-rs/core/src/default_client.rs

@@ -20,7 +20,7 @@ use std::sync::OnceLock;
 /// The full user agent string is returned from the mcp initialize response.
 /// Parenthesis will be added by Codex. This should only specify what goes inside of the parenthesis.
 pub static USER_AGENT_SUFFIX: LazyLock<Mutex<Option<String>>> = LazyLock::new(|| Mutex::new(None));
-
+pub const DEFAULT_ORIGINATOR: &str = "codex_cli_rs";
 pub const CODEX_INTERNAL_ORIGINATOR_OVERRIDE_ENV_VAR: &str = "CODEX_INTERNAL_ORIGINATOR_OVERRIDE";
 #[derive(Debug, Clone)]
 pub struct Originator {
@@ -35,10 +35,11 @@ pub enum SetOriginatorError {
     AlreadyInitialized,
 }
 
-fn init_originator_from_env() -> Originator {
-    let default = "codex_cli_rs";
+fn get_originator_value(provided: Option<String>) -> Originator {
     let value = std::env::var(CODEX_INTERNAL_ORIGINATOR_OVERRIDE_ENV_VAR)
-        .unwrap_or_else(|_| default.to_string());
+        .ok()
+        .or(provided)
+        .unwrap_or(DEFAULT_ORIGINATOR.to_string());
 
     match HeaderValue::from_str(&value) {
         Ok(header_value) => Originator {
@@ -48,31 +49,22 @@ fn init_originator_from_env() -> Originator {
         Err(e) => {
             tracing::error!("Unable to turn originator override {value} into header value: {e}");
             Originator {
-                value: default.to_string(),
-                header_value: HeaderValue::from_static(default),
+                value: DEFAULT_ORIGINATOR.to_string(),
+                header_value: HeaderValue::from_static(DEFAULT_ORIGINATOR),
             }
         }
     }
 }
 
-fn build_originator(value: String) -> Result<Originator, SetOriginatorError> {
-    let header_value =
-        HeaderValue::from_str(&value).map_err(|_| SetOriginatorError::InvalidHeaderValue)?;
-    Ok(Originator {
-        value,
-        header_value,
-    })
-}
-
-pub fn set_default_originator(value: &str) -> Result<(), SetOriginatorError> {
-    let originator = build_originator(value.to_string())?;
+pub fn set_default_originator(value: String) -> Result<(), SetOriginatorError> {
+    let originator = get_originator_value(Some(value));
     ORIGINATOR
         .set(originator)
         .map_err(|_| SetOriginatorError::AlreadyInitialized)
 }
 
 pub fn originator() -> &'static Originator {
-    ORIGINATOR.get_or_init(init_originator_from_env)
+    ORIGINATOR.get_or_init(|| get_originator_value(None))
 }
 
 pub fn get_codex_user_agent() -> String {

codex-rs/core/src/error.rs

@@ -1,6 +1,7 @@
 use crate::exec::ExecToolCallOutput;
 use crate::token_data::KnownPlan;
 use crate::token_data::PlanType;
+use crate::truncate::truncate_middle;
 use codex_protocol::ConversationId;
 use codex_protocol::protocol::RateLimitSnapshot;
 use reqwest::StatusCode;
@@ -12,6 +13,9 @@ use tokio::task::JoinError;
 
 pub type Result<T> = std::result::Result<T, CodexErr>;
 
+/// Limit UI error messages to a reasonable size while keeping useful context.
+const ERROR_MESSAGE_UI_MAX_BYTES: usize = 2 * 1024; // 4 KiB
+
 #[derive(Error, Debug)]
 pub enum SandboxErr {
     /// Error from sandbox execution
@@ -55,6 +59,11 @@ pub enum CodexErr {
     #[error("stream disconnected before completion: {0}")]
     Stream(String, Option<Duration>),
 
+    #[error(
+        "Codex ran out of room in the model's context window. Start a new conversation or clear earlier history before retrying."
+    )]
+    ContextWindowExceeded,
+
     #[error("no conversation with id: {0}")]
     ConversationNotFound(ConversationId),
 
@@ -108,6 +117,9 @@ pub enum CodexErr {
     #[error("unsupported operation: {0}")]
     UnsupportedOperation(String),
 
+    #[error("Fatal error: {0}")]
+    Fatal(String),
+
     // -----------------------------------------------------------------
     // Automatic conversions for common external error types
     // -----------------------------------------------------------------
@@ -296,21 +308,44 @@ impl CodexErr {
 }
 
 pub fn get_error_message_ui(e: &CodexErr) -> String {
-    match e {
-        CodexErr::Sandbox(SandboxErr::Denied { output }) => output.stderr.text.clone(),
+    let message = match e {
+        CodexErr::Sandbox(SandboxErr::Denied { output }) => {
+            let aggregated = output.aggregated_output.text.trim();
+            if !aggregated.is_empty() {
+                output.aggregated_output.text.clone()
+            } else {
+                let stderr = output.stderr.text.trim();
+                let stdout = output.stdout.text.trim();
+                match (stderr.is_empty(), stdout.is_empty()) {
+                    (false, false) => format!("{stderr}\n{stdout}"),
+                    (false, true) => output.stderr.text.clone(),
+                    (true, false) => output.stdout.text.clone(),
+                    (true, true) => format!(
+                        "command failed inside sandbox with exit code {}",
+                        output.exit_code
+                    ),
+                }
+            }
+        }
         // Timeouts are not sandbox errors from a UX perspective; present them plainly
-        CodexErr::Sandbox(SandboxErr::Timeout { output }) => format!(
-            "error: command timed out after {} ms",
-            output.duration.as_millis()
-        ),
+        CodexErr::Sandbox(SandboxErr::Timeout { output }) => {
+            format!(
+                "error: command timed out after {} ms",
+                output.duration.as_millis()
+            )
+        }
         _ => e.to_string(),
-    }
+    };
+
+    truncate_middle(&message, ERROR_MESSAGE_UI_MAX_BYTES).0
 }
 
 #[cfg(test)]
 mod tests {
     use super::*;
+    use crate::exec::StreamOutput;
     use codex_protocol::protocol::RateLimitWindow;
+    use pretty_assertions::assert_eq;
 
     fn rate_limit_snapshot() -> RateLimitSnapshot {
         RateLimitSnapshot {
@@ -340,6 +375,73 @@ mod tests {
         );
     }
 
+    #[test]
+    fn sandbox_denied_uses_aggregated_output_when_stderr_empty() {
+        let output = ExecToolCallOutput {
+            exit_code: 77,
+            stdout: StreamOutput::new(String::new()),
+            stderr: StreamOutput::new(String::new()),
+            aggregated_output: StreamOutput::new("aggregate detail".to_string()),
+            duration: Duration::from_millis(10),
+            timed_out: false,
+        };
+        let err = CodexErr::Sandbox(SandboxErr::Denied {
+            output: Box::new(output),
+        });
+        assert_eq!(get_error_message_ui(&err), "aggregate detail");
+    }
+
+    #[test]
+    fn sandbox_denied_reports_both_streams_when_available() {
+        let output = ExecToolCallOutput {
+            exit_code: 9,
+            stdout: StreamOutput::new("stdout detail".to_string()),
+            stderr: StreamOutput::new("stderr detail".to_string()),
+            aggregated_output: StreamOutput::new(String::new()),
+            duration: Duration::from_millis(10),
+            timed_out: false,
+        };
+        let err = CodexErr::Sandbox(SandboxErr::Denied {
+            output: Box::new(output),
+        });
+        assert_eq!(get_error_message_ui(&err), "stderr detail\nstdout detail");
+    }
+
+    #[test]
+    fn sandbox_denied_reports_stdout_when_no_stderr() {
+        let output = ExecToolCallOutput {
+            exit_code: 11,
+            stdout: StreamOutput::new("stdout only".to_string()),
+            stderr: StreamOutput::new(String::new()),
+            aggregated_output: StreamOutput::new(String::new()),
+            duration: Duration::from_millis(8),
+            timed_out: false,
+        };
+        let err = CodexErr::Sandbox(SandboxErr::Denied {
+            output: Box::new(output),
+        });
+        assert_eq!(get_error_message_ui(&err), "stdout only");
+    }
+
+    #[test]
+    fn sandbox_denied_reports_exit_code_when_no_output_available() {
+        let output = ExecToolCallOutput {
+            exit_code: 13,
+            stdout: StreamOutput::new(String::new()),
+            stderr: StreamOutput::new(String::new()),
+            aggregated_output: StreamOutput::new(String::new()),
+            duration: Duration::from_millis(5),
+            timed_out: false,
+        };
+        let err = CodexErr::Sandbox(SandboxErr::Denied {
+            output: Box::new(output),
+        });
+        assert_eq!(
+            get_error_message_ui(&err),
+            "command failed inside sandbox with exit code 13"
+        );
+    }
+
     #[test]
     fn usage_limit_reached_error_formats_free_plan() {
         let err = UsageLimitReachedError {

codex-rs/core/src/event_mapping.rs

@@ -127,6 +127,7 @@ mod tests {
     use super::map_response_item_to_event_messages;
     use crate::protocol::EventMsg;
     use crate::protocol::InputMessageKind;
+    use assert_matches::assert_matches;
     use codex_protocol::models::ContentItem;
     use codex_protocol::models::ResponseItem;
     use pretty_assertions::assert_eq;
@@ -158,7 +159,7 @@ mod tests {
         match &events[0] {
             EventMsg::UserMessage(user) => {
                 assert_eq!(user.message, "Hello world");
-                assert!(matches!(user.kind, Some(InputMessageKind::Plain)));
+                assert_matches!(user.kind, Some(InputMessageKind::Plain));
                 assert_eq!(user.images, Some(vec![img1, img2]));
             }
             other => panic!("expected UserMessage, got {other:?}"),

codex-rs/core/src/exec.rs

@@ -177,7 +177,7 @@ pub async fn process_exec_tool_call(
                 }));
             }
 
-            if exit_code != 0 && is_likely_sandbox_denied(sandbox_type, exit_code) {
+            if is_likely_sandbox_denied(sandbox_type, &exec_output) {
                 return Err(CodexErr::Sandbox(SandboxErr::Denied {
                     output: Box::new(exec_output),
                 }));
@@ -195,21 +195,57 @@ pub async fn process_exec_tool_call(
 /// We don't have a fully deterministic way to tell if our command failed
 /// because of the sandbox - a command in the user's zshrc file might hit an
 /// error, but the command itself might fail or succeed for other reasons.
-/// For now, we conservatively check for 'command not found' (exit code 127),
-/// and can add additional cases as necessary.
-fn is_likely_sandbox_denied(sandbox_type: SandboxType, exit_code: i32) -> bool {
-    if sandbox_type == SandboxType::None {
+/// For now, we conservatively check for well known command failure exit codes and
+/// also look for common sandbox denial keywords in the command output.
+fn is_likely_sandbox_denied(sandbox_type: SandboxType, exec_output: &ExecToolCallOutput) -> bool {
+    if sandbox_type == SandboxType::None || exec_output.exit_code == 0 {
         return false;
     }
 
     // Quick rejects: well-known non-sandbox shell exit codes
-    // 127: command not found, 2: misuse of shell builtins
-    if exit_code == 127 {
+    // 2: misuse of shell builtins
+    // 126: permission denied
+    // 127: command not found
+    const QUICK_REJECT_EXIT_CODES: [i32; 3] = [2, 126, 127];
+    if QUICK_REJECT_EXIT_CODES.contains(&exec_output.exit_code) {
         return false;
     }
 
-    // For all other cases, we assume the sandbox is the cause
-    true
+    const SANDBOX_DENIED_KEYWORDS: [&str; 6] = [
+        "operation not permitted",
+        "permission denied",
+        "read-only file system",
+        "seccomp",
+        "sandbox",
+        "landlock",
+    ];
+
+    if [
+        &exec_output.stderr.text,
+        &exec_output.stdout.text,
+        &exec_output.aggregated_output.text,
+    ]
+    .into_iter()
+    .any(|section| {
+        let lower = section.to_lowercase();
+        SANDBOX_DENIED_KEYWORDS
+            .iter()
+            .any(|needle| lower.contains(needle))
+    }) {
+        return true;
+    }
+
+    #[cfg(unix)]
+    {
+        const SIGSYS_CODE: i32 = libc::SIGSYS;
+        if sandbox_type == SandboxType::LinuxSeccomp
+            && exec_output.exit_code == EXIT_CODE_SIGNAL_BASE + SIGSYS_CODE
+        {
+            return true;
+        }
+    }
+
+    false
 }
 
 #[derive(Debug)]
@@ -436,3 +472,77 @@ fn synthetic_exit_status(code: i32) -> ExitStatus {
     #[expect(clippy::unwrap_used)]
     std::process::ExitStatus::from_raw(code.try_into().unwrap())
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::time::Duration;
+
+    fn make_exec_output(
+        exit_code: i32,
+        stdout: &str,
+        stderr: &str,
+        aggregated: &str,
+    ) -> ExecToolCallOutput {
+        ExecToolCallOutput {
+            exit_code,
+            stdout: StreamOutput::new(stdout.to_string()),
+            stderr: StreamOutput::new(stderr.to_string()),
+            aggregated_output: StreamOutput::new(aggregated.to_string()),
+            duration: Duration::from_millis(1),
+            timed_out: false,
+        }
+    }
+
+    #[test]
+    fn sandbox_detection_requires_keywords() {
+        let output = make_exec_output(1, "", "", "");
+        assert!(!is_likely_sandbox_denied(
+            SandboxType::LinuxSeccomp,
+            &output
+        ));
+    }
+
+    #[test]
+    fn sandbox_detection_identifies_keyword_in_stderr() {
+        let output = make_exec_output(1, "", "Operation not permitted", "");
+        assert!(is_likely_sandbox_denied(SandboxType::LinuxSeccomp, &output));
+    }
+
+    #[test]
+    fn sandbox_detection_respects_quick_reject_exit_codes() {
+        let output = make_exec_output(127, "", "command not found", "");
+        assert!(!is_likely_sandbox_denied(
+            SandboxType::LinuxSeccomp,
+            &output
+        ));
+    }
+
+    #[test]
+    fn sandbox_detection_ignores_non_sandbox_mode() {
+        let output = make_exec_output(1, "", "Operation not permitted", "");
+        assert!(!is_likely_sandbox_denied(SandboxType::None, &output));
+    }
+
+    #[test]
+    fn sandbox_detection_uses_aggregated_output() {
+        let output = make_exec_output(
+            101,
+            "",
+            "",
+            "cargo failed: Read-only file system when writing target",
+        );
+        assert!(is_likely_sandbox_denied(
+            SandboxType::MacosSeatbelt,
+            &output
+        ));
+    }
+
+    #[cfg(unix)]
+    #[test]
+    fn sandbox_detection_flags_sigsys_exit_code() {
+        let exit_code = EXIT_CODE_SIGNAL_BASE + libc::SIGSYS;
+        let output = make_exec_output(exit_code, "", "", "");
+        assert!(is_likely_sandbox_denied(SandboxType::LinuxSeccomp, &output));
+    }
+}

codex-rs/core/src/exec_command/responses_api.rs

@@ -1,7 +1,7 @@
 use std::collections::BTreeMap;
 
+use crate::client_common::tools::ResponsesApiTool;
 use crate::openai_tools::JsonSchema;
-use crate::openai_tools::ResponsesApiTool;
 
 pub const EXEC_COMMAND_TOOL_NAME: &str = "exec_command";
 pub const WRITE_STDIN_TOOL_NAME: &str = "write_stdin";
@@ -49,7 +49,7 @@ pub fn create_exec_command_tool_for_responses_api() -> ResponsesApiTool {
         parameters: JsonSchema::Object {
             properties,
             required: Some(vec!["cmd".to_string()]),
-            additional_properties: Some(false),
+            additional_properties: Some(false.into()),
         },
     }
 }
@@ -92,7 +92,7 @@ Can write control characters (\u0003 for Ctrl-C), or an empty string to just pol
         parameters: JsonSchema::Object {
             properties,
             required: Some(vec!["session_id".to_string(), "chars".to_string()]),
-            additional_properties: Some(false),
+            additional_properties: Some(false.into()),
         },
     }
 }

codex-rs/core/src/executor/backends.rs

@@ -6,6 +6,7 @@ use async_trait::async_trait;
 use crate::CODEX_APPLY_PATCH_ARG1;
 use crate::apply_patch::ApplyPatchExec;
 use crate::exec::ExecParams;
+use crate::executor::ExecutorConfig;
 use crate::function_tool::FunctionCallError;
 
 pub(crate) enum ExecutionMode {
@@ -22,6 +23,7 @@ pub(crate) trait ExecutionBackend: Send + Sync {
         params: ExecParams,
         // Required for downcasting the apply_patch.
         mode: &ExecutionMode,
+        config: &ExecutorConfig,
     ) -> Result<ExecParams, FunctionCallError>;
 
     fn stream_stdout(&self, _mode: &ExecutionMode) -> bool {
@@ -47,6 +49,7 @@ impl ExecutionBackend for ShellBackend {
         &self,
         params: ExecParams,
         mode: &ExecutionMode,
+        _config: &ExecutorConfig,
     ) -> Result<ExecParams, FunctionCallError> {
         match mode {
             ExecutionMode::Shell => Ok(params),
@@ -65,17 +68,22 @@ impl ExecutionBackend for ApplyPatchBackend {
         &self,
         params: ExecParams,
         mode: &ExecutionMode,
+        config: &ExecutorConfig,
     ) -> Result<ExecParams, FunctionCallError> {
         match mode {
             ExecutionMode::ApplyPatch(exec) => {
-                let path_to_codex = env::current_exe()
-                    .ok()
-                    .map(|p| p.to_string_lossy().to_string())
-                    .ok_or_else(|| {
-                        FunctionCallError::RespondToModel(
-                            "failed to determine path to codex executable".to_string(),
-                        )
-                    })?;
+                let path_to_codex = if let Some(exe_path) = &config.codex_exe {
+                    exe_path.to_string_lossy().to_string()
+                } else {
+                    env::current_exe()
+                        .ok()
+                        .map(|p| p.to_string_lossy().to_string())
+                        .ok_or_else(|| {
+                            FunctionCallError::RespondToModel(
+                                "failed to determine path to codex executable".to_string(),
+                            )
+                        })?
+                };
 
                 let patch = exec.action.patch.clone();
                 Ok(ExecParams {

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

@@ -10,11 +10,11 @@ pub(crate) use runner::ExecutorConfig;
 pub(crate) use runner::normalize_exec_result;
 
 pub(crate) mod linkers {
-    use crate::codex::ExecCommandContext;
     use crate::exec::ExecParams;
     use crate::exec::StdoutStream;
     use crate::executor::backends::ExecutionMode;
     use crate::executor::runner::ExecutionRequest;
+    use crate::tools::context::ExecCommandContext;
 
     pub struct PreparedExec {
         pub(crate) context: ExecCommandContext,

codex-rs/core/src/executor/runner.rs

@@ -6,7 +6,6 @@ use std::time::Duration;
 use super::backends::ExecutionMode;
 use super::backends::backend_for_mode;
 use super::cache::ApprovalCache;
-use crate::codex::ExecCommandContext;
 use crate::codex::Session;
 use crate::error::CodexErr;
 use crate::error::SandboxErr;
@@ -24,25 +23,26 @@ use crate::protocol::AskForApproval;
 use crate::protocol::ReviewDecision;
 use crate::protocol::SandboxPolicy;
 use crate::shell;
+use crate::tools::context::ExecCommandContext;
 use codex_otel::otel_event_manager::ToolDecisionSource;
 
 #[derive(Clone, Debug)]
 pub(crate) struct ExecutorConfig {
     pub(crate) sandbox_policy: SandboxPolicy,
     pub(crate) sandbox_cwd: PathBuf,
-    codex_linux_sandbox_exe: Option<PathBuf>,
+    pub(crate) codex_exe: Option<PathBuf>,
 }
 
 impl ExecutorConfig {
     pub(crate) fn new(
         sandbox_policy: SandboxPolicy,
         sandbox_cwd: PathBuf,
-        codex_linux_sandbox_exe: Option<PathBuf>,
+        codex_exe: Option<PathBuf>,
     ) -> Self {
         Self {
             sandbox_policy,
             sandbox_cwd,
-            codex_linux_sandbox_exe,
+            codex_exe,
         }
     }
 }
@@ -86,7 +86,14 @@ impl Executor {
                 maybe_translate_shell_command(request.params, session, request.use_shell_profile);
         }
 
-        // Step 1: Normalise parameters via the selected backend.
+        // Step 1: Snapshot sandbox configuration so it stays stable for this run.
+        let config = self
+            .config
+            .read()
+            .map_err(|_| ExecError::rejection("executor config poisoned"))?
+            .clone();
+
+        // Step 2: Normalise parameters via the selected backend.
         let backend = backend_for_mode(&request.mode);
         let stdout_stream = if backend.stream_stdout(&request.mode) {
             request.stdout_stream.clone()
@@ -94,16 +101,9 @@ impl Executor {
             None
         };
         request.params = backend
-            .prepare(request.params, &request.mode)
+            .prepare(request.params, &request.mode, &config)
             .map_err(ExecError::from)?;
 
-        // Step 2: Snapshot sandbox configuration so it stays stable for this run.
-        let config = self
-            .config
-            .read()
-            .map_err(|_| ExecError::rejection("executor config poisoned"))?
-            .clone();
-
         // Step 3: Decide sandbox placement, prompting for approval when needed.
         let sandbox_decision = select_sandbox(
             &request,
@@ -148,10 +148,8 @@ impl Executor {
                     )
                     .await
                 } else {
-                    Err(ExecError::rejection(format!(
-                        "failed in sandbox {:?} with execution error: {error:?}",
-                        sandbox_decision.initial_sandbox
-                    )))
+                    let message = sandbox_failure_message(error);
+                    Err(ExecError::rejection(message))
                 }
             }
             Err(err) => Err(err.into()),
@@ -229,7 +227,7 @@ impl Executor {
             sandbox,
             &config.sandbox_policy,
             &config.sandbox_cwd,
-            &config.codex_linux_sandbox_exe,
+            &config.codex_exe,
             stdout_stream,
         )
         .await
@@ -255,6 +253,12 @@ fn maybe_translate_shell_command(
     params
 }
 
+fn sandbox_failure_message(error: SandboxErr) -> String {
+    let codex_error = CodexErr::Sandbox(error);
+    let friendly = get_error_message_ui(&codex_error);
+    format!("failed in sandbox: {friendly}")
+}
+
 pub(crate) struct ExecutionRequest {
     pub params: ExecParams,
     pub approval_command: Vec<String>,
@@ -299,6 +303,7 @@ pub(crate) fn normalize_exec_result(
             let message = match err {
                 ExecError::Function(FunctionCallError::RespondToModel(msg)) => msg.clone(),
                 ExecError::Codex(e) => get_error_message_ui(e),
+                err => err.to_string(),
             };
             let synthetic = ExecToolCallOutput {
                 exit_code: -1,
@@ -358,6 +363,40 @@ mod tests {
         );
     }
 
+    #[test]
+    fn sandbox_failure_message_uses_denied_stderr() {
+        let output = ExecToolCallOutput {
+            exit_code: 101,
+            stdout: StreamOutput::new(String::new()),
+            stderr: StreamOutput::new("sandbox stderr".to_string()),
+            aggregated_output: StreamOutput::new(String::new()),
+            duration: Duration::from_millis(10),
+            timed_out: false,
+        };
+        let err = SandboxErr::Denied {
+            output: Box::new(output),
+        };
+        let message = sandbox_failure_message(err);
+        assert_eq!(message, "failed in sandbox: sandbox stderr");
+    }
+
+    #[test]
+    fn sandbox_failure_message_falls_back_to_aggregated_output() {
+        let output = ExecToolCallOutput {
+            exit_code: 101,
+            stdout: StreamOutput::new(String::new()),
+            stderr: StreamOutput::new(String::new()),
+            aggregated_output: StreamOutput::new("aggregate text".to_string()),
+            duration: Duration::from_millis(10),
+            timed_out: false,
+        };
+        let err = SandboxErr::Denied {
+            output: Box::new(output),
+        };
+        let message = sandbox_failure_message(err);
+        assert_eq!(message, "failed in sandbox: aggregate text");
+    }
+
     #[test]
     fn normalize_function_error_synthesizes_payload() {
         let err = FunctionCallError::RespondToModel("boom".to_string());

codex-rs/core/src/features.rs

@@ -0,0 +1,250 @@
+//! Centralized feature flags and metadata.
+//!
+//! This module defines a small set of toggles that gate experimental and
+//! optional behavior across the codebase. Instead of wiring individual
+//! booleans through multiple types, call sites consult a single `Features`
+//! container attached to `Config`.
+
+use crate::config::ConfigToml;
+use crate::config_profile::ConfigProfile;
+use serde::Deserialize;
+use std::collections::BTreeMap;
+use std::collections::BTreeSet;
+
+mod legacy;
+pub(crate) use legacy::LegacyFeatureToggles;
+
+/// High-level lifecycle stage for a feature.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum Stage {
+    Experimental,
+    Beta,
+    Stable,
+    Deprecated,
+    Removed,
+}
+
+/// Unique features toggled via configuration.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
+pub enum Feature {
+    /// Use the single unified PTY-backed exec tool.
+    UnifiedExec,
+    /// Use the streamable exec-command/write-stdin tool pair.
+    StreamableShell,
+    /// Use the official Rust MCP client (rmcp).
+    RmcpClient,
+    /// Include the plan tool.
+    PlanTool,
+    /// Include the freeform apply_patch tool.
+    ApplyPatchFreeform,
+    /// Include the view_image tool.
+    ViewImageTool,
+    /// Allow the model to request web searches.
+    WebSearchRequest,
+}
+
+impl Feature {
+    pub fn key(self) -> &'static str {
+        self.info().key
+    }
+
+    pub fn stage(self) -> Stage {
+        self.info().stage
+    }
+
+    pub fn default_enabled(self) -> bool {
+        self.info().default_enabled
+    }
+
+    fn info(self) -> &'static FeatureSpec {
+        FEATURES
+            .iter()
+            .find(|spec| spec.id == self)
+            .unwrap_or_else(|| unreachable!("missing FeatureSpec for {:?}", self))
+    }
+}
+
+/// Holds the effective set of enabled features.
+#[derive(Debug, Clone, Default, PartialEq)]
+pub struct Features {
+    enabled: BTreeSet<Feature>,
+}
+
+#[derive(Debug, Clone, Default)]
+pub struct FeatureOverrides {
+    pub include_plan_tool: Option<bool>,
+    pub include_apply_patch_tool: Option<bool>,
+    pub include_view_image_tool: Option<bool>,
+    pub web_search_request: Option<bool>,
+}
+
+impl FeatureOverrides {
+    fn apply(self, features: &mut Features) {
+        LegacyFeatureToggles {
+            include_plan_tool: self.include_plan_tool,
+            include_apply_patch_tool: self.include_apply_patch_tool,
+            include_view_image_tool: self.include_view_image_tool,
+            tools_web_search: self.web_search_request,
+            ..Default::default()
+        }
+        .apply(features);
+    }
+}
+
+impl Features {
+    /// Starts with built-in defaults.
+    pub fn with_defaults() -> Self {
+        let mut set = BTreeSet::new();
+        for spec in FEATURES {
+            if spec.default_enabled {
+                set.insert(spec.id);
+            }
+        }
+        Self { enabled: set }
+    }
+
+    pub fn enabled(&self, f: Feature) -> bool {
+        self.enabled.contains(&f)
+    }
+
+    pub fn enable(&mut self, f: Feature) {
+        self.enabled.insert(f);
+    }
+
+    pub fn disable(&mut self, f: Feature) {
+        self.enabled.remove(&f);
+    }
+
+    /// Apply a table of key -> bool toggles (e.g. from TOML).
+    pub fn apply_map(&mut self, m: &BTreeMap<String, bool>) {
+        for (k, v) in m {
+            match feature_for_key(k) {
+                Some(feat) => {
+                    if *v {
+                        self.enable(feat);
+                    } else {
+                        self.disable(feat);
+                    }
+                }
+                None => {
+                    tracing::warn!("unknown feature key in config: {k}");
+                }
+            }
+        }
+    }
+
+    pub fn from_config(
+        cfg: &ConfigToml,
+        config_profile: &ConfigProfile,
+        overrides: FeatureOverrides,
+    ) -> Self {
+        let mut features = Features::with_defaults();
+
+        let base_legacy = LegacyFeatureToggles {
+            experimental_use_freeform_apply_patch: cfg.experimental_use_freeform_apply_patch,
+            experimental_use_exec_command_tool: cfg.experimental_use_exec_command_tool,
+            experimental_use_unified_exec_tool: cfg.experimental_use_unified_exec_tool,
+            experimental_use_rmcp_client: cfg.experimental_use_rmcp_client,
+            tools_web_search: cfg.tools.as_ref().and_then(|t| t.web_search),
+            tools_view_image: cfg.tools.as_ref().and_then(|t| t.view_image),
+            ..Default::default()
+        };
+        base_legacy.apply(&mut features);
+
+        if let Some(base_features) = cfg.features.as_ref() {
+            features.apply_map(&base_features.entries);
+        }
+
+        let profile_legacy = LegacyFeatureToggles {
+            include_plan_tool: config_profile.include_plan_tool,
+            include_apply_patch_tool: config_profile.include_apply_patch_tool,
+            include_view_image_tool: config_profile.include_view_image_tool,
+            experimental_use_freeform_apply_patch: config_profile
+                .experimental_use_freeform_apply_patch,
+            experimental_use_exec_command_tool: config_profile.experimental_use_exec_command_tool,
+            experimental_use_unified_exec_tool: config_profile.experimental_use_unified_exec_tool,
+            experimental_use_rmcp_client: config_profile.experimental_use_rmcp_client,
+            tools_web_search: config_profile.tools_web_search,
+            tools_view_image: config_profile.tools_view_image,
+        };
+        profile_legacy.apply(&mut features);
+        if let Some(profile_features) = config_profile.features.as_ref() {
+            features.apply_map(&profile_features.entries);
+        }
+
+        overrides.apply(&mut features);
+
+        features
+    }
+}
+
+/// Keys accepted in `[features]` tables.
+fn feature_for_key(key: &str) -> Option<Feature> {
+    for spec in FEATURES {
+        if spec.key == key {
+            return Some(spec.id);
+        }
+    }
+    legacy::feature_for_key(key)
+}
+
+/// Deserializable features table for TOML.
+#[derive(Deserialize, Debug, Clone, Default, PartialEq)]
+pub struct FeaturesToml {
+    #[serde(flatten)]
+    pub entries: BTreeMap<String, bool>,
+}
+
+/// Single, easy-to-read registry of all feature definitions.
+#[derive(Debug, Clone, Copy)]
+pub struct FeatureSpec {
+    pub id: Feature,
+    pub key: &'static str,
+    pub stage: Stage,
+    pub default_enabled: bool,
+}
+
+pub const FEATURES: &[FeatureSpec] = &[
+    FeatureSpec {
+        id: Feature::UnifiedExec,
+        key: "unified_exec",
+        stage: Stage::Experimental,
+        default_enabled: false,
+    },
+    FeatureSpec {
+        id: Feature::StreamableShell,
+        key: "streamable_shell",
+        stage: Stage::Experimental,
+        default_enabled: false,
+    },
+    FeatureSpec {
+        id: Feature::RmcpClient,
+        key: "rmcp_client",
+        stage: Stage::Experimental,
+        default_enabled: false,
+    },
+    FeatureSpec {
+        id: Feature::PlanTool,
+        key: "plan_tool",
+        stage: Stage::Stable,
+        default_enabled: false,
+    },
+    FeatureSpec {
+        id: Feature::ApplyPatchFreeform,
+        key: "apply_patch_freeform",
+        stage: Stage::Beta,
+        default_enabled: false,
+    },
+    FeatureSpec {
+        id: Feature::ViewImageTool,
+        key: "view_image_tool",
+        stage: Stage::Stable,
+        default_enabled: true,
+    },
+    FeatureSpec {
+        id: Feature::WebSearchRequest,
+        key: "web_search_request",
+        stage: Stage::Stable,
+        default_enabled: false,
+    },
+];

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

@@ -0,0 +1,158 @@
+use super::Feature;
+use super::Features;
+use tracing::info;
+
+#[derive(Clone, Copy)]
+struct Alias {
+    legacy_key: &'static str,
+    feature: Feature,
+}
+
+const ALIASES: &[Alias] = &[
+    Alias {
+        legacy_key: "experimental_use_unified_exec_tool",
+        feature: Feature::UnifiedExec,
+    },
+    Alias {
+        legacy_key: "experimental_use_exec_command_tool",
+        feature: Feature::StreamableShell,
+    },
+    Alias {
+        legacy_key: "experimental_use_rmcp_client",
+        feature: Feature::RmcpClient,
+    },
+    Alias {
+        legacy_key: "experimental_use_freeform_apply_patch",
+        feature: Feature::ApplyPatchFreeform,
+    },
+    Alias {
+        legacy_key: "include_apply_patch_tool",
+        feature: Feature::ApplyPatchFreeform,
+    },
+    Alias {
+        legacy_key: "include_plan_tool",
+        feature: Feature::PlanTool,
+    },
+    Alias {
+        legacy_key: "include_view_image_tool",
+        feature: Feature::ViewImageTool,
+    },
+    Alias {
+        legacy_key: "web_search",
+        feature: Feature::WebSearchRequest,
+    },
+];
+
+pub(crate) fn feature_for_key(key: &str) -> Option<Feature> {
+    ALIASES
+        .iter()
+        .find(|alias| alias.legacy_key == key)
+        .map(|alias| {
+            log_alias(alias.legacy_key, alias.feature);
+            alias.feature
+        })
+}
+
+#[derive(Debug, Default)]
+pub struct LegacyFeatureToggles {
+    pub include_plan_tool: Option<bool>,
+    pub include_apply_patch_tool: Option<bool>,
+    pub include_view_image_tool: Option<bool>,
+    pub experimental_use_freeform_apply_patch: Option<bool>,
+    pub experimental_use_exec_command_tool: Option<bool>,
+    pub experimental_use_unified_exec_tool: Option<bool>,
+    pub experimental_use_rmcp_client: Option<bool>,
+    pub tools_web_search: Option<bool>,
+    pub tools_view_image: Option<bool>,
+}
+
+impl LegacyFeatureToggles {
+    pub fn apply(self, features: &mut Features) {
+        set_if_some(
+            features,
+            Feature::PlanTool,
+            self.include_plan_tool,
+            "include_plan_tool",
+        );
+        set_if_some(
+            features,
+            Feature::ApplyPatchFreeform,
+            self.include_apply_patch_tool,
+            "include_apply_patch_tool",
+        );
+        set_if_some(
+            features,
+            Feature::ApplyPatchFreeform,
+            self.experimental_use_freeform_apply_patch,
+            "experimental_use_freeform_apply_patch",
+        );
+        set_if_some(
+            features,
+            Feature::StreamableShell,
+            self.experimental_use_exec_command_tool,
+            "experimental_use_exec_command_tool",
+        );
+        set_if_some(
+            features,
+            Feature::UnifiedExec,
+            self.experimental_use_unified_exec_tool,
+            "experimental_use_unified_exec_tool",
+        );
+        set_if_some(
+            features,
+            Feature::RmcpClient,
+            self.experimental_use_rmcp_client,
+            "experimental_use_rmcp_client",
+        );
+        set_if_some(
+            features,
+            Feature::WebSearchRequest,
+            self.tools_web_search,
+            "tools.web_search",
+        );
+        set_if_some(
+            features,
+            Feature::ViewImageTool,
+            self.include_view_image_tool,
+            "include_view_image_tool",
+        );
+        set_if_some(
+            features,
+            Feature::ViewImageTool,
+            self.tools_view_image,
+            "tools.view_image",
+        );
+    }
+}
+
+fn set_if_some(
+    features: &mut Features,
+    feature: Feature,
+    maybe_value: Option<bool>,
+    alias_key: &'static str,
+) {
+    if let Some(enabled) = maybe_value {
+        set_feature(features, feature, enabled);
+        log_alias(alias_key, feature);
+    }
+}
+
+fn set_feature(features: &mut Features, feature: Feature, enabled: bool) {
+    if enabled {
+        features.enable(feature);
+    } else {
+        features.disable(feature);
+    }
+}
+
+fn log_alias(alias: &str, feature: Feature) {
+    let canonical = feature.key();
+    if alias == canonical {
+        return;
+    }
+    info!(
+        %alias,
+        canonical,
+        "legacy feature toggle detected; prefer `[features].{canonical}`"
+    );
+}

codex-rs/core/src/function_tool.rs

@@ -4,4 +4,8 @@ use thiserror::Error;
 pub enum FunctionCallError {
     #[error("{0}")]
     RespondToModel(String),
+    #[error("LocalShellCall without call_id or id")]
+    MissingLocalShellCallId,
+    #[error("Fatal error: {0}")]
+    Fatal(String),
 }

codex-rs/core/src/lib.rs

@@ -18,6 +18,7 @@ pub use codex_conversation::CodexConversation;
 mod command_safety;
 pub mod config;
 pub mod config_edit;
+pub mod config_loader;
 pub mod config_profile;
 pub mod config_types;
 mod conversation_history;
@@ -28,9 +29,11 @@ pub mod exec;
 mod exec_command;
 pub mod exec_env;
 pub mod executor;
+pub mod features;
 mod flags;
 pub mod git_info;
 pub mod landlock;
+pub mod mcp;
 mod mcp_connection_manager;
 mod mcp_tool_call;
 mod message_history;
@@ -57,7 +60,6 @@ pub mod default_client;
 pub mod model_family;
 mod openai_model_info;
 mod openai_tools;
-pub mod plan_tool;
 pub mod project_doc;
 mod rollout;
 pub(crate) mod safety;
@@ -65,9 +67,10 @@ pub mod seatbelt;
 pub mod shell;
 pub mod spawn;
 pub mod terminal;
-mod tool_apply_patch;
+mod tools;
 pub mod turn_diff_tracker;
 pub use rollout::ARCHIVED_SESSIONS_SUBDIR;
+pub use rollout::INTERACTIVE_SESSION_SOURCES;
 pub use rollout::RolloutRecorder;
 pub use rollout::SESSIONS_SUBDIR;
 pub use rollout::SessionMeta;

codex-rs/core/src/mcp/auth.rs

@@ -0,0 +1,58 @@
+use std::collections::HashMap;
+
+use anyhow::Result;
+use codex_protocol::protocol::McpAuthStatus;
+use codex_rmcp_client::OAuthCredentialsStoreMode;
+use codex_rmcp_client::determine_streamable_http_auth_status;
+use futures::future::join_all;
+use tracing::warn;
+
+use crate::config_types::McpServerConfig;
+use crate::config_types::McpServerTransportConfig;
+
+pub async fn compute_auth_statuses<'a, I>(
+    servers: I,
+    store_mode: OAuthCredentialsStoreMode,
+) -> HashMap<String, McpAuthStatus>
+where
+    I: IntoIterator<Item = (&'a String, &'a McpServerConfig)>,
+{
+    let futures = servers.into_iter().map(|(name, config)| {
+        let name = name.clone();
+        let config = config.clone();
+        async move {
+            let status = match compute_auth_status(&name, &config, store_mode).await {
+                Ok(status) => status,
+                Err(error) => {
+                    warn!("failed to determine auth status for MCP server `{name}`: {error:?}");
+                    McpAuthStatus::Unsupported
+                }
+            };
+            (name, status)
+        }
+    });
+
+    join_all(futures).await.into_iter().collect()
+}
+
+async fn compute_auth_status(
+    server_name: &str,
+    config: &McpServerConfig,
+    store_mode: OAuthCredentialsStoreMode,
+) -> Result<McpAuthStatus> {
+    match &config.transport {
+        McpServerTransportConfig::Stdio { .. } => Ok(McpAuthStatus::Unsupported),
+        McpServerTransportConfig::StreamableHttp {
+            url,
+            bearer_token_env_var,
+        } => {
+            determine_streamable_http_auth_status(
+                server_name,
+                url,
+                bearer_token_env_var.as_deref(),
+                store_mode,
+            )
+            .await
+        }
+    }
+}

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

@@ -0,0 +1 @@
+pub mod auth;

codex-rs/core/src/mcp_connection_manager.rs

@@ -8,6 +8,7 @@
 
 use std::collections::HashMap;
 use std::collections::HashSet;
+use std::env;
 use std::ffi::OsString;
 use std::sync::Arc;
 use std::time::Duration;
@@ -16,6 +17,7 @@ use anyhow::Context;
 use anyhow::Result;
 use anyhow::anyhow;
 use codex_mcp_client::McpClient;
+use codex_rmcp_client::OAuthCredentialsStoreMode;
 use codex_rmcp_client::RmcpClient;
 use mcp_types::ClientCapabilities;
 use mcp_types::Implementation;
@@ -108,9 +110,6 @@ impl McpClientAdapter {
         params: mcp_types::InitializeRequestParams,
         startup_timeout: Duration,
     ) -> Result<Self> {
-        info!(
-            "new_stdio_client use_rmcp_client: {use_rmcp_client} program: {program:?} args: {args:?} env: {env:?} params: {params:?} startup_timeout: {startup_timeout:?}"
-        );
         if use_rmcp_client {
             let client = Arc::new(RmcpClient::new_stdio_client(program, args, env).await?);
             client.initialize(params, Some(startup_timeout)).await?;
@@ -123,12 +122,17 @@ impl McpClientAdapter {
     }
 
     async fn new_streamable_http_client(
+        server_name: String,
         url: String,
         bearer_token: Option<String>,
         params: mcp_types::InitializeRequestParams,
         startup_timeout: Duration,
+        store_mode: OAuthCredentialsStoreMode,
     ) -> Result<Self> {
-        let client = Arc::new(RmcpClient::new_streamable_http_client(url, bearer_token)?);
+        let client = Arc::new(
+            RmcpClient::new_streamable_http_client(&server_name, &url, bearer_token, store_mode)
+                .await?,
+        );
         client.initialize(params, Some(startup_timeout)).await?;
         Ok(McpClientAdapter::Rmcp(client))
     }
@@ -182,6 +186,7 @@ impl McpConnectionManager {
     pub async fn new(
         mcp_servers: HashMap<String, McpServerConfig>,
         use_rmcp_client: bool,
+        store_mode: OAuthCredentialsStoreMode,
     ) -> Result<(Self, ClientStartErrors)> {
         // Early exit if no servers are configured.
         if mcp_servers.is_empty() {
@@ -202,22 +207,21 @@ impl McpConnectionManager {
                 continue;
             }
 
-            if matches!(
-                cfg.transport,
-                McpServerTransportConfig::StreamableHttp { .. }
-            ) && !use_rmcp_client
-            {
-                info!(
-                    "skipping MCP server `{}` configured with url because rmcp client is disabled",
-                    server_name
-                );
+            if !cfg.enabled {
                 continue;
             }
 
             let startup_timeout = cfg.startup_timeout_sec.unwrap_or(DEFAULT_STARTUP_TIMEOUT);
             let tool_timeout = cfg.tool_timeout_sec.unwrap_or(DEFAULT_TOOL_TIMEOUT);
 
-            let use_rmcp_client_flag = use_rmcp_client;
+            let resolved_bearer_token = match &cfg.transport {
+                McpServerTransportConfig::StreamableHttp {
+                    bearer_token_env_var,
+                    ..
+                } => resolve_bearer_token(&server_name, bearer_token_env_var.as_deref()),
+                _ => Ok(None),
+            };
+
             join_set.spawn(async move {
                 let McpServerConfig { transport, .. } = cfg;
                 let params = mcp_types::InitializeRequestParams {
@@ -246,21 +250,23 @@ impl McpConnectionManager {
                         let command_os: OsString = command.into();
                         let args_os: Vec<OsString> = args.into_iter().map(Into::into).collect();
                         McpClientAdapter::new_stdio_client(
-                            use_rmcp_client_flag,
+                            use_rmcp_client,
                             command_os,
                             args_os,
                             env,
-                            params.clone(),
+                            params,
                             startup_timeout,
                         )
                         .await
                     }
-                    McpServerTransportConfig::StreamableHttp { url, bearer_token } => {
+                    McpServerTransportConfig::StreamableHttp { url, .. } => {
                         McpClientAdapter::new_streamable_http_client(
+                            server_name.clone(),
                             url,
-                            bearer_token,
+                            resolved_bearer_token.unwrap_or_default(),
                             params,
                             startup_timeout,
+                            store_mode,
                         )
                         .await
                     }
@@ -348,6 +354,33 @@ impl McpConnectionManager {
     }
 }
 
+fn resolve_bearer_token(
+    server_name: &str,
+    bearer_token_env_var: Option<&str>,
+) -> Result<Option<String>> {
+    let Some(env_var) = bearer_token_env_var else {
+        return Ok(None);
+    };
+
+    match env::var(env_var) {
+        Ok(value) => {
+            if value.is_empty() {
+                Err(anyhow!(
+                    "Environment variable {env_var} for MCP server '{server_name}' is empty"
+                ))
+            } else {
+                Ok(Some(value))
+            }
+        }
+        Err(env::VarError::NotPresent) => Err(anyhow!(
+            "Environment variable {env_var} for MCP server '{server_name}' is not set"
+        )),
+        Err(env::VarError::NotUnicode(_)) => Err(anyhow!(
+            "Environment variable {env_var} for MCP server '{server_name}' contains invalid Unicode"
+        )),
+    }
+}
+
 /// Query every server for its available tools and return a single map that
 /// contains **all** tools. Each key is the fully-qualified name for the tool.
 async fn list_all_tools(clients: &HashMap<String, ManagedClient>) -> Result<Vec<ToolInfo>> {

codex-rs/core/src/model_family.rs

@@ -1,5 +1,5 @@
 use crate::config_types::ReasoningSummaryFormat;
-use crate::tool_apply_patch::ApplyPatchToolType;
+use crate::tools::handlers::apply_patch::ApplyPatchToolType;
 
 /// The `instructions` field in the payload sent to a model should always start
 /// with this content.
@@ -35,12 +35,19 @@ pub struct ModelFamily {
     // See https://platform.openai.com/docs/guides/tools-local-shell
     pub uses_local_shell_tool: bool,
 
+    /// Whether this model supports parallel tool calls when using the
+    /// Responses API.
+    pub supports_parallel_tool_calls: bool,
+
     /// Present if the model performs better when `apply_patch` is provided as
     /// a tool call instead of just a bash command
     pub apply_patch_tool_type: Option<ApplyPatchToolType>,
 
     // Instructions to use for querying the model
     pub base_instructions: String,
+
+    /// Names of beta tools that should be exposed to this model family.
+    pub experimental_supported_tools: Vec<String>,
 }
 
 macro_rules! model_family {
@@ -55,8 +62,10 @@ macro_rules! model_family {
             supports_reasoning_summaries: false,
             reasoning_summary_format: ReasoningSummaryFormat::None,
             uses_local_shell_tool: false,
+            supports_parallel_tool_calls: false,
             apply_patch_tool_type: None,
             base_instructions: BASE_INSTRUCTIONS.to_string(),
+            experimental_supported_tools: Vec::new(),
         };
         // apply overrides
         $(
@@ -68,7 +77,11 @@ macro_rules! model_family {
 
 /// Returns a `ModelFamily` for the given model slug, or `None` if the slug
 /// does not match any known model family.
-pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
+pub fn find_family_for_model(mut slug: &str) -> Option<ModelFamily> {
+    // TODO(jif) clean once we have proper feature flags
+    if matches!(std::env::var("CODEX_EXPERIMENTAL").as_deref(), Ok("1")) {
+        slug = "codex-experimental";
+    }
     if slug.starts_with("o3") {
         model_family!(
             slug, "o3",
@@ -99,12 +112,45 @@ pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
         model_family!(slug, "gpt-4o", needs_special_apply_patch_instructions: true)
     } else if slug.starts_with("gpt-3.5") {
         model_family!(slug, "gpt-3.5", needs_special_apply_patch_instructions: true)
-    } else if slug.starts_with("codex-") || slug.starts_with("gpt-5-codex") {
+    } else if slug.starts_with("test-gpt-5-codex") {
         model_family!(
             slug, slug,
             supports_reasoning_summaries: true,
             reasoning_summary_format: ReasoningSummaryFormat::Experimental,
             base_instructions: GPT_5_CODEX_INSTRUCTIONS.to_string(),
+            experimental_supported_tools: vec![
+                "grep_files".to_string(),
+                "list_dir".to_string(),
+                "read_file".to_string(),
+                "test_sync_tool".to_string(),
+            ],
+            supports_parallel_tool_calls: true,
+        )
+
+    // Internal models.
+    } else if slug.starts_with("codex-") {
+        model_family!(
+            slug, slug,
+            supports_reasoning_summaries: true,
+            reasoning_summary_format: ReasoningSummaryFormat::Experimental,
+            base_instructions: GPT_5_CODEX_INSTRUCTIONS.to_string(),
+            apply_patch_tool_type: Some(ApplyPatchToolType::Freeform),
+            experimental_supported_tools: vec![
+                "grep_files".to_string(),
+                "list_dir".to_string(),
+                "read_file".to_string(),
+            ],
+            supports_parallel_tool_calls: true,
+        )
+
+    // Production models.
+    } else if slug.starts_with("gpt-5-codex") {
+        model_family!(
+            slug, slug,
+            supports_reasoning_summaries: true,
+            reasoning_summary_format: ReasoningSummaryFormat::Experimental,
+            base_instructions: GPT_5_CODEX_INSTRUCTIONS.to_string(),
+            apply_patch_tool_type: Some(ApplyPatchToolType::Freeform),
         )
     } else if slug.starts_with("gpt-5") {
         model_family!(
@@ -125,7 +171,9 @@ pub fn derive_default_model_family(model: &str) -> ModelFamily {
         supports_reasoning_summaries: false,
         reasoning_summary_format: ReasoningSummaryFormat::None,
         uses_local_shell_tool: false,
+        supports_parallel_tool_calls: false,
         apply_patch_tool_type: None,
         base_instructions: BASE_INSTRUCTIONS.to_string(),
+        experimental_supported_tools: Vec::new(),
     }
 }

codex-rs/core/src/project_doc.rs

@@ -21,6 +21,8 @@ use tracing::error;
 
 /// Default filename scanned for project-level docs.
 pub const DEFAULT_PROJECT_DOC_FILENAME: &str = "AGENTS.md";
+/// Preferred local override for project-level docs.
+pub const LOCAL_PROJECT_DOC_FILENAME: &str = "AGENTS.override.md";
 
 /// When both `Config::instructions` and the project doc are present, they will
 /// be concatenated with the following separator.
@@ -178,7 +180,8 @@ pub fn discover_project_doc_paths(config: &Config) -> std::io::Result<Vec<PathBu
 
 fn candidate_filenames<'a>(config: &'a Config) -> Vec<&'a str> {
     let mut names: Vec<&'a str> =
-        Vec::with_capacity(1 + config.project_doc_fallback_filenames.len());
+        Vec::with_capacity(2 + config.project_doc_fallback_filenames.len());
+    names.push(LOCAL_PROJECT_DOC_FILENAME);
     names.push(DEFAULT_PROJECT_DOC_FILENAME);
     for candidate in &config.project_doc_fallback_filenames {
         let candidate = candidate.as_str();
@@ -381,6 +384,29 @@ mod tests {
         assert_eq!(res, "root doc\n\ncrate doc");
     }
 
+    /// AGENTS.override.md is preferred over AGENTS.md when both are present.
+    #[tokio::test]
+    async fn agents_local_md_preferred() {
+        let tmp = tempfile::tempdir().expect("tempdir");
+        fs::write(tmp.path().join(DEFAULT_PROJECT_DOC_FILENAME), "versioned").unwrap();
+        fs::write(tmp.path().join(LOCAL_PROJECT_DOC_FILENAME), "local").unwrap();
+
+        let cfg = make_config(&tmp, 4096, None);
+
+        let res = get_user_instructions(&cfg)
+            .await
+            .expect("local doc expected");
+
+        assert_eq!(res, "local");
+
+        let discovery = discover_project_doc_paths(&cfg).expect("discover paths");
+        assert_eq!(discovery.len(), 1);
+        assert_eq!(
+            discovery[0].file_name().unwrap().to_string_lossy(),
+            LOCAL_PROJECT_DOC_FILENAME
+        );
+    }
+
     /// When AGENTS.md is absent but a configured fallback exists, the fallback is used.
     #[tokio::test]
     async fn uses_configured_fallback_when_agents_missing() {

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

@@ -17,6 +17,7 @@ use super::SESSIONS_SUBDIR;
 use crate::protocol::EventMsg;
 use codex_protocol::protocol::RolloutItem;
 use codex_protocol::protocol::RolloutLine;
+use codex_protocol::protocol::SessionSource;
 
 /// Returned page of conversation summaries.
 #[derive(Debug, Default, PartialEq)]
@@ -52,6 +53,7 @@ struct HeadTailSummary {
     tail: Vec<serde_json::Value>,
     saw_session_meta: bool,
     saw_user_event: bool,
+    source: Option<SessionSource>,
     created_at: Option<String>,
     updated_at: Option<String>,
 }
@@ -106,6 +108,7 @@ pub(crate) async fn get_conversations(
     codex_home: &Path,
     page_size: usize,
     cursor: Option<&Cursor>,
+    allowed_sources: &[SessionSource],
 ) -> io::Result<ConversationsPage> {
     let mut root = codex_home.to_path_buf();
     root.push(SESSIONS_SUBDIR);
@@ -121,7 +124,8 @@ pub(crate) async fn get_conversations(
 
     let anchor = cursor.cloned();
 
-    let result = traverse_directories_for_paths(root.clone(), page_size, anchor).await?;
+    let result =
+        traverse_directories_for_paths(root.clone(), page_size, anchor, allowed_sources).await?;
     Ok(result)
 }
 
@@ -140,6 +144,7 @@ async fn traverse_directories_for_paths(
     root: PathBuf,
     page_size: usize,
     anchor: Option<Cursor>,
+    allowed_sources: &[SessionSource],
 ) -> io::Result<ConversationsPage> {
     let mut items: Vec<ConversationItem> = Vec::with_capacity(page_size);
     let mut scanned_files = 0usize;
@@ -196,6 +201,13 @@ async fn traverse_directories_for_paths(
                     let summary = read_head_and_tail(&path, HEAD_RECORD_LIMIT, TAIL_RECORD_LIMIT)
                         .await
                         .unwrap_or_default();
+                    if !allowed_sources.is_empty()
+                        && !summary
+                            .source
+                            .is_some_and(|source| allowed_sources.iter().any(|s| s == &source))
+                    {
+                        continue;
+                    }
                     // Apply filters: must have session meta and at least one user message event
                     if summary.saw_session_meta && summary.saw_user_event {
                         let HeadTailSummary {
@@ -341,6 +353,7 @@ async fn read_head_and_tail(
 
         match rollout_line.item {
             RolloutItem::SessionMeta(session_meta_line) => {
+                summary.source = Some(session_meta_line.meta.source);
                 summary.created_at = summary
                     .created_at
                     .clone()

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

@@ -1,7 +1,11 @@
 //! Rollout module: persistence and discovery of session rollout files.
 
+use codex_protocol::protocol::SessionSource;
+
 pub const SESSIONS_SUBDIR: &str = "sessions";
 pub const ARCHIVED_SESSIONS_SUBDIR: &str = "archived_sessions";
+pub const INTERACTIVE_SESSION_SOURCES: &[SessionSource] =
+    &[SessionSource::Cli, SessionSource::VSCode];
 
 pub mod list;
 pub(crate) mod policy;

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

@@ -70,6 +70,7 @@ pub(crate) fn should_persist_event_msg(ev: &EventMsg) -> bool {
         | EventMsg::ListCustomPromptsResponse(_)
         | EventMsg::PlanUpdate(_)
         | EventMsg::ShutdownComplete
+        | EventMsg::ViewImageToolCall(_)
         | EventMsg::ConversationPath(_) => false,
     }
 }

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

@@ -32,6 +32,7 @@ use codex_protocol::protocol::RolloutItem;
 use codex_protocol::protocol::RolloutLine;
 use codex_protocol::protocol::SessionMeta;
 use codex_protocol::protocol::SessionMetaLine;
+use codex_protocol::protocol::SessionSource;
 
 /// Records all [`ResponseItem`]s for a session and flushes them to disk after
 /// every update.
@@ -53,6 +54,7 @@ pub enum RolloutRecorderParams {
     Create {
         conversation_id: ConversationId,
         instructions: Option<String>,
+        source: SessionSource,
     },
     Resume {
         path: PathBuf,
@@ -71,10 +73,15 @@ enum RolloutCmd {
 }
 
 impl RolloutRecorderParams {
-    pub fn new(conversation_id: ConversationId, instructions: Option<String>) -> Self {
+    pub fn new(
+        conversation_id: ConversationId,
+        instructions: Option<String>,
+        source: SessionSource,
+    ) -> Self {
         Self::Create {
             conversation_id,
             instructions,
+            source,
         }
     }
 
@@ -89,8 +96,9 @@ impl RolloutRecorder {
         codex_home: &Path,
         page_size: usize,
         cursor: Option<&Cursor>,
+        allowed_sources: &[SessionSource],
     ) -> std::io::Result<ConversationsPage> {
-        get_conversations(codex_home, page_size, cursor).await
+        get_conversations(codex_home, page_size, cursor, allowed_sources).await
     }
 
     /// Attempt to create a new [`RolloutRecorder`]. If the sessions directory
@@ -101,6 +109,7 @@ impl RolloutRecorder {
             RolloutRecorderParams::Create {
                 conversation_id,
                 instructions,
+                source,
             } => {
                 let LogFileInfo {
                     file,
@@ -127,6 +136,7 @@ impl RolloutRecorder {
                         originator: originator().value.clone(),
                         cli_version: env!("CARGO_PKG_VERSION").to_string(),
                         instructions,
+                        source,
                     }),
                 )
             }

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

@@ -12,6 +12,7 @@ use time::format_description::FormatItem;
 use time::macros::format_description;
 use uuid::Uuid;
 
+use crate::rollout::INTERACTIVE_SESSION_SOURCES;
 use crate::rollout::list::ConversationItem;
 use crate::rollout::list::ConversationsPage;
 use crate::rollout::list::Cursor;
@@ -28,13 +29,17 @@ use codex_protocol::protocol::RolloutItem;
 use codex_protocol::protocol::RolloutLine;
 use codex_protocol::protocol::SessionMeta;
 use codex_protocol::protocol::SessionMetaLine;
+use codex_protocol::protocol::SessionSource;
 use codex_protocol::protocol::UserMessageEvent;
 
+const NO_SOURCE_FILTER: &[SessionSource] = &[];
+
 fn write_session_file(
     root: &Path,
     ts_str: &str,
     uuid: Uuid,
     num_records: usize,
+    source: Option<SessionSource>,
 ) -> std::io::Result<(OffsetDateTime, Uuid)> {
     let format: &[FormatItem] =
         format_description!("[year]-[month]-[day]T[hour]-[minute]-[second]");
@@ -52,17 +57,23 @@ fn write_session_file(
     let file_path = dir.join(filename);
     let mut file = File::create(file_path)?;
 
+    let mut payload = serde_json::json!({
+        "id": uuid,
+        "timestamp": ts_str,
+        "instructions": null,
+        "cwd": ".",
+        "originator": "test_originator",
+        "cli_version": "test_version",
+    });
+
+    if let Some(source) = source {
+        payload["source"] = serde_json::to_value(source).unwrap();
+    }
+
     let meta = serde_json::json!({
         "timestamp": ts_str,
         "type": "session_meta",
-        "payload": {
-            "id": uuid,
-            "timestamp": ts_str,
-            "instructions": null,
-            "cwd": ".",
-            "originator": "test_originator",
-            "cli_version": "test_version"
-        }
+        "payload": payload,
     });
     writeln!(file, "{meta}")?;
 
@@ -99,11 +110,34 @@ async fn test_list_conversations_latest_first() {
     let u3 = Uuid::from_u128(3);
 
     // Create three sessions across three days
-    write_session_file(home, "2025-01-01T12-00-00", u1, 3).unwrap();
-    write_session_file(home, "2025-01-02T12-00-00", u2, 3).unwrap();
-    write_session_file(home, "2025-01-03T12-00-00", u3, 3).unwrap();
+    write_session_file(
+        home,
+        "2025-01-01T12-00-00",
+        u1,
+        3,
+        Some(SessionSource::VSCode),
+    )
+    .unwrap();
+    write_session_file(
+        home,
+        "2025-01-02T12-00-00",
+        u2,
+        3,
+        Some(SessionSource::VSCode),
+    )
+    .unwrap();
+    write_session_file(
+        home,
+        "2025-01-03T12-00-00",
+        u3,
+        3,
+        Some(SessionSource::VSCode),
+    )
+    .unwrap();
 
-    let page = get_conversations(home, 10, None).await.unwrap();
+    let page = get_conversations(home, 10, None, INTERACTIVE_SESSION_SOURCES)
+        .await
+        .unwrap();
 
     // Build expected objects
     let p1 = home
@@ -131,7 +165,8 @@ async fn test_list_conversations_latest_first() {
         "instructions": null,
         "cwd": ".",
         "originator": "test_originator",
-        "cli_version": "test_version"
+        "cli_version": "test_version",
+        "source": "vscode",
     })];
     let head_2 = vec![serde_json::json!({
         "id": u2,
@@ -139,7 +174,8 @@ async fn test_list_conversations_latest_first() {
         "instructions": null,
         "cwd": ".",
         "originator": "test_originator",
-        "cli_version": "test_version"
+        "cli_version": "test_version",
+        "source": "vscode",
     })];
     let head_1 = vec![serde_json::json!({
         "id": u1,
@@ -147,7 +183,8 @@ async fn test_list_conversations_latest_first() {
         "instructions": null,
         "cwd": ".",
         "originator": "test_originator",
-        "cli_version": "test_version"
+        "cli_version": "test_version",
+        "source": "vscode",
     })];
 
     let expected_cursor: Cursor =
@@ -198,13 +235,50 @@ async fn test_pagination_cursor() {
     let u5 = Uuid::from_u128(55);
 
     // Oldest to newest
-    write_session_file(home, "2025-03-01T09-00-00", u1, 1).unwrap();
-    write_session_file(home, "2025-03-02T09-00-00", u2, 1).unwrap();
-    write_session_file(home, "2025-03-03T09-00-00", u3, 1).unwrap();
-    write_session_file(home, "2025-03-04T09-00-00", u4, 1).unwrap();
-    write_session_file(home, "2025-03-05T09-00-00", u5, 1).unwrap();
+    write_session_file(
+        home,
+        "2025-03-01T09-00-00",
+        u1,
+        1,
+        Some(SessionSource::VSCode),
+    )
+    .unwrap();
+    write_session_file(
+        home,
+        "2025-03-02T09-00-00",
+        u2,
+        1,
+        Some(SessionSource::VSCode),
+    )
+    .unwrap();
+    write_session_file(
+        home,
+        "2025-03-03T09-00-00",
+        u3,
+        1,
+        Some(SessionSource::VSCode),
+    )
+    .unwrap();
+    write_session_file(
+        home,
+        "2025-03-04T09-00-00",
+        u4,
+        1,
+        Some(SessionSource::VSCode),
+    )
+    .unwrap();
+    write_session_file(
+        home,
+        "2025-03-05T09-00-00",
+        u5,
+        1,
+        Some(SessionSource::VSCode),
+    )
+    .unwrap();
 
-    let page1 = get_conversations(home, 2, None).await.unwrap();
+    let page1 = get_conversations(home, 2, None, INTERACTIVE_SESSION_SOURCES)
+        .await
+        .unwrap();
     let p5 = home
         .join("sessions")
         .join("2025")
@@ -223,7 +297,8 @@ async fn test_pagination_cursor() {
         "instructions": null,
         "cwd": ".",
         "originator": "test_originator",
-        "cli_version": "test_version"
+        "cli_version": "test_version",
+        "source": "vscode",
     })];
     let head_4 = vec![serde_json::json!({
         "id": u4,
@@ -231,7 +306,8 @@ async fn test_pagination_cursor() {
         "instructions": null,
         "cwd": ".",
         "originator": "test_originator",
-        "cli_version": "test_version"
+        "cli_version": "test_version",
+        "source": "vscode",
     })];
     let expected_cursor1: Cursor =
         serde_json::from_str(&format!("\"2025-03-04T09-00-00|{u4}\"")).unwrap();
@@ -258,9 +334,14 @@ async fn test_pagination_cursor() {
     };
     assert_eq!(page1, expected_page1);
 
-    let page2 = get_conversations(home, 2, page1.next_cursor.as_ref())
-        .await
-        .unwrap();
+    let page2 = get_conversations(
+        home,
+        2,
+        page1.next_cursor.as_ref(),
+        INTERACTIVE_SESSION_SOURCES,
+    )
+    .await
+    .unwrap();
     let p3 = home
         .join("sessions")
         .join("2025")
@@ -279,7 +360,8 @@ async fn test_pagination_cursor() {
         "instructions": null,
         "cwd": ".",
         "originator": "test_originator",
-        "cli_version": "test_version"
+        "cli_version": "test_version",
+        "source": "vscode",
     })];
     let head_2 = vec![serde_json::json!({
         "id": u2,
@@ -287,7 +369,8 @@ async fn test_pagination_cursor() {
         "instructions": null,
         "cwd": ".",
         "originator": "test_originator",
-        "cli_version": "test_version"
+        "cli_version": "test_version",
+        "source": "vscode",
     })];
     let expected_cursor2: Cursor =
         serde_json::from_str(&format!("\"2025-03-02T09-00-00|{u2}\"")).unwrap();
@@ -314,9 +397,14 @@ async fn test_pagination_cursor() {
     };
     assert_eq!(page2, expected_page2);
 
-    let page3 = get_conversations(home, 2, page2.next_cursor.as_ref())
-        .await
-        .unwrap();
+    let page3 = get_conversations(
+        home,
+        2,
+        page2.next_cursor.as_ref(),
+        INTERACTIVE_SESSION_SOURCES,
+    )
+    .await
+    .unwrap();
     let p1 = home
         .join("sessions")
         .join("2025")
@@ -329,7 +417,8 @@ async fn test_pagination_cursor() {
         "instructions": null,
         "cwd": ".",
         "originator": "test_originator",
-        "cli_version": "test_version"
+        "cli_version": "test_version",
+        "source": "vscode",
     })];
     let expected_cursor3: Cursor =
         serde_json::from_str(&format!("\"2025-03-01T09-00-00|{u1}\"")).unwrap();
@@ -355,9 +444,11 @@ async fn test_get_conversation_contents() {
 
     let uuid = Uuid::new_v4();
     let ts = "2025-04-01T10-30-00";
-    write_session_file(home, ts, uuid, 2).unwrap();
+    write_session_file(home, ts, uuid, 2, Some(SessionSource::VSCode)).unwrap();
 
-    let page = get_conversations(home, 1, None).await.unwrap();
+    let page = get_conversations(home, 1, None, INTERACTIVE_SESSION_SOURCES)
+        .await
+        .unwrap();
     let path = &page.items[0].path;
 
     let content = get_conversation(path).await.unwrap();
@@ -375,7 +466,8 @@ async fn test_get_conversation_contents() {
         "instructions": null,
         "cwd": ".",
         "originator": "test_originator",
-        "cli_version": "test_version"
+        "cli_version": "test_version",
+        "source": "vscode",
     })];
     let expected_cursor: Cursor = serde_json::from_str(&format!("\"{ts}|{uuid}\"")).unwrap();
     let expected_page = ConversationsPage {
@@ -393,7 +485,19 @@ async fn test_get_conversation_contents() {
     assert_eq!(page, expected_page);
 
     // Entire file contents equality
-    let meta = serde_json::json!({"timestamp": ts, "type": "session_meta", "payload": {"id": uuid, "timestamp": ts, "instructions": null, "cwd": ".", "originator": "test_originator", "cli_version": "test_version"}});
+    let meta = serde_json::json!({
+        "timestamp": ts,
+        "type": "session_meta",
+        "payload": {
+            "id": uuid,
+            "timestamp": ts,
+            "instructions": null,
+            "cwd": ".",
+            "originator": "test_originator",
+            "cli_version": "test_version",
+            "source": "vscode",
+        }
+    });
     let user_event = serde_json::json!({
         "timestamp": ts,
         "type": "event_msg",
@@ -428,6 +532,7 @@ async fn test_tail_includes_last_response_items() -> Result<()> {
                 cwd: ".".into(),
                 originator: "test_originator".into(),
                 cli_version: "test_version".into(),
+                source: SessionSource::VSCode,
             },
             git: None,
         }),
@@ -460,7 +565,7 @@ async fn test_tail_includes_last_response_items() -> Result<()> {
     }
     drop(file);
 
-    let page = get_conversations(home, 1, None).await?;
+    let page = get_conversations(home, 1, None, INTERACTIVE_SESSION_SOURCES).await?;
     let item = page.items.first().expect("conversation item");
     let tail_len = item.tail.len();
     assert_eq!(tail_len, 10usize.min(total_messages));
@@ -511,6 +616,7 @@ async fn test_tail_handles_short_sessions() -> Result<()> {
                 cwd: ".".into(),
                 originator: "test_originator".into(),
                 cli_version: "test_version".into(),
+                source: SessionSource::VSCode,
             },
             git: None,
         }),
@@ -542,7 +648,7 @@ async fn test_tail_handles_short_sessions() -> Result<()> {
     }
     drop(file);
 
-    let page = get_conversations(home, 1, None).await?;
+    let page = get_conversations(home, 1, None, INTERACTIVE_SESSION_SOURCES).await?;
     let tail = &page.items.first().expect("conversation item").tail;
 
     assert_eq!(tail.len(), 3);
@@ -595,6 +701,7 @@ async fn test_tail_skips_trailing_non_responses() -> Result<()> {
                 cwd: ".".into(),
                 originator: "test_originator".into(),
                 cli_version: "test_version".into(),
+                source: SessionSource::VSCode,
             },
             git: None,
         }),
@@ -640,7 +747,7 @@ async fn test_tail_skips_trailing_non_responses() -> Result<()> {
     writeln!(file, "{}", serde_json::to_string(&shutdown_event)?)?;
     drop(file);
 
-    let page = get_conversations(home, 1, None).await?;
+    let page = get_conversations(home, 1, None, INTERACTIVE_SESSION_SOURCES).await?;
     let tail = &page.items.first().expect("conversation item").tail;
 
     let expected: Vec<serde_json::Value> = (0..4)
@@ -678,11 +785,13 @@ async fn test_stable_ordering_same_second_pagination() {
     let u2 = Uuid::from_u128(2);
     let u3 = Uuid::from_u128(3);
 
-    write_session_file(home, ts, u1, 0).unwrap();
-    write_session_file(home, ts, u2, 0).unwrap();
-    write_session_file(home, ts, u3, 0).unwrap();
+    write_session_file(home, ts, u1, 0, Some(SessionSource::VSCode)).unwrap();
+    write_session_file(home, ts, u2, 0, Some(SessionSource::VSCode)).unwrap();
+    write_session_file(home, ts, u3, 0, Some(SessionSource::VSCode)).unwrap();
 
-    let page1 = get_conversations(home, 2, None).await.unwrap();
+    let page1 = get_conversations(home, 2, None, INTERACTIVE_SESSION_SOURCES)
+        .await
+        .unwrap();
 
     let p3 = home
         .join("sessions")
@@ -703,7 +812,8 @@ async fn test_stable_ordering_same_second_pagination() {
             "instructions": null,
             "cwd": ".",
             "originator": "test_originator",
-            "cli_version": "test_version"
+            "cli_version": "test_version",
+            "source": "vscode",
         })]
     };
     let expected_cursor1: Cursor = serde_json::from_str(&format!("\"{ts}|{u2}\"")).unwrap();
@@ -730,9 +840,14 @@ async fn test_stable_ordering_same_second_pagination() {
     };
     assert_eq!(page1, expected_page1);
 
-    let page2 = get_conversations(home, 2, page1.next_cursor.as_ref())
-        .await
-        .unwrap();
+    let page2 = get_conversations(
+        home,
+        2,
+        page1.next_cursor.as_ref(),
+        INTERACTIVE_SESSION_SOURCES,
+    )
+    .await
+    .unwrap();
     let p1 = home
         .join("sessions")
         .join("2025")
@@ -754,3 +869,59 @@ async fn test_stable_ordering_same_second_pagination() {
     };
     assert_eq!(page2, expected_page2);
 }
+
+#[tokio::test]
+async fn test_source_filter_excludes_non_matching_sessions() {
+    let temp = TempDir::new().unwrap();
+    let home = temp.path();
+
+    let interactive_id = Uuid::from_u128(42);
+    let non_interactive_id = Uuid::from_u128(77);
+
+    write_session_file(
+        home,
+        "2025-08-02T10-00-00",
+        interactive_id,
+        2,
+        Some(SessionSource::Cli),
+    )
+    .unwrap();
+    write_session_file(
+        home,
+        "2025-08-01T10-00-00",
+        non_interactive_id,
+        2,
+        Some(SessionSource::Exec),
+    )
+    .unwrap();
+
+    let interactive_only = get_conversations(home, 10, None, INTERACTIVE_SESSION_SOURCES)
+        .await
+        .unwrap();
+    let paths: Vec<_> = interactive_only
+        .items
+        .iter()
+        .map(|item| item.path.as_path())
+        .collect();
+
+    assert_eq!(paths.len(), 1);
+    assert!(paths.iter().all(|path| {
+        path.ends_with("rollout-2025-08-02T10-00-00-00000000-0000-0000-0000-00000000002a.jsonl")
+    }));
+
+    let all_sessions = get_conversations(home, 10, None, NO_SOURCE_FILTER)
+        .await
+        .unwrap();
+    let all_paths: Vec<_> = all_sessions
+        .items
+        .into_iter()
+        .map(|item| item.path)
+        .collect();
+    assert_eq!(all_paths.len(), 2);
+    assert!(all_paths.iter().any(|path| {
+        path.ends_with("rollout-2025-08-02T10-00-00-00000000-0000-0000-0000-00000000002a.jsonl")
+    }));
+    assert!(all_paths.iter().any(|path| {
+        path.ends_with("rollout-2025-08-01T10-00-00-00000000-0000-0000-0000-00000000004d.jsonl")
+    }));
+}

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

@@ -64,5 +64,14 @@ impl SessionState {
         (self.token_info.clone(), self.latest_rate_limits.clone())
     }
 
+    pub(crate) fn set_token_usage_full(&mut self, context_window: u64) {
+        match &mut self.token_info {
+            Some(info) => info.fill_to_context_window(context_window),
+            None => {
+                self.token_info = Some(TokenUsageInfo::full_context_window(context_window));
+            }
+        }
+    }
+
     // Pending input/approval moved to TurnState.
 }

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

@@ -34,6 +34,16 @@ pub(crate) enum TaskKind {
     Compact,
 }
 
+impl TaskKind {
+    pub(crate) fn header_value(self) -> &'static str {
+        match self {
+            TaskKind::Regular => "standard",
+            TaskKind::Review => "review",
+            TaskKind::Compact => "compact",
+        }
+    }
+}
+
 #[derive(Clone)]
 pub(crate) struct RunningTask {
     pub(crate) handle: AbortHandle,
@@ -113,3 +123,15 @@ impl ActiveTurn {
         }
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::TaskKind;
+
+    #[test]
+    fn header_value_matches_expected_labels() {
+        assert_eq!(TaskKind::Regular.header_value(), "standard");
+        assert_eq!(TaskKind::Review.header_value(), "review");
+        assert_eq!(TaskKind::Compact.header_value(), "compact");
+    }
+}

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

@@ -27,6 +27,6 @@ impl SessionTask for RegularTask {
         input: Vec<InputItem>,
     ) -> Option<String> {
         let sess = session.clone_session();
-        run_task(sess, ctx, sub_id, input).await
+        run_task(sess, ctx, sub_id, input, TaskKind::Regular).await
     }
 }

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

@@ -28,7 +28,7 @@ impl SessionTask for ReviewTask {
         input: Vec<InputItem>,
     ) -> Option<String> {
         let sess = session.clone_session();
-        run_task(sess, ctx, sub_id, input).await
+        run_task(sess, ctx, sub_id, input, TaskKind::Review).await
     }
 
     async fn abort(&self, session: Arc<SessionTaskContext>, sub_id: &str) {

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

@@ -0,0 +1,249 @@
+use crate::codex::Session;
+use crate::codex::TurnContext;
+use crate::tools::TELEMETRY_PREVIEW_MAX_BYTES;
+use crate::tools::TELEMETRY_PREVIEW_MAX_LINES;
+use crate::tools::TELEMETRY_PREVIEW_TRUNCATION_NOTICE;
+use crate::turn_diff_tracker::TurnDiffTracker;
+use codex_otel::otel_event_manager::OtelEventManager;
+use codex_protocol::models::FunctionCallOutputPayload;
+use codex_protocol::models::ResponseInputItem;
+use codex_protocol::models::ShellToolCallParams;
+use codex_protocol::protocol::FileChange;
+use codex_utils_string::take_bytes_at_char_boundary;
+use mcp_types::CallToolResult;
+use std::borrow::Cow;
+use std::collections::HashMap;
+use std::path::PathBuf;
+use std::sync::Arc;
+use tokio::sync::Mutex;
+
+pub type SharedTurnDiffTracker = Arc<Mutex<TurnDiffTracker>>;
+
+#[derive(Clone)]
+pub struct ToolInvocation {
+    pub session: Arc<Session>,
+    pub turn: Arc<TurnContext>,
+    pub tracker: SharedTurnDiffTracker,
+    pub sub_id: String,
+    pub call_id: String,
+    pub tool_name: String,
+    pub payload: ToolPayload,
+}
+
+#[derive(Clone)]
+pub enum ToolPayload {
+    Function {
+        arguments: String,
+    },
+    Custom {
+        input: String,
+    },
+    LocalShell {
+        params: ShellToolCallParams,
+    },
+    UnifiedExec {
+        arguments: String,
+    },
+    Mcp {
+        server: String,
+        tool: String,
+        raw_arguments: String,
+    },
+}
+
+impl ToolPayload {
+    pub fn log_payload(&self) -> Cow<'_, str> {
+        match self {
+            ToolPayload::Function { arguments } => Cow::Borrowed(arguments),
+            ToolPayload::Custom { input } => Cow::Borrowed(input),
+            ToolPayload::LocalShell { params } => Cow::Owned(params.command.join(" ")),
+            ToolPayload::UnifiedExec { arguments } => Cow::Borrowed(arguments),
+            ToolPayload::Mcp { raw_arguments, .. } => Cow::Borrowed(raw_arguments),
+        }
+    }
+}
+
+#[derive(Clone)]
+pub enum ToolOutput {
+    Function {
+        content: String,
+        success: Option<bool>,
+    },
+    Mcp {
+        result: Result<CallToolResult, String>,
+    },
+}
+
+impl ToolOutput {
+    pub fn log_preview(&self) -> String {
+        match self {
+            ToolOutput::Function { content, .. } => telemetry_preview(content),
+            ToolOutput::Mcp { result } => format!("{result:?}"),
+        }
+    }
+
+    pub fn success_for_logging(&self) -> bool {
+        match self {
+            ToolOutput::Function { success, .. } => success.unwrap_or(true),
+            ToolOutput::Mcp { result } => result.is_ok(),
+        }
+    }
+
+    pub fn into_response(self, call_id: &str, payload: &ToolPayload) -> ResponseInputItem {
+        match self {
+            ToolOutput::Function { content, success } => {
+                if matches!(payload, ToolPayload::Custom { .. }) {
+                    ResponseInputItem::CustomToolCallOutput {
+                        call_id: call_id.to_string(),
+                        output: content,
+                    }
+                } else {
+                    ResponseInputItem::FunctionCallOutput {
+                        call_id: call_id.to_string(),
+                        output: FunctionCallOutputPayload { content, success },
+                    }
+                }
+            }
+            ToolOutput::Mcp { result } => ResponseInputItem::McpToolCallOutput {
+                call_id: call_id.to_string(),
+                result,
+            },
+        }
+    }
+}
+
+fn telemetry_preview(content: &str) -> String {
+    let truncated_slice = take_bytes_at_char_boundary(content, TELEMETRY_PREVIEW_MAX_BYTES);
+    let truncated_by_bytes = truncated_slice.len() < content.len();
+
+    let mut preview = String::new();
+    let mut lines_iter = truncated_slice.lines();
+    for idx in 0..TELEMETRY_PREVIEW_MAX_LINES {
+        match lines_iter.next() {
+            Some(line) => {
+                if idx > 0 {
+                    preview.push('\n');
+                }
+                preview.push_str(line);
+            }
+            None => break,
+        }
+    }
+    let truncated_by_lines = lines_iter.next().is_some();
+
+    if !truncated_by_bytes && !truncated_by_lines {
+        return content.to_string();
+    }
+
+    if preview.len() < truncated_slice.len()
+        && truncated_slice
+            .as_bytes()
+            .get(preview.len())
+            .is_some_and(|byte| *byte == b'\n')
+    {
+        preview.push('\n');
+    }
+
+    if !preview.is_empty() && !preview.ends_with('\n') {
+        preview.push('\n');
+    }
+    preview.push_str(TELEMETRY_PREVIEW_TRUNCATION_NOTICE);
+
+    preview
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use pretty_assertions::assert_eq;
+
+    #[test]
+    fn custom_tool_calls_should_roundtrip_as_custom_outputs() {
+        let payload = ToolPayload::Custom {
+            input: "patch".to_string(),
+        };
+        let response = ToolOutput::Function {
+            content: "patched".to_string(),
+            success: Some(true),
+        }
+        .into_response("call-42", &payload);
+
+        match response {
+            ResponseInputItem::CustomToolCallOutput { call_id, output } => {
+                assert_eq!(call_id, "call-42");
+                assert_eq!(output, "patched");
+            }
+            other => panic!("expected CustomToolCallOutput, got {other:?}"),
+        }
+    }
+
+    #[test]
+    fn function_payloads_remain_function_outputs() {
+        let payload = ToolPayload::Function {
+            arguments: "{}".to_string(),
+        };
+        let response = ToolOutput::Function {
+            content: "ok".to_string(),
+            success: Some(true),
+        }
+        .into_response("fn-1", &payload);
+
+        match response {
+            ResponseInputItem::FunctionCallOutput { call_id, output } => {
+                assert_eq!(call_id, "fn-1");
+                assert_eq!(output.content, "ok");
+                assert_eq!(output.success, Some(true));
+            }
+            other => panic!("expected FunctionCallOutput, got {other:?}"),
+        }
+    }
+
+    #[test]
+    fn telemetry_preview_returns_original_within_limits() {
+        let content = "short output";
+        assert_eq!(telemetry_preview(content), content);
+    }
+
+    #[test]
+    fn telemetry_preview_truncates_by_bytes() {
+        let content = "x".repeat(TELEMETRY_PREVIEW_MAX_BYTES + 8);
+        let preview = telemetry_preview(&content);
+
+        assert!(preview.contains(TELEMETRY_PREVIEW_TRUNCATION_NOTICE));
+        assert!(
+            preview.len()
+                <= TELEMETRY_PREVIEW_MAX_BYTES + TELEMETRY_PREVIEW_TRUNCATION_NOTICE.len() + 1
+        );
+    }
+
+    #[test]
+    fn telemetry_preview_truncates_by_lines() {
+        let content = (0..(TELEMETRY_PREVIEW_MAX_LINES + 5))
+            .map(|idx| format!("line {idx}"))
+            .collect::<Vec<_>>()
+            .join("\n");
+
+        let preview = telemetry_preview(&content);
+        let lines: Vec<&str> = preview.lines().collect();
+
+        assert!(lines.len() <= TELEMETRY_PREVIEW_MAX_LINES + 1);
+        assert_eq!(lines.last(), Some(&TELEMETRY_PREVIEW_TRUNCATION_NOTICE));
+    }
+}
+
+#[derive(Clone, Debug)]
+pub(crate) struct ExecCommandContext {
+    pub(crate) sub_id: String,
+    pub(crate) call_id: String,
+    pub(crate) command_for_display: Vec<String>,
+    pub(crate) cwd: PathBuf,
+    pub(crate) apply_patch: Option<ApplyPatchCommandContext>,
+    pub(crate) tool_name: String,
+    pub(crate) otel_event_manager: OtelEventManager,
+}
+
+#[derive(Clone, Debug)]
+pub(crate) struct ApplyPatchCommandContext {
+    pub(crate) user_explicitly_approved_this_action: bool,
+    pub(crate) changes: HashMap<PathBuf, FileChange>,
+}

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

@@ -1,15 +1,97 @@
+use std::collections::BTreeMap;
+use std::collections::HashMap;
+use std::sync::Arc;
+
+use crate::client_common::tools::FreeformTool;
+use crate::client_common::tools::FreeformToolFormat;
+use crate::client_common::tools::ResponsesApiTool;
+use crate::client_common::tools::ToolSpec;
+use crate::exec::ExecParams;
+use crate::function_tool::FunctionCallError;
+use crate::openai_tools::JsonSchema;
+use crate::tools::context::ToolInvocation;
+use crate::tools::context::ToolOutput;
+use crate::tools::context::ToolPayload;
+use crate::tools::handle_container_exec_with_params;
+use crate::tools::registry::ToolHandler;
+use crate::tools::registry::ToolKind;
+use crate::tools::spec::ApplyPatchToolArgs;
+use async_trait::async_trait;
 use serde::Deserialize;
 use serde::Serialize;
-use std::collections::BTreeMap;
 
-use crate::openai_tools::FreeformTool;
-use crate::openai_tools::FreeformToolFormat;
-use crate::openai_tools::JsonSchema;
-use crate::openai_tools::OpenAiTool;
-use crate::openai_tools::ResponsesApiTool;
+pub struct ApplyPatchHandler;
 
 const APPLY_PATCH_LARK_GRAMMAR: &str = include_str!("tool_apply_patch.lark");
 
+#[async_trait]
+impl ToolHandler for ApplyPatchHandler {
+    fn kind(&self) -> ToolKind {
+        ToolKind::Function
+    }
+
+    fn matches_kind(&self, payload: &ToolPayload) -> bool {
+        matches!(
+            payload,
+            ToolPayload::Function { .. } | ToolPayload::Custom { .. }
+        )
+    }
+
+    async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError> {
+        let ToolInvocation {
+            session,
+            turn,
+            tracker,
+            sub_id,
+            call_id,
+            tool_name,
+            payload,
+        } = invocation;
+
+        let patch_input = match payload {
+            ToolPayload::Function { arguments } => {
+                let args: ApplyPatchToolArgs = serde_json::from_str(&arguments).map_err(|e| {
+                    FunctionCallError::RespondToModel(format!(
+                        "failed to parse function arguments: {e:?}"
+                    ))
+                })?;
+                args.input
+            }
+            ToolPayload::Custom { input } => input,
+            _ => {
+                return Err(FunctionCallError::RespondToModel(
+                    "apply_patch handler received unsupported payload".to_string(),
+                ));
+            }
+        };
+
+        let exec_params = ExecParams {
+            command: vec!["apply_patch".to_string(), patch_input.clone()],
+            cwd: turn.cwd.clone(),
+            timeout_ms: None,
+            env: HashMap::new(),
+            with_escalated_permissions: None,
+            justification: None,
+        };
+
+        let content = handle_container_exec_with_params(
+            tool_name.as_str(),
+            exec_params,
+            Arc::clone(&session),
+            Arc::clone(&turn),
+            Arc::clone(&tracker),
+            sub_id.clone(),
+            call_id.clone(),
+        )
+        .await?;
+
+        Ok(ToolOutput::Function {
+            content,
+            success: Some(true),
+        })
+    }
+}
+
 #[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq, Hash)]
 #[serde(rename_all = "snake_case")]
 pub enum ApplyPatchToolType {
@@ -19,10 +101,10 @@ pub enum ApplyPatchToolType {
 
 /// Returns a custom tool that can be used to edit files. Well-suited for GPT-5 models
 /// https://platform.openai.com/docs/guides/function-calling#custom-tools
-pub(crate) fn create_apply_patch_freeform_tool() -> OpenAiTool {
-    OpenAiTool::Freeform(FreeformTool {
+pub(crate) fn create_apply_patch_freeform_tool() -> ToolSpec {
+    ToolSpec::Freeform(FreeformTool {
         name: "apply_patch".to_string(),
-        description: "Use the `apply_patch` tool to edit files".to_string(),
+        description: "Use the `apply_patch` tool to edit files. This is a FREEFORM tool, so do not wrap the patch in JSON.".to_string(),
         format: FreeformToolFormat {
             r#type: "grammar".to_string(),
             syntax: "lark".to_string(),
@@ -32,7 +114,7 @@ pub(crate) fn create_apply_patch_freeform_tool() -> OpenAiTool {
 }
 
 /// Returns a json tool that can be used to edit files. Should only be used with gpt-oss models
-pub(crate) fn create_apply_patch_json_tool() -> OpenAiTool {
+pub(crate) fn create_apply_patch_json_tool() -> ToolSpec {
     let mut properties = BTreeMap::new();
     properties.insert(
         "input".to_string(),
@@ -41,7 +123,7 @@ pub(crate) fn create_apply_patch_json_tool() -> OpenAiTool {
         },
     );
 
-    OpenAiTool::Function(ResponsesApiTool {
+    ToolSpec::Function(ResponsesApiTool {
         name: "apply_patch".to_string(),
         description: r#"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:
@@ -111,12 +193,12 @@ It is important to remember:
 - You must prefix new lines with `+` even when creating a new file
 - File references can only be relative, NEVER ABSOLUTE.
 "#
-        .to_string(),
+            .to_string(),
         strict: false,
         parameters: JsonSchema::Object {
             properties,
             required: Some(vec!["input".to_string()]),
-            additional_properties: Some(false),
+            additional_properties: Some(false.into()),
         },
     })
 }

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

@@ -0,0 +1,68 @@
+use async_trait::async_trait;
+
+use crate::exec_command::EXEC_COMMAND_TOOL_NAME;
+use crate::exec_command::ExecCommandParams;
+use crate::exec_command::WRITE_STDIN_TOOL_NAME;
+use crate::exec_command::WriteStdinParams;
+use crate::function_tool::FunctionCallError;
+use crate::tools::context::ToolInvocation;
+use crate::tools::context::ToolOutput;
+use crate::tools::context::ToolPayload;
+use crate::tools::registry::ToolHandler;
+use crate::tools::registry::ToolKind;
+
+pub struct ExecStreamHandler;
+
+#[async_trait]
+impl ToolHandler for ExecStreamHandler {
+    fn kind(&self) -> ToolKind {
+        ToolKind::Function
+    }
+
+    async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError> {
+        let ToolInvocation {
+            session,
+            tool_name,
+            payload,
+            ..
+        } = invocation;
+
+        let arguments = match payload {
+            ToolPayload::Function { arguments } => arguments,
+            _ => {
+                return Err(FunctionCallError::RespondToModel(
+                    "exec_stream handler received unsupported payload".to_string(),
+                ));
+            }
+        };
+
+        let content = match tool_name.as_str() {
+            EXEC_COMMAND_TOOL_NAME => {
+                let params: ExecCommandParams = serde_json::from_str(&arguments).map_err(|e| {
+                    FunctionCallError::RespondToModel(format!(
+                        "failed to parse function arguments: {e:?}"
+                    ))
+                })?;
+                session.handle_exec_command_tool(params).await?
+            }
+            WRITE_STDIN_TOOL_NAME => {
+                let params: WriteStdinParams = serde_json::from_str(&arguments).map_err(|e| {
+                    FunctionCallError::RespondToModel(format!(
+                        "failed to parse function arguments: {e:?}"
+                    ))
+                })?;
+                session.handle_write_stdin_tool(params).await?
+            }
+            _ => {
+                return Err(FunctionCallError::RespondToModel(format!(
+                    "exec_stream handler does not support tool {tool_name}"
+                )));
+            }
+        };
+
+        Ok(ToolOutput::Function {
+            content,
+            success: Some(true),
+        })
+    }
+}

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

@@ -0,0 +1,272 @@
+use std::path::Path;
+use std::time::Duration;
+
+use async_trait::async_trait;
+use serde::Deserialize;
+use tokio::process::Command;
+use tokio::time::timeout;
+
+use crate::function_tool::FunctionCallError;
+use crate::tools::context::ToolInvocation;
+use crate::tools::context::ToolOutput;
+use crate::tools::context::ToolPayload;
+use crate::tools::registry::ToolHandler;
+use crate::tools::registry::ToolKind;
+
+pub struct GrepFilesHandler;
+
+const DEFAULT_LIMIT: usize = 100;
+const MAX_LIMIT: usize = 2000;
+const COMMAND_TIMEOUT: Duration = Duration::from_secs(30);
+
+fn default_limit() -> usize {
+    DEFAULT_LIMIT
+}
+
+#[derive(Deserialize)]
+struct GrepFilesArgs {
+    pattern: String,
+    #[serde(default)]
+    include: Option<String>,
+    #[serde(default)]
+    path: Option<String>,
+    #[serde(default = "default_limit")]
+    limit: usize,
+}
+
+#[async_trait]
+impl ToolHandler for GrepFilesHandler {
+    fn kind(&self) -> ToolKind {
+        ToolKind::Function
+    }
+
+    async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError> {
+        let ToolInvocation { payload, turn, .. } = invocation;
+
+        let arguments = match payload {
+            ToolPayload::Function { arguments } => arguments,
+            _ => {
+                return Err(FunctionCallError::RespondToModel(
+                    "grep_files handler received unsupported payload".to_string(),
+                ));
+            }
+        };
+
+        let args: GrepFilesArgs = serde_json::from_str(&arguments).map_err(|err| {
+            FunctionCallError::RespondToModel(format!(
+                "failed to parse function arguments: {err:?}"
+            ))
+        })?;
+
+        let pattern = args.pattern.trim();
+        if pattern.is_empty() {
+            return Err(FunctionCallError::RespondToModel(
+                "pattern must not be empty".to_string(),
+            ));
+        }
+
+        if args.limit == 0 {
+            return Err(FunctionCallError::RespondToModel(
+                "limit must be greater than zero".to_string(),
+            ));
+        }
+
+        let limit = args.limit.min(MAX_LIMIT);
+        let search_path = turn.resolve_path(args.path.clone());
+
+        verify_path_exists(&search_path).await?;
+
+        let include = args.include.as_deref().map(str::trim).and_then(|val| {
+            if val.is_empty() {
+                None
+            } else {
+                Some(val.to_string())
+            }
+        });
+
+        let search_results =
+            run_rg_search(pattern, include.as_deref(), &search_path, limit, &turn.cwd).await?;
+
+        if search_results.is_empty() {
+            Ok(ToolOutput::Function {
+                content: "No matches found.".to_string(),
+                success: Some(false),
+            })
+        } else {
+            Ok(ToolOutput::Function {
+                content: search_results.join("\n"),
+                success: Some(true),
+            })
+        }
+    }
+}
+
+async fn verify_path_exists(path: &Path) -> Result<(), FunctionCallError> {
+    tokio::fs::metadata(path).await.map_err(|err| {
+        FunctionCallError::RespondToModel(format!("unable to access `{}`: {err}", path.display()))
+    })?;
+    Ok(())
+}
+
+async fn run_rg_search(
+    pattern: &str,
+    include: Option<&str>,
+    search_path: &Path,
+    limit: usize,
+    cwd: &Path,
+) -> Result<Vec<String>, FunctionCallError> {
+    let mut command = Command::new("rg");
+    command
+        .current_dir(cwd)
+        .arg("--files-with-matches")
+        .arg("--sortr=modified")
+        .arg("--regexp")
+        .arg(pattern)
+        .arg("--no-messages");
+
+    if let Some(glob) = include {
+        command.arg("--glob").arg(glob);
+    }
+
+    command.arg("--").arg(search_path);
+
+    let output = timeout(COMMAND_TIMEOUT, command.output())
+        .await
+        .map_err(|_| {
+            FunctionCallError::RespondToModel("rg timed out after 30 seconds".to_string())
+        })?
+        .map_err(|err| {
+            FunctionCallError::RespondToModel(format!(
+                "failed to launch rg: {err}. Ensure ripgrep is installed and on PATH."
+            ))
+        })?;
+
+    match output.status.code() {
+        Some(0) => Ok(parse_results(&output.stdout, limit)),
+        Some(1) => Ok(Vec::new()),
+        _ => {
+            let stderr = String::from_utf8_lossy(&output.stderr);
+            Err(FunctionCallError::RespondToModel(format!(
+                "rg failed: {stderr}"
+            )))
+        }
+    }
+}
+
+fn parse_results(stdout: &[u8], limit: usize) -> Vec<String> {
+    let mut results = Vec::new();
+    for line in stdout.split(|byte| *byte == b'\n') {
+        if line.is_empty() {
+            continue;
+        }
+        if let Ok(text) = std::str::from_utf8(line) {
+            if text.is_empty() {
+                continue;
+            }
+            results.push(text.to_string());
+            if results.len() == limit {
+                break;
+            }
+        }
+    }
+    results
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::process::Command as StdCommand;
+    use tempfile::tempdir;
+
+    #[test]
+    fn parses_basic_results() {
+        let stdout = b"/tmp/file_a.rs\n/tmp/file_b.rs\n";
+        let parsed = parse_results(stdout, 10);
+        assert_eq!(
+            parsed,
+            vec!["/tmp/file_a.rs".to_string(), "/tmp/file_b.rs".to_string()]
+        );
+    }
+
+    #[test]
+    fn parse_truncates_after_limit() {
+        let stdout = b"/tmp/file_a.rs\n/tmp/file_b.rs\n/tmp/file_c.rs\n";
+        let parsed = parse_results(stdout, 2);
+        assert_eq!(
+            parsed,
+            vec!["/tmp/file_a.rs".to_string(), "/tmp/file_b.rs".to_string()]
+        );
+    }
+
+    #[tokio::test]
+    async fn run_search_returns_results() -> anyhow::Result<()> {
+        if !rg_available() {
+            return Ok(());
+        }
+        let temp = tempdir().expect("create temp dir");
+        let dir = temp.path();
+        std::fs::write(dir.join("match_one.txt"), "alpha beta gamma").unwrap();
+        std::fs::write(dir.join("match_two.txt"), "alpha delta").unwrap();
+        std::fs::write(dir.join("other.txt"), "omega").unwrap();
+
+        let results = run_rg_search("alpha", None, dir, 10, dir).await?;
+        assert_eq!(results.len(), 2);
+        assert!(results.iter().any(|path| path.ends_with("match_one.txt")));
+        assert!(results.iter().any(|path| path.ends_with("match_two.txt")));
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn run_search_with_glob_filter() -> anyhow::Result<()> {
+        if !rg_available() {
+            return Ok(());
+        }
+        let temp = tempdir().expect("create temp dir");
+        let dir = temp.path();
+        std::fs::write(dir.join("match_one.rs"), "alpha beta gamma").unwrap();
+        std::fs::write(dir.join("match_two.txt"), "alpha delta").unwrap();
+
+        let results = run_rg_search("alpha", Some("*.rs"), dir, 10, dir).await?;
+        assert_eq!(results.len(), 1);
+        assert!(results.iter().all(|path| path.ends_with("match_one.rs")));
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn run_search_respects_limit() -> anyhow::Result<()> {
+        if !rg_available() {
+            return Ok(());
+        }
+        let temp = tempdir().expect("create temp dir");
+        let dir = temp.path();
+        std::fs::write(dir.join("one.txt"), "alpha one").unwrap();
+        std::fs::write(dir.join("two.txt"), "alpha two").unwrap();
+        std::fs::write(dir.join("three.txt"), "alpha three").unwrap();
+
+        let results = run_rg_search("alpha", None, dir, 2, dir).await?;
+        assert_eq!(results.len(), 2);
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn run_search_handles_no_matches() -> anyhow::Result<()> {
+        if !rg_available() {
+            return Ok(());
+        }
+        let temp = tempdir().expect("create temp dir");
+        let dir = temp.path();
+        std::fs::write(dir.join("one.txt"), "omega").unwrap();
+
+        let results = run_rg_search("alpha", None, dir, 5, dir).await?;
+        assert!(results.is_empty());
+        Ok(())
+    }
+
+    fn rg_available() -> bool {
+        StdCommand::new("rg")
+            .arg("--version")
+            .output()
+            .map(|output| output.status.success())
+            .unwrap_or(false)
+    }
+}

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

@@ -0,0 +1,476 @@
+use std::collections::VecDeque;
+use std::ffi::OsStr;
+use std::fs::FileType;
+use std::path::Path;
+use std::path::PathBuf;
+
+use async_trait::async_trait;
+use codex_utils_string::take_bytes_at_char_boundary;
+use serde::Deserialize;
+use tokio::fs;
+
+use crate::function_tool::FunctionCallError;
+use crate::tools::context::ToolInvocation;
+use crate::tools::context::ToolOutput;
+use crate::tools::context::ToolPayload;
+use crate::tools::registry::ToolHandler;
+use crate::tools::registry::ToolKind;
+
+pub struct ListDirHandler;
+
+const MAX_ENTRY_LENGTH: usize = 500;
+const INDENTATION_SPACES: usize = 2;
+
+fn default_offset() -> usize {
+    1
+}
+
+fn default_limit() -> usize {
+    25
+}
+
+fn default_depth() -> usize {
+    2
+}
+
+#[derive(Deserialize)]
+struct ListDirArgs {
+    dir_path: String,
+    #[serde(default = "default_offset")]
+    offset: usize,
+    #[serde(default = "default_limit")]
+    limit: usize,
+    #[serde(default = "default_depth")]
+    depth: usize,
+}
+
+#[async_trait]
+impl ToolHandler for ListDirHandler {
+    fn kind(&self) -> ToolKind {
+        ToolKind::Function
+    }
+
+    async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError> {
+        let ToolInvocation { payload, .. } = invocation;
+
+        let arguments = match payload {
+            ToolPayload::Function { arguments } => arguments,
+            _ => {
+                return Err(FunctionCallError::RespondToModel(
+                    "list_dir handler received unsupported payload".to_string(),
+                ));
+            }
+        };
+
+        let args: ListDirArgs = serde_json::from_str(&arguments).map_err(|err| {
+            FunctionCallError::RespondToModel(format!(
+                "failed to parse function arguments: {err:?}"
+            ))
+        })?;
+
+        let ListDirArgs {
+            dir_path,
+            offset,
+            limit,
+            depth,
+        } = args;
+
+        if offset == 0 {
+            return Err(FunctionCallError::RespondToModel(
+                "offset must be a 1-indexed entry number".to_string(),
+            ));
+        }
+
+        if limit == 0 {
+            return Err(FunctionCallError::RespondToModel(
+                "limit must be greater than zero".to_string(),
+            ));
+        }
+
+        if depth == 0 {
+            return Err(FunctionCallError::RespondToModel(
+                "depth must be greater than zero".to_string(),
+            ));
+        }
+
+        let path = PathBuf::from(&dir_path);
+        if !path.is_absolute() {
+            return Err(FunctionCallError::RespondToModel(
+                "dir_path must be an absolute path".to_string(),
+            ));
+        }
+
+        let entries = list_dir_slice(&path, offset, limit, depth).await?;
+        let mut output = Vec::with_capacity(entries.len() + 1);
+        output.push(format!("Absolute path: {}", path.display()));
+        output.extend(entries);
+        Ok(ToolOutput::Function {
+            content: output.join("\n"),
+            success: Some(true),
+        })
+    }
+}
+
+async fn list_dir_slice(
+    path: &Path,
+    offset: usize,
+    limit: usize,
+    depth: usize,
+) -> Result<Vec<String>, FunctionCallError> {
+    let mut entries = Vec::new();
+    collect_entries(path, Path::new(""), depth, &mut entries).await?;
+
+    if entries.is_empty() {
+        return Ok(Vec::new());
+    }
+
+    let start_index = offset - 1;
+    if start_index >= entries.len() {
+        return Err(FunctionCallError::RespondToModel(
+            "offset exceeds directory entry count".to_string(),
+        ));
+    }
+
+    let remaining_entries = entries.len() - start_index;
+    let capped_limit = limit.min(remaining_entries);
+    let end_index = start_index + capped_limit;
+    let mut selected_entries = entries[start_index..end_index].to_vec();
+    selected_entries.sort_unstable_by(|a, b| a.name.cmp(&b.name));
+    let mut formatted = Vec::with_capacity(selected_entries.len());
+
+    for entry in &selected_entries {
+        formatted.push(format_entry_line(entry));
+    }
+
+    if end_index < entries.len() {
+        formatted.push(format!("More than {capped_limit} entries found"));
+    }
+
+    Ok(formatted)
+}
+
+async fn collect_entries(
+    dir_path: &Path,
+    relative_prefix: &Path,
+    depth: usize,
+    entries: &mut Vec<DirEntry>,
+) -> Result<(), FunctionCallError> {
+    let mut queue = VecDeque::new();
+    queue.push_back((dir_path.to_path_buf(), relative_prefix.to_path_buf(), depth));
+
+    while let Some((current_dir, prefix, remaining_depth)) = queue.pop_front() {
+        let mut read_dir = fs::read_dir(&current_dir).await.map_err(|err| {
+            FunctionCallError::RespondToModel(format!("failed to read directory: {err}"))
+        })?;
+
+        let mut dir_entries = Vec::new();
+
+        while let Some(entry) = read_dir.next_entry().await.map_err(|err| {
+            FunctionCallError::RespondToModel(format!("failed to read directory: {err}"))
+        })? {
+            let file_type = entry.file_type().await.map_err(|err| {
+                FunctionCallError::RespondToModel(format!("failed to inspect entry: {err}"))
+            })?;
+
+            let file_name = entry.file_name();
+            let relative_path = if prefix.as_os_str().is_empty() {
+                PathBuf::from(&file_name)
+            } else {
+                prefix.join(&file_name)
+            };
+
+            let display_name = format_entry_component(&file_name);
+            let display_depth = prefix.components().count();
+            let sort_key = format_entry_name(&relative_path);
+            let kind = DirEntryKind::from(&file_type);
+            dir_entries.push((
+                entry.path(),
+                relative_path,
+                kind,
+                DirEntry {
+                    name: sort_key,
+                    display_name,
+                    depth: display_depth,
+                    kind,
+                },
+            ));
+        }
+
+        dir_entries.sort_unstable_by(|a, b| a.3.name.cmp(&b.3.name));
+
+        for (entry_path, relative_path, kind, dir_entry) in dir_entries {
+            if kind == DirEntryKind::Directory && remaining_depth > 1 {
+                queue.push_back((entry_path, relative_path, remaining_depth - 1));
+            }
+            entries.push(dir_entry);
+        }
+    }
+
+    Ok(())
+}
+
+fn format_entry_name(path: &Path) -> String {
+    let normalized = path.to_string_lossy().replace("\\", "/");
+    if normalized.len() > MAX_ENTRY_LENGTH {
+        take_bytes_at_char_boundary(&normalized, MAX_ENTRY_LENGTH).to_string()
+    } else {
+        normalized
+    }
+}
+
+fn format_entry_component(name: &OsStr) -> String {
+    let normalized = name.to_string_lossy();
+    if normalized.len() > MAX_ENTRY_LENGTH {
+        take_bytes_at_char_boundary(&normalized, MAX_ENTRY_LENGTH).to_string()
+    } else {
+        normalized.to_string()
+    }
+}
+
+fn format_entry_line(entry: &DirEntry) -> String {
+    let indent = " ".repeat(entry.depth * INDENTATION_SPACES);
+    let mut name = entry.display_name.clone();
+    match entry.kind {
+        DirEntryKind::Directory => name.push('/'),
+        DirEntryKind::Symlink => name.push('@'),
+        DirEntryKind::Other => name.push('?'),
+        DirEntryKind::File => {}
+    }
+    format!("{indent}{name}")
+}
+
+#[derive(Clone)]
+struct DirEntry {
+    name: String,
+    display_name: String,
+    depth: usize,
+    kind: DirEntryKind,
+}
+
+#[derive(Clone, Copy, PartialEq, Eq)]
+enum DirEntryKind {
+    Directory,
+    File,
+    Symlink,
+    Other,
+}
+
+impl From<&FileType> for DirEntryKind {
+    fn from(file_type: &FileType) -> Self {
+        if file_type.is_symlink() {
+            DirEntryKind::Symlink
+        } else if file_type.is_dir() {
+            DirEntryKind::Directory
+        } else if file_type.is_file() {
+            DirEntryKind::File
+        } else {
+            DirEntryKind::Other
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use tempfile::tempdir;
+
+    #[tokio::test]
+    async fn lists_directory_entries() {
+        let temp = tempdir().expect("create tempdir");
+        let dir_path = temp.path();
+
+        let sub_dir = dir_path.join("nested");
+        tokio::fs::create_dir(&sub_dir)
+            .await
+            .expect("create sub dir");
+
+        let deeper_dir = sub_dir.join("deeper");
+        tokio::fs::create_dir(&deeper_dir)
+            .await
+            .expect("create deeper dir");
+
+        tokio::fs::write(dir_path.join("entry.txt"), b"content")
+            .await
+            .expect("write file");
+        tokio::fs::write(sub_dir.join("child.txt"), b"child")
+            .await
+            .expect("write child");
+        tokio::fs::write(deeper_dir.join("grandchild.txt"), b"grandchild")
+            .await
+            .expect("write grandchild");
+
+        #[cfg(unix)]
+        {
+            use std::os::unix::fs::symlink;
+            let link_path = dir_path.join("link");
+            symlink(dir_path.join("entry.txt"), &link_path).expect("create symlink");
+        }
+
+        let entries = list_dir_slice(dir_path, 1, 20, 3)
+            .await
+            .expect("list directory");
+
+        #[cfg(unix)]
+        let expected = vec![
+            "entry.txt".to_string(),
+            "link@".to_string(),
+            "nested/".to_string(),
+            "  child.txt".to_string(),
+            "  deeper/".to_string(),
+            "    grandchild.txt".to_string(),
+        ];
+
+        #[cfg(not(unix))]
+        let expected = vec![
+            "entry.txt".to_string(),
+            "nested/".to_string(),
+            "  child.txt".to_string(),
+            "  deeper/".to_string(),
+            "    grandchild.txt".to_string(),
+        ];
+
+        assert_eq!(entries, expected);
+    }
+
+    #[tokio::test]
+    async fn errors_when_offset_exceeds_entries() {
+        let temp = tempdir().expect("create tempdir");
+        let dir_path = temp.path();
+        tokio::fs::create_dir(dir_path.join("nested"))
+            .await
+            .expect("create sub dir");
+
+        let err = list_dir_slice(dir_path, 10, 1, 2)
+            .await
+            .expect_err("offset exceeds entries");
+        assert_eq!(
+            err,
+            FunctionCallError::RespondToModel("offset exceeds directory entry count".to_string())
+        );
+    }
+
+    #[tokio::test]
+    async fn respects_depth_parameter() {
+        let temp = tempdir().expect("create tempdir");
+        let dir_path = temp.path();
+        let nested = dir_path.join("nested");
+        let deeper = nested.join("deeper");
+        tokio::fs::create_dir(&nested).await.expect("create nested");
+        tokio::fs::create_dir(&deeper).await.expect("create deeper");
+        tokio::fs::write(dir_path.join("root.txt"), b"root")
+            .await
+            .expect("write root");
+        tokio::fs::write(nested.join("child.txt"), b"child")
+            .await
+            .expect("write nested");
+        tokio::fs::write(deeper.join("grandchild.txt"), b"deep")
+            .await
+            .expect("write deeper");
+
+        let entries_depth_one = list_dir_slice(dir_path, 1, 10, 1)
+            .await
+            .expect("list depth 1");
+        assert_eq!(
+            entries_depth_one,
+            vec!["nested/".to_string(), "root.txt".to_string(),]
+        );
+
+        let entries_depth_two = list_dir_slice(dir_path, 1, 20, 2)
+            .await
+            .expect("list depth 2");
+        assert_eq!(
+            entries_depth_two,
+            vec![
+                "nested/".to_string(),
+                "  child.txt".to_string(),
+                "  deeper/".to_string(),
+                "root.txt".to_string(),
+            ]
+        );
+
+        let entries_depth_three = list_dir_slice(dir_path, 1, 30, 3)
+            .await
+            .expect("list depth 3");
+        assert_eq!(
+            entries_depth_three,
+            vec![
+                "nested/".to_string(),
+                "  child.txt".to_string(),
+                "  deeper/".to_string(),
+                "    grandchild.txt".to_string(),
+                "root.txt".to_string(),
+            ]
+        );
+    }
+
+    #[tokio::test]
+    async fn handles_large_limit_without_overflow() {
+        let temp = tempdir().expect("create tempdir");
+        let dir_path = temp.path();
+        tokio::fs::write(dir_path.join("alpha.txt"), b"alpha")
+            .await
+            .expect("write alpha");
+        tokio::fs::write(dir_path.join("beta.txt"), b"beta")
+            .await
+            .expect("write beta");
+        tokio::fs::write(dir_path.join("gamma.txt"), b"gamma")
+            .await
+            .expect("write gamma");
+
+        let entries = list_dir_slice(dir_path, 2, usize::MAX, 1)
+            .await
+            .expect("list without overflow");
+        assert_eq!(
+            entries,
+            vec!["beta.txt".to_string(), "gamma.txt".to_string(),]
+        );
+    }
+
+    #[tokio::test]
+    async fn indicates_truncated_results() {
+        let temp = tempdir().expect("create tempdir");
+        let dir_path = temp.path();
+
+        for idx in 0..40 {
+            let file = dir_path.join(format!("file_{idx:02}.txt"));
+            tokio::fs::write(file, b"content")
+                .await
+                .expect("write file");
+        }
+
+        let entries = list_dir_slice(dir_path, 1, 25, 1)
+            .await
+            .expect("list directory");
+        assert_eq!(entries.len(), 26);
+        assert_eq!(
+            entries.last(),
+            Some(&"More than 25 entries found".to_string())
+        );
+    }
+
+    #[tokio::test]
+    async fn bfs_truncation() -> anyhow::Result<()> {
+        let temp = tempdir()?;
+        let dir_path = temp.path();
+        let nested = dir_path.join("nested");
+        let deeper = nested.join("deeper");
+        tokio::fs::create_dir(&nested).await?;
+        tokio::fs::create_dir(&deeper).await?;
+        tokio::fs::write(dir_path.join("root.txt"), b"root").await?;
+        tokio::fs::write(nested.join("child.txt"), b"child").await?;
+        tokio::fs::write(deeper.join("grandchild.txt"), b"deep").await?;
+
+        let entries_depth_three = list_dir_slice(dir_path, 1, 3, 3).await?;
+        assert_eq!(
+            entries_depth_three,
+            vec![
+                "nested/".to_string(),
+                "  child.txt".to_string(),
+                "root.txt".to_string(),
+                "More than 3 entries found".to_string()
+            ]
+        );
+
+        Ok(())
+    }
+}

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

@@ -0,0 +1,67 @@
+use async_trait::async_trait;
+
+use crate::function_tool::FunctionCallError;
+use crate::mcp_tool_call::handle_mcp_tool_call;
+use crate::tools::context::ToolInvocation;
+use crate::tools::context::ToolOutput;
+use crate::tools::context::ToolPayload;
+use crate::tools::registry::ToolHandler;
+use crate::tools::registry::ToolKind;
+
+pub struct McpHandler;
+
+#[async_trait]
+impl ToolHandler for McpHandler {
+    fn kind(&self) -> ToolKind {
+        ToolKind::Mcp
+    }
+
+    async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError> {
+        let ToolInvocation {
+            session,
+            sub_id,
+            call_id,
+            payload,
+            ..
+        } = invocation;
+
+        let payload = match payload {
+            ToolPayload::Mcp {
+                server,
+                tool,
+                raw_arguments,
+            } => (server, tool, raw_arguments),
+            _ => {
+                return Err(FunctionCallError::RespondToModel(
+                    "mcp handler received unsupported payload".to_string(),
+                ));
+            }
+        };
+
+        let (server, tool, raw_arguments) = payload;
+        let arguments_str = raw_arguments;
+
+        let response = handle_mcp_tool_call(
+            session.as_ref(),
+            &sub_id,
+            call_id.clone(),
+            server,
+            tool,
+            arguments_str,
+        )
+        .await;
+
+        match response {
+            codex_protocol::models::ResponseInputItem::McpToolCallOutput { result, .. } => {
+                Ok(ToolOutput::Mcp { result })
+            }
+            codex_protocol::models::ResponseInputItem::FunctionCallOutput { output, .. } => {
+                let codex_protocol::models::FunctionCallOutputPayload { content, success } = output;
+                Ok(ToolOutput::Function { content, success })
+            }
+            _ => Err(FunctionCallError::RespondToModel(
+                "mcp handler received unexpected response variant".to_string(),
+            )),
+        }
+    }
+}

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

@@ -0,0 +1,25 @@
+pub mod apply_patch;
+mod exec_stream;
+mod grep_files;
+mod list_dir;
+mod mcp;
+mod plan;
+mod read_file;
+mod shell;
+mod test_sync;
+mod unified_exec;
+mod view_image;
+
+pub use plan::PLAN_TOOL;
+
+pub use apply_patch::ApplyPatchHandler;
+pub use exec_stream::ExecStreamHandler;
+pub use grep_files::GrepFilesHandler;
+pub use list_dir::ListDirHandler;
+pub use mcp::McpHandler;
+pub use plan::PlanHandler;
+pub use read_file::ReadFileHandler;
+pub use shell::ShellHandler;
+pub use test_sync::TestSyncHandler;
+pub use unified_exec::UnifiedExecHandler;
+pub use view_image::ViewImageHandler;

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

@@ -1,23 +1,23 @@
-use std::collections::BTreeMap;
-use std::sync::LazyLock;
-
+use crate::client_common::tools::ResponsesApiTool;
+use crate::client_common::tools::ToolSpec;
 use crate::codex::Session;
 use crate::function_tool::FunctionCallError;
 use crate::openai_tools::JsonSchema;
-use crate::openai_tools::OpenAiTool;
-use crate::openai_tools::ResponsesApiTool;
-use crate::protocol::Event;
-use crate::protocol::EventMsg;
+use crate::tools::context::ToolInvocation;
+use crate::tools::context::ToolOutput;
+use crate::tools::context::ToolPayload;
+use crate::tools::registry::ToolHandler;
+use crate::tools::registry::ToolKind;
+use async_trait::async_trait;
+use codex_protocol::plan_tool::UpdatePlanArgs;
+use codex_protocol::protocol::Event;
+use codex_protocol::protocol::EventMsg;
+use std::collections::BTreeMap;
+use std::sync::LazyLock;
 
-// Use the canonical plan tool types from the protocol crate to ensure
-// type-identity matches events transported via `codex_protocol`.
-pub use codex_protocol::plan_tool::PlanItemArg;
-pub use codex_protocol::plan_tool::StepStatus;
-pub use codex_protocol::plan_tool::UpdatePlanArgs;
+pub struct PlanHandler;
 
-// Types for the TODO tool arguments matching codex-vscode/todo-mcp/src/main.rs
-
-pub(crate) static PLAN_TOOL: LazyLock<OpenAiTool> = LazyLock::new(|| {
+pub static PLAN_TOOL: LazyLock<ToolSpec> = LazyLock::new(|| {
     let mut plan_item_props = BTreeMap::new();
     plan_item_props.insert("step".to_string(), JsonSchema::String { description: None });
     plan_item_props.insert(
@@ -32,7 +32,7 @@ pub(crate) static PLAN_TOOL: LazyLock<OpenAiTool> = LazyLock::new(|| {
         items: Box::new(JsonSchema::Object {
             properties: plan_item_props,
             required: Some(vec!["step".to_string(), "status".to_string()]),
-            additional_properties: Some(false),
+            additional_properties: Some(false.into()),
         }),
     };
 
@@ -43,7 +43,7 @@ pub(crate) static PLAN_TOOL: LazyLock<OpenAiTool> = LazyLock::new(|| {
     );
     properties.insert("plan".to_string(), plan_items_schema);
 
-    OpenAiTool::Function(ResponsesApiTool {
+    ToolSpec::Function(ResponsesApiTool {
         name: "update_plan".to_string(),
         description: r#"Updates the task plan.
 Provide an optional explanation and a list of plan items, each with a step and status.
@@ -54,11 +54,45 @@ At most one step can be in_progress at a time.
         parameters: JsonSchema::Object {
             properties,
             required: Some(vec!["plan".to_string()]),
-            additional_properties: Some(false),
+            additional_properties: Some(false.into()),
         },
     })
 });
 
+#[async_trait]
+impl ToolHandler for PlanHandler {
+    fn kind(&self) -> ToolKind {
+        ToolKind::Function
+    }
+
+    async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError> {
+        let ToolInvocation {
+            session,
+            sub_id,
+            call_id,
+            payload,
+            ..
+        } = invocation;
+
+        let arguments = match payload {
+            ToolPayload::Function { arguments } => arguments,
+            _ => {
+                return Err(FunctionCallError::RespondToModel(
+                    "update_plan handler received unsupported payload".to_string(),
+                ));
+            }
+        };
+
+        let content =
+            handle_update_plan(session.as_ref(), arguments, sub_id.clone(), call_id).await?;
+
+        Ok(ToolOutput::Function {
+            content,
+            success: Some(true),
+        })
+    }
+}
+
 /// This function doesn't do anything useful. However, it gives the model a structured way to record its plan that clients can read and render.
 /// So it's the _inputs_ to this function that are useful to clients, not the outputs and neither are actually useful for the model other
 /// than forcing it to come up and document a plan (TBD how that affects performance).

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

@@ -0,0 +1,998 @@
+use std::collections::VecDeque;
+use std::path::PathBuf;
+
+use async_trait::async_trait;
+use codex_utils_string::take_bytes_at_char_boundary;
+use serde::Deserialize;
+
+use crate::function_tool::FunctionCallError;
+use crate::tools::context::ToolInvocation;
+use crate::tools::context::ToolOutput;
+use crate::tools::context::ToolPayload;
+use crate::tools::registry::ToolHandler;
+use crate::tools::registry::ToolKind;
+
+pub struct ReadFileHandler;
+
+const MAX_LINE_LENGTH: usize = 500;
+const TAB_WIDTH: usize = 4;
+
+// TODO(jif) add support for block comments
+const COMMENT_PREFIXES: &[&str] = &["#", "//", "--"];
+
+/// JSON arguments accepted by the `read_file` tool handler.
+#[derive(Deserialize)]
+struct ReadFileArgs {
+    /// Absolute path to the file that will be read.
+    file_path: String,
+    /// 1-indexed line number to start reading from; defaults to 1.
+    #[serde(default = "defaults::offset")]
+    offset: usize,
+    /// Maximum number of lines to return; defaults to 2000.
+    #[serde(default = "defaults::limit")]
+    limit: usize,
+    /// Determines whether the handler reads a simple slice or indentation-aware block.
+    #[serde(default)]
+    mode: ReadMode,
+    /// Optional indentation configuration used when `mode` is `Indentation`.
+    #[serde(default)]
+    indentation: Option<IndentationArgs>,
+}
+
+#[derive(Deserialize)]
+#[serde(rename_all = "snake_case")]
+enum ReadMode {
+    Slice,
+    Indentation,
+}
+/// Additional configuration for indentation-aware reads.
+#[derive(Deserialize, Clone)]
+struct IndentationArgs {
+    /// Optional explicit anchor line; defaults to `offset` when omitted.
+    #[serde(default)]
+    anchor_line: Option<usize>,
+    /// Maximum indentation depth to collect; `0` means unlimited.
+    #[serde(default = "defaults::max_levels")]
+    max_levels: usize,
+    /// Whether to include sibling blocks at the same indentation level.
+    #[serde(default = "defaults::include_siblings")]
+    include_siblings: bool,
+    /// Whether to include header lines above the anchor block. This made on a best effort basis.
+    #[serde(default = "defaults::include_header")]
+    include_header: bool,
+    /// Optional hard cap on returned lines; defaults to the global `limit`.
+    #[serde(default)]
+    max_lines: Option<usize>,
+}
+
+#[derive(Clone, Debug)]
+struct LineRecord {
+    number: usize,
+    raw: String,
+    display: String,
+    indent: usize,
+}
+
+impl LineRecord {
+    fn trimmed(&self) -> &str {
+        self.raw.trim_start()
+    }
+
+    fn is_blank(&self) -> bool {
+        self.trimmed().is_empty()
+    }
+
+    fn is_comment(&self) -> bool {
+        COMMENT_PREFIXES
+            .iter()
+            .any(|prefix| self.raw.trim().starts_with(prefix))
+    }
+}
+
+#[async_trait]
+impl ToolHandler for ReadFileHandler {
+    fn kind(&self) -> ToolKind {
+        ToolKind::Function
+    }
+
+    async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError> {
+        let ToolInvocation { payload, .. } = invocation;
+
+        let arguments = match payload {
+            ToolPayload::Function { arguments } => arguments,
+            _ => {
+                return Err(FunctionCallError::RespondToModel(
+                    "read_file handler received unsupported payload".to_string(),
+                ));
+            }
+        };
+
+        let args: ReadFileArgs = serde_json::from_str(&arguments).map_err(|err| {
+            FunctionCallError::RespondToModel(format!(
+                "failed to parse function arguments: {err:?}"
+            ))
+        })?;
+
+        let ReadFileArgs {
+            file_path,
+            offset,
+            limit,
+            mode,
+            indentation,
+        } = args;
+
+        if offset == 0 {
+            return Err(FunctionCallError::RespondToModel(
+                "offset must be a 1-indexed line number".to_string(),
+            ));
+        }
+
+        if limit == 0 {
+            return Err(FunctionCallError::RespondToModel(
+                "limit must be greater than zero".to_string(),
+            ));
+        }
+
+        let path = PathBuf::from(&file_path);
+        if !path.is_absolute() {
+            return Err(FunctionCallError::RespondToModel(
+                "file_path must be an absolute path".to_string(),
+            ));
+        }
+
+        let collected = match mode {
+            ReadMode::Slice => slice::read(&path, offset, limit).await?,
+            ReadMode::Indentation => {
+                let indentation = indentation.unwrap_or_default();
+                indentation::read_block(&path, offset, limit, indentation).await?
+            }
+        };
+        Ok(ToolOutput::Function {
+            content: collected.join("\n"),
+            success: Some(true),
+        })
+    }
+}
+
+mod slice {
+    use crate::function_tool::FunctionCallError;
+    use crate::tools::handlers::read_file::format_line;
+    use std::path::Path;
+    use tokio::fs::File;
+    use tokio::io::AsyncBufReadExt;
+    use tokio::io::BufReader;
+
+    pub async fn read(
+        path: &Path,
+        offset: usize,
+        limit: usize,
+    ) -> Result<Vec<String>, FunctionCallError> {
+        let file = File::open(path).await.map_err(|err| {
+            FunctionCallError::RespondToModel(format!("failed to read file: {err}"))
+        })?;
+
+        let mut reader = BufReader::new(file);
+        let mut collected = Vec::new();
+        let mut seen = 0usize;
+        let mut buffer = Vec::new();
+
+        loop {
+            buffer.clear();
+            let bytes_read = reader.read_until(b'\n', &mut buffer).await.map_err(|err| {
+                FunctionCallError::RespondToModel(format!("failed to read file: {err}"))
+            })?;
+
+            if bytes_read == 0 {
+                break;
+            }
+
+            if buffer.last() == Some(&b'\n') {
+                buffer.pop();
+                if buffer.last() == Some(&b'\r') {
+                    buffer.pop();
+                }
+            }
+
+            seen += 1;
+
+            if seen < offset {
+                continue;
+            }
+
+            if collected.len() == limit {
+                break;
+            }
+
+            let formatted = format_line(&buffer);
+            collected.push(format!("L{seen}: {formatted}"));
+
+            if collected.len() == limit {
+                break;
+            }
+        }
+
+        if seen < offset {
+            return Err(FunctionCallError::RespondToModel(
+                "offset exceeds file length".to_string(),
+            ));
+        }
+
+        Ok(collected)
+    }
+}
+
+mod indentation {
+    use crate::function_tool::FunctionCallError;
+    use crate::tools::handlers::read_file::IndentationArgs;
+    use crate::tools::handlers::read_file::LineRecord;
+    use crate::tools::handlers::read_file::TAB_WIDTH;
+    use crate::tools::handlers::read_file::format_line;
+    use crate::tools::handlers::read_file::trim_empty_lines;
+    use std::collections::VecDeque;
+    use std::path::Path;
+    use tokio::fs::File;
+    use tokio::io::AsyncBufReadExt;
+    use tokio::io::BufReader;
+
+    pub async fn read_block(
+        path: &Path,
+        offset: usize,
+        limit: usize,
+        options: IndentationArgs,
+    ) -> Result<Vec<String>, FunctionCallError> {
+        let anchor_line = options.anchor_line.unwrap_or(offset);
+        if anchor_line == 0 {
+            return Err(FunctionCallError::RespondToModel(
+                "anchor_line must be a 1-indexed line number".to_string(),
+            ));
+        }
+
+        let guard_limit = options.max_lines.unwrap_or(limit);
+        if guard_limit == 0 {
+            return Err(FunctionCallError::RespondToModel(
+                "max_lines must be greater than zero".to_string(),
+            ));
+        }
+
+        let collected = collect_file_lines(path).await?;
+        if collected.is_empty() || anchor_line > collected.len() {
+            return Err(FunctionCallError::RespondToModel(
+                "anchor_line exceeds file length".to_string(),
+            ));
+        }
+
+        let anchor_index = anchor_line - 1;
+        let effective_indents = compute_effective_indents(&collected);
+        let anchor_indent = effective_indents[anchor_index];
+
+        // Compute the min indent
+        let min_indent = if options.max_levels == 0 {
+            0
+        } else {
+            anchor_indent.saturating_sub(options.max_levels * TAB_WIDTH)
+        };
+
+        // Cap requested lines by guard_limit and file length
+        let final_limit = limit.min(guard_limit).min(collected.len());
+
+        if final_limit == 1 {
+            return Ok(vec![format!(
+                "L{}: {}",
+                collected[anchor_index].number, collected[anchor_index].display
+            )]);
+        }
+
+        // Cursors
+        let mut i: isize = anchor_index as isize - 1; // up (inclusive)
+        let mut j: usize = anchor_index + 1; // down (inclusive)
+        let mut i_counter_min_indent = 0;
+        let mut j_counter_min_indent = 0;
+
+        let mut out = VecDeque::with_capacity(limit);
+        out.push_back(&collected[anchor_index]);
+
+        while out.len() < final_limit {
+            let mut progressed = 0;
+
+            // Up.
+            if i >= 0 {
+                let iu = i as usize;
+                if effective_indents[iu] >= min_indent {
+                    out.push_front(&collected[iu]);
+                    progressed += 1;
+                    i -= 1;
+
+                    // We do not include the siblings (not applied to comments).
+                    if effective_indents[iu] == min_indent && !options.include_siblings {
+                        let allow_header_comment =
+                            options.include_header && collected[iu].is_comment();
+                        let can_take_line = allow_header_comment || i_counter_min_indent == 0;
+
+                        if can_take_line {
+                            i_counter_min_indent += 1;
+                        } else {
+                            // This line shouldn't have been taken.
+                            out.pop_front();
+                            progressed -= 1;
+                            i = -1; // consider using Option<usize> or a control flag instead of a sentinel
+                        }
+                    }
+
+                    // Short-cut.
+                    if out.len() >= final_limit {
+                        break;
+                    }
+                } else {
+                    // Stop moving up.
+                    i = -1;
+                }
+            }
+
+            // Down.
+            if j < collected.len() {
+                let ju = j;
+                if effective_indents[ju] >= min_indent {
+                    out.push_back(&collected[ju]);
+                    progressed += 1;
+                    j += 1;
+
+                    // We do not include the siblings (applied to comments).
+                    if effective_indents[ju] == min_indent && !options.include_siblings {
+                        if j_counter_min_indent > 0 {
+                            // This line shouldn't have been taken.
+                            out.pop_back();
+                            progressed -= 1;
+                            j = collected.len();
+                        }
+                        j_counter_min_indent += 1;
+                    }
+                } else {
+                    // Stop moving down.
+                    j = collected.len();
+                }
+            }
+
+            if progressed == 0 {
+                break;
+            }
+        }
+
+        // Trim empty lines
+        trim_empty_lines(&mut out);
+
+        Ok(out
+            .into_iter()
+            .map(|record| format!("L{}: {}", record.number, record.display))
+            .collect())
+    }
+
+    async fn collect_file_lines(path: &Path) -> Result<Vec<LineRecord>, FunctionCallError> {
+        let file = File::open(path).await.map_err(|err| {
+            FunctionCallError::RespondToModel(format!("failed to read file: {err}"))
+        })?;
+
+        let mut reader = BufReader::new(file);
+        let mut buffer = Vec::new();
+        let mut lines = Vec::new();
+        let mut number = 0usize;
+
+        loop {
+            buffer.clear();
+            let bytes_read = reader.read_until(b'\n', &mut buffer).await.map_err(|err| {
+                FunctionCallError::RespondToModel(format!("failed to read file: {err}"))
+            })?;
+
+            if bytes_read == 0 {
+                break;
+            }
+
+            if buffer.last() == Some(&b'\n') {
+                buffer.pop();
+                if buffer.last() == Some(&b'\r') {
+                    buffer.pop();
+                }
+            }
+
+            number += 1;
+            let raw = String::from_utf8_lossy(&buffer).into_owned();
+            let indent = measure_indent(&raw);
+            let display = format_line(&buffer);
+            lines.push(LineRecord {
+                number,
+                raw,
+                display,
+                indent,
+            });
+        }
+
+        Ok(lines)
+    }
+
+    fn compute_effective_indents(records: &[LineRecord]) -> Vec<usize> {
+        let mut effective = Vec::with_capacity(records.len());
+        let mut previous_indent = 0usize;
+        for record in records {
+            if record.is_blank() {
+                effective.push(previous_indent);
+            } else {
+                previous_indent = record.indent;
+                effective.push(previous_indent);
+            }
+        }
+        effective
+    }
+
+    fn measure_indent(line: &str) -> usize {
+        line.chars()
+            .take_while(|c| matches!(c, ' ' | '\t'))
+            .map(|c| if c == '\t' { TAB_WIDTH } else { 1 })
+            .sum()
+    }
+}
+
+fn format_line(bytes: &[u8]) -> String {
+    let decoded = String::from_utf8_lossy(bytes);
+    if decoded.len() > MAX_LINE_LENGTH {
+        take_bytes_at_char_boundary(&decoded, MAX_LINE_LENGTH).to_string()
+    } else {
+        decoded.into_owned()
+    }
+}
+
+fn trim_empty_lines(out: &mut VecDeque<&LineRecord>) {
+    while matches!(out.front(), Some(line) if line.raw.trim().is_empty()) {
+        out.pop_front();
+    }
+    while matches!(out.back(), Some(line) if line.raw.trim().is_empty()) {
+        out.pop_back();
+    }
+}
+
+mod defaults {
+    use super::*;
+
+    impl Default for IndentationArgs {
+        fn default() -> Self {
+            Self {
+                anchor_line: None,
+                max_levels: max_levels(),
+                include_siblings: include_siblings(),
+                include_header: include_header(),
+                max_lines: None,
+            }
+        }
+    }
+
+    impl Default for ReadMode {
+        fn default() -> Self {
+            Self::Slice
+        }
+    }
+
+    pub fn offset() -> usize {
+        1
+    }
+
+    pub fn limit() -> usize {
+        2000
+    }
+
+    pub fn max_levels() -> usize {
+        0
+    }
+
+    pub fn include_siblings() -> bool {
+        false
+    }
+
+    pub fn include_header() -> bool {
+        true
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::indentation::read_block;
+    use super::slice::read;
+    use super::*;
+    use pretty_assertions::assert_eq;
+    use tempfile::NamedTempFile;
+
+    #[tokio::test]
+    async fn reads_requested_range() -> anyhow::Result<()> {
+        let mut temp = NamedTempFile::new()?;
+        use std::io::Write as _;
+        write!(
+            temp,
+            "alpha
+beta
+gamma
+"
+        )?;
+
+        let lines = read(temp.path(), 2, 2).await?;
+        assert_eq!(lines, vec!["L2: beta".to_string(), "L3: gamma".to_string()]);
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn errors_when_offset_exceeds_length() -> anyhow::Result<()> {
+        let mut temp = NamedTempFile::new()?;
+        use std::io::Write as _;
+        writeln!(temp, "only")?;
+
+        let err = read(temp.path(), 3, 1)
+            .await
+            .expect_err("offset exceeds length");
+        assert_eq!(
+            err,
+            FunctionCallError::RespondToModel("offset exceeds file length".to_string())
+        );
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn reads_non_utf8_lines() -> anyhow::Result<()> {
+        let mut temp = NamedTempFile::new()?;
+        use std::io::Write as _;
+        temp.as_file_mut().write_all(b"\xff\xfe\nplain\n")?;
+
+        let lines = read(temp.path(), 1, 2).await?;
+        let expected_first = format!("L1: {}{}", '\u{FFFD}', '\u{FFFD}');
+        assert_eq!(lines, vec![expected_first, "L2: plain".to_string()]);
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn trims_crlf_endings() -> anyhow::Result<()> {
+        let mut temp = NamedTempFile::new()?;
+        use std::io::Write as _;
+        write!(temp, "one\r\ntwo\r\n")?;
+
+        let lines = read(temp.path(), 1, 2).await?;
+        assert_eq!(lines, vec!["L1: one".to_string(), "L2: two".to_string()]);
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn respects_limit_even_with_more_lines() -> anyhow::Result<()> {
+        let mut temp = NamedTempFile::new()?;
+        use std::io::Write as _;
+        write!(
+            temp,
+            "first
+second
+third
+"
+        )?;
+
+        let lines = read(temp.path(), 1, 2).await?;
+        assert_eq!(
+            lines,
+            vec!["L1: first".to_string(), "L2: second".to_string()]
+        );
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn truncates_lines_longer_than_max_length() -> anyhow::Result<()> {
+        let mut temp = NamedTempFile::new()?;
+        use std::io::Write as _;
+        let long_line = "x".repeat(MAX_LINE_LENGTH + 50);
+        writeln!(temp, "{long_line}")?;
+
+        let lines = read(temp.path(), 1, 1).await?;
+        let expected = "x".repeat(MAX_LINE_LENGTH);
+        assert_eq!(lines, vec![format!("L1: {expected}")]);
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn indentation_mode_captures_block() -> anyhow::Result<()> {
+        let mut temp = NamedTempFile::new()?;
+        use std::io::Write as _;
+        write!(
+            temp,
+            "fn outer() {{
+    if cond {{
+        inner();
+    }}
+    tail();
+}}
+"
+        )?;
+
+        let options = IndentationArgs {
+            anchor_line: Some(3),
+            include_siblings: false,
+            max_levels: 1,
+            ..Default::default()
+        };
+
+        let lines = read_block(temp.path(), 3, 10, options).await?;
+
+        assert_eq!(
+            lines,
+            vec![
+                "L2:     if cond {".to_string(),
+                "L3:         inner();".to_string(),
+                "L4:     }".to_string()
+            ]
+        );
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn indentation_mode_expands_parents() -> anyhow::Result<()> {
+        let mut temp = NamedTempFile::new()?;
+        use std::io::Write as _;
+        write!(
+            temp,
+            "mod root {{
+    fn outer() {{
+        if cond {{
+            inner();
+        }}
+    }}
+}}
+"
+        )?;
+
+        let mut options = IndentationArgs {
+            anchor_line: Some(4),
+            max_levels: 2,
+            ..Default::default()
+        };
+
+        let lines = read_block(temp.path(), 4, 50, options.clone()).await?;
+        assert_eq!(
+            lines,
+            vec![
+                "L2:     fn outer() {".to_string(),
+                "L3:         if cond {".to_string(),
+                "L4:             inner();".to_string(),
+                "L5:         }".to_string(),
+                "L6:     }".to_string(),
+            ]
+        );
+
+        options.max_levels = 3;
+        let expanded = read_block(temp.path(), 4, 50, options).await?;
+        assert_eq!(
+            expanded,
+            vec![
+                "L1: mod root {".to_string(),
+                "L2:     fn outer() {".to_string(),
+                "L3:         if cond {".to_string(),
+                "L4:             inner();".to_string(),
+                "L5:         }".to_string(),
+                "L6:     }".to_string(),
+                "L7: }".to_string(),
+            ]
+        );
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn indentation_mode_respects_sibling_flag() -> anyhow::Result<()> {
+        let mut temp = NamedTempFile::new()?;
+        use std::io::Write as _;
+        write!(
+            temp,
+            "fn wrapper() {{
+    if first {{
+        do_first();
+    }}
+    if second {{
+        do_second();
+    }}
+}}
+"
+        )?;
+
+        let mut options = IndentationArgs {
+            anchor_line: Some(3),
+            include_siblings: false,
+            max_levels: 1,
+            ..Default::default()
+        };
+
+        let lines = read_block(temp.path(), 3, 50, options.clone()).await?;
+        assert_eq!(
+            lines,
+            vec![
+                "L2:     if first {".to_string(),
+                "L3:         do_first();".to_string(),
+                "L4:     }".to_string(),
+            ]
+        );
+
+        options.include_siblings = true;
+        let with_siblings = read_block(temp.path(), 3, 50, options).await?;
+        assert_eq!(
+            with_siblings,
+            vec![
+                "L2:     if first {".to_string(),
+                "L3:         do_first();".to_string(),
+                "L4:     }".to_string(),
+                "L5:     if second {".to_string(),
+                "L6:         do_second();".to_string(),
+                "L7:     }".to_string(),
+            ]
+        );
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn indentation_mode_handles_python_sample() -> anyhow::Result<()> {
+        let mut temp = NamedTempFile::new()?;
+        use std::io::Write as _;
+        write!(
+            temp,
+            "class Foo:
+    def __init__(self, size):
+        self.size = size
+    def double(self, value):
+        if value is None:
+            return 0
+        result = value * self.size
+        return result
+class Bar:
+    def compute(self):
+        helper = Foo(2)
+        return helper.double(5)
+"
+        )?;
+
+        let options = IndentationArgs {
+            anchor_line: Some(7),
+            include_siblings: true,
+            max_levels: 1,
+            ..Default::default()
+        };
+
+        let lines = read_block(temp.path(), 1, 200, options).await?;
+        assert_eq!(
+            lines,
+            vec![
+                "L2:     def __init__(self, size):".to_string(),
+                "L3:         self.size = size".to_string(),
+                "L4:     def double(self, value):".to_string(),
+                "L5:         if value is None:".to_string(),
+                "L6:             return 0".to_string(),
+                "L7:         result = value * self.size".to_string(),
+                "L8:         return result".to_string(),
+            ]
+        );
+        Ok(())
+    }
+
+    #[tokio::test]
+    #[ignore]
+    async fn indentation_mode_handles_javascript_sample() -> anyhow::Result<()> {
+        let mut temp = NamedTempFile::new()?;
+        use std::io::Write as _;
+        write!(
+            temp,
+            "export function makeThing() {{
+    const cache = new Map();
+    function ensure(key) {{
+        if (!cache.has(key)) {{
+            cache.set(key, []);
+        }}
+        return cache.get(key);
+    }}
+    const handlers = {{
+        init() {{
+            console.log(\"init\");
+        }},
+        run() {{
+            if (Math.random() > 0.5) {{
+                return \"heads\";
+            }}
+            return \"tails\";
+        }},
+    }};
+    return {{ cache, handlers }};
+}}
+export function other() {{
+    return makeThing();
+}}
+"
+        )?;
+
+        let options = IndentationArgs {
+            anchor_line: Some(15),
+            max_levels: 1,
+            ..Default::default()
+        };
+
+        let lines = read_block(temp.path(), 15, 200, options).await?;
+        assert_eq!(
+            lines,
+            vec![
+                "L10:         init() {".to_string(),
+                "L11:             console.log(\"init\");".to_string(),
+                "L12:         },".to_string(),
+                "L13:         run() {".to_string(),
+                "L14:             if (Math.random() > 0.5) {".to_string(),
+                "L15:                 return \"heads\";".to_string(),
+                "L16:             }".to_string(),
+                "L17:             return \"tails\";".to_string(),
+                "L18:         },".to_string(),
+            ]
+        );
+        Ok(())
+    }
+
+    fn write_cpp_sample() -> anyhow::Result<NamedTempFile> {
+        let mut temp = NamedTempFile::new()?;
+        use std::io::Write as _;
+        write!(
+            temp,
+            "#include <vector>
+#include <string>
+
+namespace sample {{
+class Runner {{
+public:
+    void setup() {{
+        if (enabled_) {{
+            init();
+        }}
+    }}
+
+    // Run the code
+    int run() const {{
+        switch (mode_) {{
+            case Mode::Fast:
+                return fast();
+            case Mode::Slow:
+                return slow();
+            default:
+                return fallback();
+        }}
+    }}
+
+private:
+    bool enabled_ = false;
+    Mode mode_ = Mode::Fast;
+
+    int fast() const {{
+        return 1;
+    }}
+}};
+}}  // namespace sample
+"
+        )?;
+        Ok(temp)
+    }
+
+    #[tokio::test]
+    async fn indentation_mode_handles_cpp_sample_shallow() -> anyhow::Result<()> {
+        let temp = write_cpp_sample()?;
+
+        let options = IndentationArgs {
+            include_siblings: false,
+            anchor_line: Some(18),
+            max_levels: 1,
+            ..Default::default()
+        };
+
+        let lines = read_block(temp.path(), 18, 200, options).await?;
+        assert_eq!(
+            lines,
+            vec![
+                "L15:         switch (mode_) {".to_string(),
+                "L16:             case Mode::Fast:".to_string(),
+                "L17:                 return fast();".to_string(),
+                "L18:             case Mode::Slow:".to_string(),
+                "L19:                 return slow();".to_string(),
+                "L20:             default:".to_string(),
+                "L21:                 return fallback();".to_string(),
+                "L22:         }".to_string(),
+            ]
+        );
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn indentation_mode_handles_cpp_sample() -> anyhow::Result<()> {
+        let temp = write_cpp_sample()?;
+
+        let options = IndentationArgs {
+            include_siblings: false,
+            anchor_line: Some(18),
+            max_levels: 2,
+            ..Default::default()
+        };
+
+        let lines = read_block(temp.path(), 18, 200, options).await?;
+        assert_eq!(
+            lines,
+            vec![
+                "L13:     // Run the code".to_string(),
+                "L14:     int run() const {".to_string(),
+                "L15:         switch (mode_) {".to_string(),
+                "L16:             case Mode::Fast:".to_string(),
+                "L17:                 return fast();".to_string(),
+                "L18:             case Mode::Slow:".to_string(),
+                "L19:                 return slow();".to_string(),
+                "L20:             default:".to_string(),
+                "L21:                 return fallback();".to_string(),
+                "L22:         }".to_string(),
+                "L23:     }".to_string(),
+            ]
+        );
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn indentation_mode_handles_cpp_sample_no_headers() -> anyhow::Result<()> {
+        let temp = write_cpp_sample()?;
+
+        let options = IndentationArgs {
+            include_siblings: false,
+            include_header: false,
+            anchor_line: Some(18),
+            max_levels: 2,
+            ..Default::default()
+        };
+
+        let lines = read_block(temp.path(), 18, 200, options).await?;
+        assert_eq!(
+            lines,
+            vec![
+                "L14:     int run() const {".to_string(),
+                "L15:         switch (mode_) {".to_string(),
+                "L16:             case Mode::Fast:".to_string(),
+                "L17:                 return fast();".to_string(),
+                "L18:             case Mode::Slow:".to_string(),
+                "L19:                 return slow();".to_string(),
+                "L20:             default:".to_string(),
+                "L21:                 return fallback();".to_string(),
+                "L22:         }".to_string(),
+                "L23:     }".to_string(),
+            ]
+        );
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn indentation_mode_handles_cpp_sample_siblings() -> anyhow::Result<()> {
+        let temp = write_cpp_sample()?;
+
+        let options = IndentationArgs {
+            include_siblings: true,
+            include_header: false,
+            anchor_line: Some(18),
+            max_levels: 2,
+            ..Default::default()
+        };
+
+        let lines = read_block(temp.path(), 18, 200, options).await?;
+        assert_eq!(
+            lines,
+            vec![
+                "L7:     void setup() {".to_string(),
+                "L8:         if (enabled_) {".to_string(),
+                "L9:             init();".to_string(),
+                "L10:         }".to_string(),
+                "L11:     }".to_string(),
+                "L12: ".to_string(),
+                "L13:     // Run the code".to_string(),
+                "L14:     int run() const {".to_string(),
+                "L15:         switch (mode_) {".to_string(),
+                "L16:             case Mode::Fast:".to_string(),
+                "L17:                 return fast();".to_string(),
+                "L18:             case Mode::Slow:".to_string(),
+                "L19:                 return slow();".to_string(),
+                "L20:             default:".to_string(),
+                "L21:                 return fallback();".to_string(),
+                "L22:         }".to_string(),
+                "L23:     }".to_string(),
+            ]
+        );
+        Ok(())
+    }
+}

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

@@ -0,0 +1,101 @@
+use async_trait::async_trait;
+use codex_protocol::models::ShellToolCallParams;
+use std::sync::Arc;
+
+use crate::codex::TurnContext;
+use crate::exec::ExecParams;
+use crate::exec_env::create_env;
+use crate::function_tool::FunctionCallError;
+use crate::tools::context::ToolInvocation;
+use crate::tools::context::ToolOutput;
+use crate::tools::context::ToolPayload;
+use crate::tools::handle_container_exec_with_params;
+use crate::tools::registry::ToolHandler;
+use crate::tools::registry::ToolKind;
+
+pub struct ShellHandler;
+
+impl ShellHandler {
+    fn to_exec_params(params: ShellToolCallParams, turn_context: &TurnContext) -> ExecParams {
+        ExecParams {
+            command: params.command,
+            cwd: turn_context.resolve_path(params.workdir.clone()),
+            timeout_ms: params.timeout_ms,
+            env: create_env(&turn_context.shell_environment_policy),
+            with_escalated_permissions: params.with_escalated_permissions,
+            justification: params.justification,
+        }
+    }
+}
+
+#[async_trait]
+impl ToolHandler for ShellHandler {
+    fn kind(&self) -> ToolKind {
+        ToolKind::Function
+    }
+
+    fn matches_kind(&self, payload: &ToolPayload) -> bool {
+        matches!(
+            payload,
+            ToolPayload::Function { .. } | ToolPayload::LocalShell { .. }
+        )
+    }
+
+    async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError> {
+        let ToolInvocation {
+            session,
+            turn,
+            tracker,
+            sub_id,
+            call_id,
+            tool_name,
+            payload,
+        } = invocation;
+
+        match payload {
+            ToolPayload::Function { arguments } => {
+                let params: ShellToolCallParams =
+                    serde_json::from_str(&arguments).map_err(|e| {
+                        FunctionCallError::RespondToModel(format!(
+                            "failed to parse function arguments: {e:?}"
+                        ))
+                    })?;
+                let exec_params = Self::to_exec_params(params, turn.as_ref());
+                let content = handle_container_exec_with_params(
+                    tool_name.as_str(),
+                    exec_params,
+                    Arc::clone(&session),
+                    Arc::clone(&turn),
+                    Arc::clone(&tracker),
+                    sub_id.clone(),
+                    call_id.clone(),
+                )
+                .await?;
+                Ok(ToolOutput::Function {
+                    content,
+                    success: Some(true),
+                })
+            }
+            ToolPayload::LocalShell { params } => {
+                let exec_params = Self::to_exec_params(params, turn.as_ref());
+                let content = handle_container_exec_with_params(
+                    tool_name.as_str(),
+                    exec_params,
+                    Arc::clone(&session),
+                    Arc::clone(&turn),
+                    Arc::clone(&tracker),
+                    sub_id.clone(),
+                    call_id.clone(),
+                )
+                .await?;
+                Ok(ToolOutput::Function {
+                    content,
+                    success: Some(true),
+                })
+            }
+            _ => Err(FunctionCallError::RespondToModel(format!(
+                "unsupported payload for shell handler: {tool_name}"
+            ))),
+        }
+    }
+}

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

@@ -0,0 +1,158 @@
+use std::collections::HashMap;
+use std::collections::hash_map::Entry;
+use std::sync::Arc;
+use std::sync::OnceLock;
+use std::time::Duration;
+
+use async_trait::async_trait;
+use serde::Deserialize;
+use tokio::sync::Barrier;
+use tokio::time::sleep;
+
+use crate::function_tool::FunctionCallError;
+use crate::tools::context::ToolInvocation;
+use crate::tools::context::ToolOutput;
+use crate::tools::context::ToolPayload;
+use crate::tools::registry::ToolHandler;
+use crate::tools::registry::ToolKind;
+
+pub struct TestSyncHandler;
+
+const DEFAULT_TIMEOUT_MS: u64 = 1_000;
+
+static BARRIERS: OnceLock<tokio::sync::Mutex<HashMap<String, BarrierState>>> = OnceLock::new();
+
+struct BarrierState {
+    barrier: Arc<Barrier>,
+    participants: usize,
+}
+
+#[derive(Debug, Deserialize)]
+struct BarrierArgs {
+    id: String,
+    participants: usize,
+    #[serde(default = "default_timeout_ms")]
+    timeout_ms: u64,
+}
+
+#[derive(Debug, Deserialize)]
+struct TestSyncArgs {
+    #[serde(default)]
+    sleep_before_ms: Option<u64>,
+    #[serde(default)]
+    sleep_after_ms: Option<u64>,
+    #[serde(default)]
+    barrier: Option<BarrierArgs>,
+}
+
+fn default_timeout_ms() -> u64 {
+    DEFAULT_TIMEOUT_MS
+}
+
+fn barrier_map() -> &'static tokio::sync::Mutex<HashMap<String, BarrierState>> {
+    BARRIERS.get_or_init(|| tokio::sync::Mutex::new(HashMap::new()))
+}
+
+#[async_trait]
+impl ToolHandler for TestSyncHandler {
+    fn kind(&self) -> ToolKind {
+        ToolKind::Function
+    }
+
+    async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError> {
+        let ToolInvocation { payload, .. } = invocation;
+
+        let arguments = match payload {
+            ToolPayload::Function { arguments } => arguments,
+            _ => {
+                return Err(FunctionCallError::RespondToModel(
+                    "test_sync_tool handler received unsupported payload".to_string(),
+                ));
+            }
+        };
+
+        let args: TestSyncArgs = serde_json::from_str(&arguments).map_err(|err| {
+            FunctionCallError::RespondToModel(format!(
+                "failed to parse function arguments: {err:?}"
+            ))
+        })?;
+
+        if let Some(delay) = args.sleep_before_ms
+            && delay > 0
+        {
+            sleep(Duration::from_millis(delay)).await;
+        }
+
+        if let Some(barrier) = args.barrier {
+            wait_on_barrier(barrier).await?;
+        }
+
+        if let Some(delay) = args.sleep_after_ms
+            && delay > 0
+        {
+            sleep(Duration::from_millis(delay)).await;
+        }
+
+        Ok(ToolOutput::Function {
+            content: "ok".to_string(),
+            success: Some(true),
+        })
+    }
+}
+
+async fn wait_on_barrier(args: BarrierArgs) -> Result<(), FunctionCallError> {
+    if args.participants == 0 {
+        return Err(FunctionCallError::RespondToModel(
+            "barrier participants must be greater than zero".to_string(),
+        ));
+    }
+
+    if args.timeout_ms == 0 {
+        return Err(FunctionCallError::RespondToModel(
+            "barrier timeout must be greater than zero".to_string(),
+        ));
+    }
+
+    let barrier_id = args.id.clone();
+    let barrier = {
+        let mut map = barrier_map().lock().await;
+        match map.entry(barrier_id.clone()) {
+            Entry::Occupied(entry) => {
+                let state = entry.get();
+                if state.participants != args.participants {
+                    let existing = state.participants;
+                    return Err(FunctionCallError::RespondToModel(format!(
+                        "barrier {barrier_id} already registered with {existing} participants"
+                    )));
+                }
+                state.barrier.clone()
+            }
+            Entry::Vacant(entry) => {
+                let barrier = Arc::new(Barrier::new(args.participants));
+                entry.insert(BarrierState {
+                    barrier: barrier.clone(),
+                    participants: args.participants,
+                });
+                barrier
+            }
+        }
+    };
+
+    let timeout = Duration::from_millis(args.timeout_ms);
+    let wait_result = tokio::time::timeout(timeout, barrier.wait())
+        .await
+        .map_err(|_| {
+            FunctionCallError::RespondToModel("test_sync_tool barrier wait timed out".to_string())
+        })?;
+
+    if wait_result.is_leader() {
+        let mut map = barrier_map().lock().await;
+        if let Some(state) = map.get(&barrier_id)
+            && Arc::ptr_eq(&state.barrier, &barrier)
+        {
+            map.remove(&barrier_id);
+        }
+    }
+
+    Ok(())
+}

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

@@ -0,0 +1,109 @@
+use async_trait::async_trait;
+use serde::Deserialize;
+
+use crate::function_tool::FunctionCallError;
+use crate::tools::context::ToolInvocation;
+use crate::tools::context::ToolOutput;
+use crate::tools::context::ToolPayload;
+use crate::tools::registry::ToolHandler;
+use crate::tools::registry::ToolKind;
+use crate::unified_exec::UnifiedExecRequest;
+
+pub struct UnifiedExecHandler;
+
+#[derive(Deserialize)]
+struct UnifiedExecArgs {
+    input: Vec<String>,
+    #[serde(default)]
+    session_id: Option<String>,
+    #[serde(default)]
+    timeout_ms: Option<u64>,
+}
+
+#[async_trait]
+impl ToolHandler for UnifiedExecHandler {
+    fn kind(&self) -> ToolKind {
+        ToolKind::UnifiedExec
+    }
+
+    fn matches_kind(&self, payload: &ToolPayload) -> bool {
+        matches!(
+            payload,
+            ToolPayload::UnifiedExec { .. } | ToolPayload::Function { .. }
+        )
+    }
+
+    async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError> {
+        let ToolInvocation {
+            session, payload, ..
+        } = invocation;
+
+        let args = match payload {
+            ToolPayload::UnifiedExec { arguments } | ToolPayload::Function { arguments } => {
+                serde_json::from_str::<UnifiedExecArgs>(&arguments).map_err(|err| {
+                    FunctionCallError::RespondToModel(format!(
+                        "failed to parse function arguments: {err:?}"
+                    ))
+                })?
+            }
+            _ => {
+                return Err(FunctionCallError::RespondToModel(
+                    "unified_exec handler received unsupported payload".to_string(),
+                ));
+            }
+        };
+
+        let UnifiedExecArgs {
+            input,
+            session_id,
+            timeout_ms,
+        } = args;
+
+        let parsed_session_id = if let Some(session_id) = session_id {
+            match session_id.parse::<i32>() {
+                Ok(parsed) => Some(parsed),
+                Err(output) => {
+                    return Err(FunctionCallError::RespondToModel(format!(
+                        "invalid session_id: {session_id} due to error {output:?}"
+                    )));
+                }
+            }
+        } else {
+            None
+        };
+
+        let request = UnifiedExecRequest {
+            session_id: parsed_session_id,
+            input_chunks: &input,
+            timeout_ms,
+        };
+
+        let value = session
+            .run_unified_exec_request(request)
+            .await
+            .map_err(|err| {
+                FunctionCallError::RespondToModel(format!("unified exec failed: {err:?}"))
+            })?;
+
+        #[derive(serde::Serialize)]
+        struct SerializedUnifiedExecResult {
+            session_id: Option<String>,
+            output: String,
+        }
+
+        let content = serde_json::to_string(&SerializedUnifiedExecResult {
+            session_id: value.session_id.map(|id| id.to_string()),
+            output: value.output,
+        })
+        .map_err(|err| {
+            FunctionCallError::RespondToModel(format!(
+                "failed to serialize unified exec output: {err:?}"
+            ))
+        })?;
+
+        Ok(ToolOutput::Function {
+            content,
+            success: Some(true),
+        })
+    }
+}

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

@@ -0,0 +1,93 @@
+use async_trait::async_trait;
+use serde::Deserialize;
+use tokio::fs;
+
+use crate::function_tool::FunctionCallError;
+use crate::protocol::Event;
+use crate::protocol::EventMsg;
+use crate::protocol::InputItem;
+use crate::protocol::ViewImageToolCallEvent;
+use crate::tools::context::ToolInvocation;
+use crate::tools::context::ToolOutput;
+use crate::tools::context::ToolPayload;
+use crate::tools::registry::ToolHandler;
+use crate::tools::registry::ToolKind;
+
+pub struct ViewImageHandler;
+
+#[derive(Deserialize)]
+struct ViewImageArgs {
+    path: String,
+}
+
+#[async_trait]
+impl ToolHandler for ViewImageHandler {
+    fn kind(&self) -> ToolKind {
+        ToolKind::Function
+    }
+
+    async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError> {
+        let ToolInvocation {
+            session,
+            turn,
+            payload,
+            sub_id,
+            call_id,
+            ..
+        } = invocation;
+
+        let arguments = match payload {
+            ToolPayload::Function { arguments } => arguments,
+            _ => {
+                return Err(FunctionCallError::RespondToModel(
+                    "view_image handler received unsupported payload".to_string(),
+                ));
+            }
+        };
+
+        let args: ViewImageArgs = serde_json::from_str(&arguments).map_err(|e| {
+            FunctionCallError::RespondToModel(format!("failed to parse function arguments: {e:?}"))
+        })?;
+
+        let abs_path = turn.resolve_path(Some(args.path));
+
+        let metadata = fs::metadata(&abs_path).await.map_err(|error| {
+            FunctionCallError::RespondToModel(format!(
+                "unable to locate image at `{}`: {error}",
+                abs_path.display()
+            ))
+        })?;
+
+        if !metadata.is_file() {
+            return Err(FunctionCallError::RespondToModel(format!(
+                "image path `{}` is not a file",
+                abs_path.display()
+            )));
+        }
+        let event_path = abs_path.clone();
+
+        session
+            .inject_input(vec![InputItem::LocalImage { path: abs_path }])
+            .await
+            .map_err(|_| {
+                FunctionCallError::RespondToModel(
+                    "unable to attach image (no active task)".to_string(),
+                )
+            })?;
+
+        session
+            .send_event(Event {
+                id: sub_id.to_string(),
+                msg: EventMsg::ViewImageToolCall(ViewImageToolCallEvent {
+                    call_id,
+                    path: event_path,
+                }),
+            })
+            .await;
+
+        Ok(ToolOutput::Function {
+            content: "attached local image path".to_string(),
+            success: Some(true),
+        })
+    }
+}

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

@@ -0,0 +1,387 @@
+pub mod context;
+pub(crate) mod handlers;
+pub mod parallel;
+pub mod registry;
+pub mod router;
+pub mod spec;
+
+use crate::apply_patch;
+use crate::apply_patch::ApplyPatchExec;
+use crate::apply_patch::InternalApplyPatchInvocation;
+use crate::apply_patch::convert_apply_patch_to_protocol;
+use crate::codex::Session;
+use crate::codex::TurnContext;
+use crate::error::CodexErr;
+use crate::error::SandboxErr;
+use crate::exec::ExecParams;
+use crate::exec::ExecToolCallOutput;
+use crate::exec::StdoutStream;
+use crate::executor::ExecutionMode;
+use crate::executor::errors::ExecError;
+use crate::executor::linkers::PreparedExec;
+use crate::function_tool::FunctionCallError;
+use crate::tools::context::ApplyPatchCommandContext;
+use crate::tools::context::ExecCommandContext;
+use crate::tools::context::SharedTurnDiffTracker;
+use codex_apply_patch::MaybeApplyPatchVerified;
+use codex_apply_patch::maybe_parse_apply_patch_verified;
+use codex_protocol::protocol::AskForApproval;
+use codex_utils_string::take_bytes_at_char_boundary;
+use codex_utils_string::take_last_bytes_at_char_boundary;
+pub use router::ToolRouter;
+use serde::Serialize;
+use std::sync::Arc;
+use tracing::trace;
+
+// 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;
+
+// Telemetry preview limits: keep log events smaller than model budgets.
+pub(crate) const TELEMETRY_PREVIEW_MAX_BYTES: usize = 2 * 1024; // 2 KiB
+pub(crate) const TELEMETRY_PREVIEW_MAX_LINES: usize = 64; // lines
+pub(crate) const TELEMETRY_PREVIEW_TRUNCATION_NOTICE: &str =
+    "[... telemetry preview truncated ...]";
+
+// TODO(jif) break this down
+pub(crate) async fn handle_container_exec_with_params(
+    tool_name: &str,
+    params: ExecParams,
+    sess: Arc<Session>,
+    turn_context: Arc<TurnContext>,
+    turn_diff_tracker: SharedTurnDiffTracker,
+    sub_id: String,
+    call_id: String,
+) -> Result<String, FunctionCallError> {
+    let otel_event_manager = turn_context.client.get_otel_event_manager();
+
+    if params.with_escalated_permissions.unwrap_or(false)
+        && !matches!(turn_context.approval_policy, AskForApproval::OnRequest)
+    {
+        return Err(FunctionCallError::RespondToModel(format!(
+            "approval policy is {policy:?}; reject command — you should not ask for escalated permissions if the approval policy is {policy:?}",
+            policy = turn_context.approval_policy
+        )));
+    }
+
+    // check if this was a patch, and apply it if so
+    let apply_patch_exec = match maybe_parse_apply_patch_verified(&params.command, &params.cwd) {
+        MaybeApplyPatchVerified::Body(changes) => {
+            match apply_patch::apply_patch(
+                sess.as_ref(),
+                turn_context.as_ref(),
+                &sub_id,
+                &call_id,
+                changes,
+            )
+            .await
+            {
+                InternalApplyPatchInvocation::Output(item) => return item,
+                InternalApplyPatchInvocation::DelegateToExec(apply_patch_exec) => {
+                    Some(apply_patch_exec)
+                }
+            }
+        }
+        MaybeApplyPatchVerified::CorrectnessError(parse_error) => {
+            // It looks like an invocation of `apply_patch`, but we
+            // could not resolve it into a patch that would apply
+            // cleanly. Return to model for resample.
+            return Err(FunctionCallError::RespondToModel(format!(
+                "apply_patch verification failed: {parse_error}"
+            )));
+        }
+        MaybeApplyPatchVerified::ShellParseError(error) => {
+            trace!("Failed to parse shell command, {error:?}");
+            None
+        }
+        MaybeApplyPatchVerified::NotApplyPatch => None,
+    };
+
+    let command_for_display = if let Some(exec) = apply_patch_exec.as_ref() {
+        vec!["apply_patch".to_string(), exec.action.patch.clone()]
+    } else {
+        params.command.clone()
+    };
+
+    let exec_command_context = ExecCommandContext {
+        sub_id: sub_id.clone(),
+        call_id: call_id.clone(),
+        command_for_display: command_for_display.clone(),
+        cwd: params.cwd.clone(),
+        apply_patch: apply_patch_exec.as_ref().map(
+            |ApplyPatchExec {
+                 action,
+                 user_explicitly_approved_this_action,
+             }| ApplyPatchCommandContext {
+                user_explicitly_approved_this_action: *user_explicitly_approved_this_action,
+                changes: convert_apply_patch_to_protocol(action),
+            },
+        ),
+        tool_name: tool_name.to_string(),
+        otel_event_manager,
+    };
+
+    let mode = match apply_patch_exec {
+        Some(exec) => ExecutionMode::ApplyPatch(exec),
+        None => ExecutionMode::Shell,
+    };
+
+    sess.services.executor.update_environment(
+        turn_context.sandbox_policy.clone(),
+        turn_context.cwd.clone(),
+    );
+
+    let prepared_exec = PreparedExec::new(
+        exec_command_context,
+        params,
+        command_for_display,
+        mode,
+        Some(StdoutStream {
+            sub_id: sub_id.clone(),
+            call_id: call_id.clone(),
+            tx_event: sess.get_tx_event(),
+        }),
+        turn_context.shell_environment_policy.use_profile,
+    );
+
+    let output_result = sess
+        .run_exec_with_events(
+            turn_diff_tracker.clone(),
+            prepared_exec,
+            turn_context.approval_policy,
+        )
+        .await;
+
+    // always make sure to truncate the output if its length isn't controlled.
+    match output_result {
+        Ok(output) => {
+            let ExecToolCallOutput { exit_code, .. } = &output;
+            let content = format_exec_output_apply_patch(&output);
+            if *exit_code == 0 {
+                Ok(content)
+            } else {
+                Err(FunctionCallError::RespondToModel(content))
+            }
+        }
+        Err(ExecError::Function(err)) => Err(truncate_function_error(err)),
+        Err(ExecError::Codex(CodexErr::Sandbox(SandboxErr::Timeout { output }))) => Err(
+            FunctionCallError::RespondToModel(format_exec_output_apply_patch(&output)),
+        ),
+        Err(ExecError::Codex(err)) => {
+            let message = format!("execution error: {err:?}");
+            Err(FunctionCallError::RespondToModel(format_exec_output(
+                &message,
+            )))
+        }
+    }
+}
+
+pub fn format_exec_output_apply_patch(exec_output: &ExecToolCallOutput) -> String {
+    let ExecToolCallOutput {
+        exit_code,
+        duration,
+        ..
+    } = exec_output;
+
+    #[derive(Serialize)]
+    struct ExecMetadata {
+        exit_code: i32,
+        duration_seconds: f32,
+    }
+
+    #[derive(Serialize)]
+    struct ExecOutput<'a> {
+        output: &'a str,
+        metadata: ExecMetadata,
+    }
+
+    // round to 1 decimal place
+    let duration_seconds = ((duration.as_secs_f32()) * 10.0).round() / 10.0;
+
+    let formatted_output = format_exec_output_str(exec_output);
+
+    let payload = ExecOutput {
+        output: &formatted_output,
+        metadata: ExecMetadata {
+            exit_code: *exit_code,
+            duration_seconds,
+        },
+    };
+
+    #[expect(clippy::expect_used)]
+    serde_json::to_string(&payload).expect("serialize ExecOutput")
+}
+
+pub fn format_exec_output_str(exec_output: &ExecToolCallOutput) -> String {
+    let ExecToolCallOutput {
+        aggregated_output, ..
+    } = exec_output;
+
+    let content = aggregated_output.text.as_str();
+
+    if exec_output.timed_out {
+        let prefixed = format!(
+            "command timed out after {} milliseconds\n{content}",
+            exec_output.duration.as_millis()
+        );
+        return format_exec_output(&prefixed);
+    }
+
+    format_exec_output(content)
+}
+
+fn truncate_function_error(err: FunctionCallError) -> FunctionCallError {
+    match err {
+        FunctionCallError::RespondToModel(msg) => {
+            FunctionCallError::RespondToModel(format_exec_output(&msg))
+        }
+        FunctionCallError::Fatal(msg) => FunctionCallError::Fatal(format_exec_output(&msg)),
+        other => other,
+    }
+}
+
+fn format_exec_output(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 marker = format!("\n[... omitted {omitted} of {total_lines} lines ...]\n\n");
+
+    // Byte budgets for head/tail around the marker
+    let mut head_budget = MODEL_FORMAT_HEAD_BYTES.min(MODEL_FORMAT_MAX_BYTES);
+    let tail_budget = MODEL_FORMAT_MAX_BYTES.saturating_sub(head_budget + marker.len());
+    if tail_budget == 0 && marker.len() >= MODEL_FORMAT_MAX_BYTES {
+        // Degenerate case: marker alone exceeds budget; return a clipped marker
+        return take_bytes_at_char_boundary(&marker, MODEL_FORMAT_MAX_BYTES).to_string();
+    }
+    if tail_budget == 0 {
+        // Make room for the marker by shrinking head
+        head_budget = MODEL_FORMAT_MAX_BYTES.saturating_sub(marker.len());
+    }
+
+    let head_slice = &content[..head_slice_end];
+    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);
+    result.push_str(&marker);
+
+    let remaining = MODEL_FORMAT_MAX_BYTES.saturating_sub(result.len());
+    if remaining == 0 {
+        return result;
+    }
+
+    let tail_slice = &content[tail_slice_start..];
+    let tail_part = take_last_bytes_at_char_boundary(tail_slice, remaining);
+    result.push_str(tail_part);
+
+    result
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use regex_lite::Regex;
+
+    fn assert_truncated_message_matches(message: &str, line: &str, total_lines: usize) {
+        let pattern = truncated_message_pattern(line, total_lines);
+        let regex = Regex::new(&pattern).unwrap_or_else(|err| {
+            panic!("failed to compile regex {pattern}: {err}");
+        });
+        let captures = regex
+            .captures(message)
+            .unwrap_or_else(|| panic!("message failed to match pattern {pattern}: {message}"));
+        let body = captures
+            .name("body")
+            .expect("missing body capture")
+            .as_str();
+        assert!(
+            body.len() <= MODEL_FORMAT_MAX_BYTES,
+            "body exceeds byte limit: {} bytes",
+            body.len()
+        );
+    }
+
+    fn truncated_message_pattern(line: &str, total_lines: usize) -> String {
+        let head_take = MODEL_FORMAT_HEAD_LINES.min(total_lines);
+        let tail_take = MODEL_FORMAT_TAIL_LINES.min(total_lines.saturating_sub(head_take));
+        let omitted = total_lines.saturating_sub(head_take + tail_take);
+        let escaped_line = regex_lite::escape(line);
+        format!(
+            r"(?s)^Total output lines: {total_lines}\n\n(?P<body>{escaped_line}.*\n\[\.{{3}} omitted {omitted} of {total_lines} lines \.{{3}}]\n\n.*)$",
+        )
+    }
+
+    #[test]
+    fn truncate_formatted_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 = format_exec_output(&large_error);
+
+        let total_lines = large_error.lines().count();
+        assert_truncated_message_matches(&truncated, line, total_lines);
+        assert_ne!(truncated, large_error);
+    }
+
+    #[test]
+    fn truncate_function_error_trims_respond_to_model() {
+        let line = "respond-to-model error that should be truncated\n";
+        let huge = line.repeat(3_000);
+        let total_lines = huge.lines().count();
+
+        let err = truncate_function_error(FunctionCallError::RespondToModel(huge));
+        match err {
+            FunctionCallError::RespondToModel(message) => {
+                assert_truncated_message_matches(&message, line, total_lines);
+            }
+            other => panic!("unexpected error variant: {other:?}"),
+        }
+    }
+
+    #[test]
+    fn truncate_function_error_trims_fatal() {
+        let line = "fatal error output that should be truncated\n";
+        let huge = line.repeat(3_000);
+        let total_lines = huge.lines().count();
+
+        let err = truncate_function_error(FunctionCallError::Fatal(huge));
+        match err {
+            FunctionCallError::Fatal(message) => {
+                assert_truncated_message_matches(&message, line, total_lines);
+            }
+            other => panic!("unexpected error variant: {other:?}"),
+        }
+    }
+}

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

@@ -0,0 +1,80 @@
+use std::sync::Arc;
+
+use tokio::sync::RwLock;
+use tokio_util::either::Either;
+use tokio_util::task::AbortOnDropHandle;
+
+use crate::codex::Session;
+use crate::codex::TurnContext;
+use crate::error::CodexErr;
+use crate::function_tool::FunctionCallError;
+use crate::tools::context::SharedTurnDiffTracker;
+use crate::tools::router::ToolCall;
+use crate::tools::router::ToolRouter;
+use codex_protocol::models::ResponseInputItem;
+
+pub(crate) struct ToolCallRuntime {
+    router: Arc<ToolRouter>,
+    session: Arc<Session>,
+    turn_context: Arc<TurnContext>,
+    tracker: SharedTurnDiffTracker,
+    sub_id: String,
+    parallel_execution: Arc<RwLock<()>>,
+}
+
+impl ToolCallRuntime {
+    pub(crate) fn new(
+        router: Arc<ToolRouter>,
+        session: Arc<Session>,
+        turn_context: Arc<TurnContext>,
+        tracker: SharedTurnDiffTracker,
+        sub_id: String,
+    ) -> Self {
+        Self {
+            router,
+            session,
+            turn_context,
+            tracker,
+            sub_id,
+            parallel_execution: Arc::new(RwLock::new(())),
+        }
+    }
+
+    pub(crate) fn handle_tool_call(
+        &self,
+        call: ToolCall,
+    ) -> impl std::future::Future<Output = Result<ResponseInputItem, CodexErr>> {
+        let supports_parallel = self.router.tool_supports_parallel(&call.tool_name);
+
+        let router = Arc::clone(&self.router);
+        let session = Arc::clone(&self.session);
+        let turn = Arc::clone(&self.turn_context);
+        let tracker = Arc::clone(&self.tracker);
+        let sub_id = self.sub_id.clone();
+        let lock = Arc::clone(&self.parallel_execution);
+
+        let handle: AbortOnDropHandle<Result<ResponseInputItem, FunctionCallError>> =
+            AbortOnDropHandle::new(tokio::spawn(async move {
+                let _guard = if supports_parallel {
+                    Either::Left(lock.read().await)
+                } else {
+                    Either::Right(lock.write().await)
+                };
+
+                router
+                    .dispatch_tool_call(session, turn, tracker, sub_id, call)
+                    .await
+            }));
+
+        async move {
+            match handle.await {
+                Ok(Ok(response)) => Ok(response),
+                Ok(Err(FunctionCallError::Fatal(message))) => Err(CodexErr::Fatal(message)),
+                Ok(Err(other)) => Err(CodexErr::Fatal(other.to_string())),
+                Err(err) => Err(CodexErr::Fatal(format!(
+                    "tool task failed to receive: {err:?}"
+                ))),
+            }
+        }
+    }
+}