remove unecessary comment

use tree-sitter-highlight instead of custom logic over the tree-sitter
tree to highlight bash.

.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/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

@@ -27,7 +27,7 @@ jobs:
       - name: Install dependencies
         run: pnpm install --frozen-lockfile
 
-      # build_npm_package.py requires DotSlash when staging releases.
+      # stage_npm_packages.py requires DotSlash when staging releases.
       - uses: facebook/install-dotslash@v2
 
       - name: Stage npm package
@@ -37,10 +37,12 @@ jobs:
         run: |
           set -euo pipefail
           CODEX_VERSION=0.40.0
-          PACK_OUTPUT="${RUNNER_TEMP}/codex-npm.tgz"
-          python3 ./codex-cli/scripts/build_npm_package.py \
+          OUTPUT_DIR="${RUNNER_TEMP}"
+          python3 ./scripts/stage_npm_packages.py \
             --release-version "$CODEX_VERSION" \
-            --pack-output "$PACK_OUTPUT"
+            --package codex \
+            --output-dir "$OUTPUT_DIR"
+          PACK_OUTPUT="${OUTPUT_DIR}/codex-npm-${CODEX_VERSION}.tgz"
           echo "pack_output=$PACK_OUTPUT" >> "$GITHUB_OUTPUT"
 
       - name: Upload staged npm package artifact
@@ -58,3 +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-release.yml

@@ -216,31 +216,30 @@ jobs:
             echo "npm_tag=" >> "$GITHUB_OUTPUT"
           fi
 
-      # build_npm_package.py requires DotSlash when staging releases.
-      - uses: facebook/install-dotslash@v2
-      - name: Stage codex CLI npm package
-        env:
-          GH_TOKEN: ${{ github.token }}
-        run: |
-          set -euo pipefail
-          TMP_DIR="${RUNNER_TEMP}/npm-stage"
-          ./codex-cli/scripts/build_npm_package.py \
-            --package codex \
-            --release-version "${{ steps.release_name.outputs.name }}" \
-            --staging-dir "${TMP_DIR}" \
-            --pack-output "${GITHUB_WORKSPACE}/dist/npm/codex-npm-${{ steps.release_name.outputs.name }}.tgz"
+      - name: Setup pnpm
+        uses: pnpm/action-setup@v4
+        with:
+          run_install: false
 
-      - name: Stage responses API proxy npm package
+      - name: Setup Node.js for npm packaging
+        uses: actions/setup-node@v5
+        with:
+          node-version: 22
+
+      - name: Install dependencies
+        run: pnpm install --frozen-lockfile
+
+      # stage_npm_packages.py requires DotSlash when staging releases.
+      - uses: facebook/install-dotslash@v2
+      - name: Stage npm packages
         env:
           GH_TOKEN: ${{ github.token }}
         run: |
-          set -euo pipefail
-          TMP_DIR="${RUNNER_TEMP}/npm-stage-responses"
-          ./codex-cli/scripts/build_npm_package.py \
-            --package codex-responses-api-proxy \
+          ./scripts/stage_npm_packages.py \
             --release-version "${{ steps.release_name.outputs.name }}" \
-            --staging-dir "${TMP_DIR}" \
-            --pack-output "${GITHUB_WORKSPACE}/dist/npm/codex-responses-api-proxy-npm-${{ steps.release_name.outputs.name }}.tgz"
+            --package codex \
+            --package codex-responses-api-proxy \
+            --package codex-sdk
 
       - name: Create GitHub Release
         uses: softprops/action-gh-release@v2
@@ -300,6 +299,10 @@ jobs:
             --repo "${GITHUB_REPOSITORY}" \
             --pattern "codex-responses-api-proxy-npm-${version}.tgz" \
             --dir dist/npm
+          gh release download "$tag" \
+            --repo "${GITHUB_REPOSITORY}" \
+            --pattern "codex-sdk-npm-${version}.tgz" \
+            --dir dist/npm
 
       # No NODE_AUTH_TOKEN needed because we use OIDC.
       - name: Publish to npm
@@ -316,6 +319,7 @@ jobs:
           tarballs=(
             "codex-npm-${VERSION}.tgz"
             "codex-responses-api-proxy-npm-${VERSION}.tgz"
+            "codex-sdk-npm-${VERSION}.tgz"
           )
 
           for tarball in "${tarballs[@]}"; do

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/scripts/README.md

@@ -1,11 +1,19 @@
 # npm releases
 
-Run the following:
-
-To build the 0.2.x or later version of the npm module, which runs the Rust version of the CLI, build it as follows:
+Use the staging helper in the repo root to generate npm tarballs for a release. For
+example, to stage the CLI, responses proxy, and SDK packages for version `0.6.0`:
 
 ```bash
-./codex-cli/scripts/build_npm_package.py --release-version 0.6.0
+./scripts/stage_npm_packages.py \
+  --release-version 0.6.0 \
+  --package codex \
+  --package codex-responses-api-proxy \
+  --package codex-sdk
 ```
 
-Note this will create `./codex-cli/vendor/` as a side-effect.
+This downloads the native artifacts once, hydrates `vendor/` for each package, and writes
+tarballs to `dist/npm/`.
+
+If you need to invoke `build_npm_package.py` directly, run
+`codex-cli/scripts/install_native_deps.py` first and pass `--vendor-src` pointing to the
+directory that contains the populated `vendor/` tree.

codex-cli/scripts/build_npm_package.py

@@ -3,7 +3,6 @@
 
 import argparse
 import json
-import re
 import shutil
 import subprocess
 import sys
@@ -14,19 +13,25 @@ SCRIPT_DIR = Path(__file__).resolve().parent
 CODEX_CLI_ROOT = SCRIPT_DIR.parent
 REPO_ROOT = CODEX_CLI_ROOT.parent
 RESPONSES_API_PROXY_NPM_ROOT = REPO_ROOT / "codex-rs" / "responses-api-proxy" / "npm"
-GITHUB_REPO = "openai/codex"
+CODEX_SDK_ROOT = REPO_ROOT / "sdk" / "typescript"
 
-# The docs are not clear on what the expected value/format of
-# workflow/workflowName is:
-# https://cli.github.com/manual/gh_run_list
-WORKFLOW_NAME = ".github/workflows/rust-release.yml"
+PACKAGE_NATIVE_COMPONENTS: dict[str, list[str]] = {
+    "codex": ["codex", "rg"],
+    "codex-responses-api-proxy": ["codex-responses-api-proxy"],
+    "codex-sdk": ["codex"],
+}
+COMPONENT_DEST_DIR: dict[str, str] = {
+    "codex": "codex",
+    "codex-responses-api-proxy": "codex-responses-api-proxy",
+    "rg": "path",
+}
 
 
 def parse_args() -> argparse.Namespace:
     parser = argparse.ArgumentParser(description="Build or stage the Codex CLI npm package.")
     parser.add_argument(
         "--package",
-        choices=("codex", "codex-responses-api-proxy"),
+        choices=("codex", "codex-responses-api-proxy", "codex-sdk"),
         default="codex",
         help="Which npm package to stage (default: codex).",
     )
@@ -37,14 +42,9 @@ def parse_args() -> argparse.Namespace:
     parser.add_argument(
         "--release-version",
         help=(
-            "Version to stage for npm release. When provided, the script also resolves the "
-            "matching rust-release workflow unless --workflow-url is supplied."
+            "Version to stage for npm release."
         ),
     )
-    parser.add_argument(
-        "--workflow-url",
-        help="Optional GitHub Actions workflow run URL used to download native binaries.",
-    )
     parser.add_argument(
         "--staging-dir",
         type=Path,
@@ -64,6 +64,11 @@ def parse_args() -> argparse.Namespace:
         type=Path,
         help="Path where the generated npm tarball should be written.",
     )
+    parser.add_argument(
+        "--vendor-src",
+        type=Path,
+        help="Directory containing pre-installed native binaries to bundle (vendor root).",
+    )
     return parser.parse_args()
 
 
@@ -86,29 +91,19 @@ def main() -> int:
     try:
         stage_sources(staging_dir, version, package)
 
-        workflow_url = args.workflow_url
-        resolved_head_sha: str | None = None
-        if not workflow_url:
-            if release_version:
-                workflow = resolve_release_workflow(version)
-                workflow_url = workflow["url"]
-                resolved_head_sha = workflow.get("headSha")
-            else:
-                workflow_url = resolve_latest_alpha_workflow_url()
-        elif release_version:
-            try:
-                workflow = resolve_release_workflow(version)
-                resolved_head_sha = workflow.get("headSha")
-            except Exception:
-                resolved_head_sha = None
+        vendor_src = args.vendor_src.resolve() if args.vendor_src else None
+        native_components = PACKAGE_NATIVE_COMPONENTS.get(package, [])
 
-        if release_version and resolved_head_sha:
-            print(f"should `git checkout {resolved_head_sha}`")
+        if native_components:
+            if vendor_src is None:
+                components_str = ", ".join(native_components)
+                raise RuntimeError(
+                    "Native components "
+                    f"({components_str}) required for package '{package}'. Provide --vendor-src "
+                    "pointing to a directory containing pre-installed binaries."
+                )
 
-        if not workflow_url:
-            raise RuntimeError("Unable to determine workflow URL for native binaries.")
-
-        install_native_binaries(staging_dir, workflow_url, package)
+            copy_native_binaries(vendor_src, staging_dir, native_components)
 
         if release_version:
             staging_dir_str = str(staging_dir)
@@ -119,12 +114,20 @@ def main() -> int:
                     f"    node {staging_dir_str}/bin/codex.js --version\n"
                     f"    node {staging_dir_str}/bin/codex.js --help\n\n"
                 )
-            else:
+            elif package == "codex-responses-api-proxy":
                 print(
                     f"Staged version {version} for release in {staging_dir_str}\n\n"
                     "Verify the responses API proxy:\n"
                     f"    node {staging_dir_str}/bin/codex-responses-api-proxy.js --help\n\n"
                 )
+            else:
+                print(
+                    f"Staged version {version} for release in {staging_dir_str}\n\n"
+                    "Verify the SDK contents:\n"
+                    f"    ls {staging_dir_str}/dist\n"
+                    f"    ls {staging_dir_str}/vendor\n"
+                    "    node -e \"import('./dist/index.js').then(() => console.log('ok'))\"\n\n"
+                )
         else:
             print(f"Staged package in {staging_dir}")
 
@@ -152,10 +155,9 @@ def prepare_staging_dir(staging_dir: Path | None) -> tuple[Path, bool]:
 
 
 def stage_sources(staging_dir: Path, version: str, package: str) -> None:
-    bin_dir = staging_dir / "bin"
-    bin_dir.mkdir(parents=True, exist_ok=True)
-
     if package == "codex":
+        bin_dir = staging_dir / "bin"
+        bin_dir.mkdir(parents=True, exist_ok=True)
         shutil.copy2(CODEX_CLI_ROOT / "bin" / "codex.js", bin_dir / "codex.js")
         rg_manifest = CODEX_CLI_ROOT / "bin" / "rg"
         if rg_manifest.exists():
@@ -167,6 +169,8 @@ def stage_sources(staging_dir: Path, version: str, package: str) -> None:
 
         package_json_path = CODEX_CLI_ROOT / "package.json"
     elif package == "codex-responses-api-proxy":
+        bin_dir = staging_dir / "bin"
+        bin_dir.mkdir(parents=True, exist_ok=True)
         launcher_src = RESPONSES_API_PROXY_NPM_ROOT / "bin" / "codex-responses-api-proxy.js"
         shutil.copy2(launcher_src, bin_dir / "codex-responses-api-proxy.js")
 
@@ -175,6 +179,9 @@ def stage_sources(staging_dir: Path, version: str, package: str) -> None:
             shutil.copy2(readme_src, staging_dir / "README.md")
 
         package_json_path = RESPONSES_API_PROXY_NPM_ROOT / "package.json"
+    elif package == "codex-sdk":
+        package_json_path = CODEX_SDK_ROOT / "package.json"
+        stage_codex_sdk_sources(staging_dir)
     else:
         raise RuntimeError(f"Unknown package '{package}'.")
 
@@ -182,91 +189,85 @@ def stage_sources(staging_dir: Path, version: str, package: str) -> None:
         package_json = json.load(fh)
     package_json["version"] = version
 
+    if package == "codex-sdk":
+        scripts = package_json.get("scripts")
+        if isinstance(scripts, dict):
+            scripts.pop("prepare", None)
+
+        files = package_json.get("files")
+        if isinstance(files, list):
+            if "vendor" not in files:
+                files.append("vendor")
+        else:
+            package_json["files"] = ["dist", "vendor"]
+
     with open(staging_dir / "package.json", "w", encoding="utf-8") as out:
         json.dump(package_json, out, indent=2)
         out.write("\n")
 
 
-def install_native_binaries(staging_dir: Path, workflow_url: str, package: str) -> None:
-    package_components = {
-        "codex": ["codex", "rg"],
-        "codex-responses-api-proxy": ["codex-responses-api-proxy"],
-    }
-
-    components = package_components.get(package)
-    if components is None:
-        raise RuntimeError(f"Unknown package '{package}'.")
-
-    cmd = ["./scripts/install_native_deps.py", "--workflow-url", workflow_url]
-    for component in components:
-        cmd.extend(["--component", component])
-    cmd.append(str(staging_dir))
-    subprocess.check_call(cmd, cwd=CODEX_CLI_ROOT)
+def run_command(cmd: list[str], cwd: Path | None = None) -> None:
+    print("+", " ".join(cmd))
+    subprocess.run(cmd, cwd=cwd, check=True)
 
 
-def resolve_latest_alpha_workflow_url() -> str:
-    version = determine_latest_alpha_version()
-    workflow = resolve_release_workflow(version)
-    return workflow["url"]
+def stage_codex_sdk_sources(staging_dir: Path) -> None:
+    package_root = CODEX_SDK_ROOT
+
+    run_command(["pnpm", "install", "--frozen-lockfile"], cwd=package_root)
+    run_command(["pnpm", "run", "build"], cwd=package_root)
+
+    dist_src = package_root / "dist"
+    if not dist_src.exists():
+        raise RuntimeError("codex-sdk build did not produce a dist directory.")
+
+    shutil.copytree(dist_src, staging_dir / "dist")
+
+    readme_src = package_root / "README.md"
+    if readme_src.exists():
+        shutil.copy2(readme_src, staging_dir / "README.md")
+
+    license_src = REPO_ROOT / "LICENSE"
+    if license_src.exists():
+        shutil.copy2(license_src, staging_dir / "LICENSE")
 
 
-def determine_latest_alpha_version() -> str:
-    releases = list_releases()
-    best_key: tuple[int, int, int, int] | None = None
-    best_version: str | None = None
-    pattern = re.compile(r"^rust-v(\d+)\.(\d+)\.(\d+)-alpha\.(\d+)$")
-    for release in releases:
-        tag = release.get("tag_name", "")
-        match = pattern.match(tag)
-        if not match:
+def copy_native_binaries(vendor_src: Path, staging_dir: Path, components: list[str]) -> None:
+    vendor_src = vendor_src.resolve()
+    if not vendor_src.exists():
+        raise RuntimeError(f"Vendor source directory not found: {vendor_src}")
+
+    components_set = {component for component in components if component in COMPONENT_DEST_DIR}
+    if not components_set:
+        return
+
+    vendor_dest = staging_dir / "vendor"
+    if vendor_dest.exists():
+        shutil.rmtree(vendor_dest)
+    vendor_dest.mkdir(parents=True, exist_ok=True)
+
+    for target_dir in vendor_src.iterdir():
+        if not target_dir.is_dir():
             continue
-        key = tuple(int(match.group(i)) for i in range(1, 5))
-        if best_key is None or key > best_key:
-            best_key = key
-            best_version = (
-                f"{match.group(1)}.{match.group(2)}.{match.group(3)}-alpha.{match.group(4)}"
-            )
 
-    if best_version is None:
-        raise RuntimeError("No alpha releases found when resolving workflow URL.")
-    return best_version
+        dest_target_dir = vendor_dest / target_dir.name
+        dest_target_dir.mkdir(parents=True, exist_ok=True)
 
+        for component in components_set:
+            dest_dir_name = COMPONENT_DEST_DIR.get(component)
+            if dest_dir_name is None:
+                continue
 
-def list_releases() -> list[dict]:
-    stdout = subprocess.check_output(
-        ["gh", "api", f"/repos/{GITHUB_REPO}/releases?per_page=100"],
-        text=True,
-    )
-    try:
-        releases = json.loads(stdout or "[]")
-    except json.JSONDecodeError as exc:
-        raise RuntimeError("Unable to parse releases JSON.") from exc
-    if not isinstance(releases, list):
-        raise RuntimeError("Unexpected response when listing releases.")
-    return releases
+            src_component_dir = target_dir / dest_dir_name
+            if not src_component_dir.exists():
+                raise RuntimeError(
+                    f"Missing native component '{component}' in vendor source: {src_component_dir}"
+                )
 
-
-def resolve_release_workflow(version: str) -> dict:
-    stdout = subprocess.check_output(
-        [
-            "gh",
-            "run",
-            "list",
-            "--branch",
-            f"rust-v{version}",
-            "--json",
-            "workflowName,url,headSha",
-            "--workflow",
-            WORKFLOW_NAME,
-            "--jq",
-            "first(.[])",
-        ],
-        text=True,
-    )
-    workflow = json.loads(stdout or "[]")
-    if not workflow:
-        raise RuntimeError(f"Unable to find rust-release workflow for version {version}.")
-    return workflow
+            dest_component_dir = dest_target_dir / dest_dir_name
+            if dest_component_dir.exists():
+                shutil.rmtree(dest_component_dir)
+            shutil.copytree(src_component_dir, dest_component_dir)
 
 
 def run_npm_pack(staging_dir: Path, output_path: Path) -> Path:

codex-rs/Cargo.toml

@@ -3,6 +3,7 @@ members = [
     "backend-client",
     "ansi-escape",
     "app-server",
+    "app-server-protocol",
     "apply-patch",
     "arg0",
     "codex-backend-openapi-models",
@@ -31,6 +32,7 @@ members = [
     "git-apply",
     "utils/json-to-toml",
     "utils/readiness",
+    "utils/string",
 ]
 resolver = "2"
 
@@ -47,6 +49,7 @@ edition = "2024"
 app_test_support = { path = "app-server/tests/common" }
 codex-ansi-escape = { path = "ansi-escape" }
 codex-app-server = { path = "app-server" }
+codex-app-server-protocol = { path = "app-server-protocol" }
 codex-apply-patch = { path = "apply-patch" }
 codex-arg0 = { path = "arg0" }
 codex-chatgpt = { path = "chatgpt" }
@@ -69,6 +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" }
@@ -79,10 +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"
@@ -95,11 +101,12 @@ derive_more = "2"
 diffy = "0.4.2"
 dirs = "6"
 dotenvy = "0.15.7"
+dunce = "1.0.4"
 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"
@@ -107,6 +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"
@@ -123,6 +131,7 @@ opentelemetry-semantic-conventions = "0.30.0"
 opentelemetry_sdk = "0.30.0"
 os_info = "3.12.0"
 owo-colors = "4.2.0"
+paste = "1.0.15"
 path-absolutize = "3.1.1"
 path-clean = "1.0.1"
 pathdiff = "0.2"
@@ -134,11 +143,13 @@ 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"
@@ -164,8 +175,9 @@ 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"
@@ -233,5 +245,9 @@ 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/app-server-protocol/Cargo.toml

@@ -0,0 +1,24 @@
+[package]
+edition = "2024"
+name = "codex-app-server-protocol"
+version = { workspace = true }
+
+[lib]
+name = "codex_app_server_protocol"
+path = "src/lib.rs"
+
+[lints]
+workspace = true
+
+[dependencies]
+codex-protocol = { workspace = true }
+paste = { workspace = true }
+serde = { workspace = true, features = ["derive"] }
+serde_json = { workspace = true }
+strum_macros = { workspace = true }
+ts-rs = { workspace = true }
+uuid = { workspace = true, features = ["serde", "v7"] }
+
+[dev-dependencies]
+anyhow = { workspace = true }
+pretty_assertions = { workspace = true }

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

@@ -0,0 +1,67 @@
+//! We do not do true JSON-RPC 2.0, as we neither send nor expect the
+//! "jsonrpc": "2.0" field.
+
+use serde::Deserialize;
+use serde::Serialize;
+use ts_rs::TS;
+
+pub const JSONRPC_VERSION: &str = "2.0";
+
+#[derive(Debug, Clone, PartialEq, Deserialize, Serialize, Hash, Eq, TS)]
+#[serde(untagged)]
+pub enum RequestId {
+    String(String),
+    #[ts(type = "number")]
+    Integer(i64),
+}
+
+pub type Result = serde_json::Value;
+
+/// Refers to any valid JSON-RPC object that can be decoded off the wire, or encoded to be sent.
+#[derive(Debug, Clone, PartialEq, Deserialize, Serialize, TS)]
+#[serde(untagged)]
+pub enum JSONRPCMessage {
+    Request(JSONRPCRequest),
+    Notification(JSONRPCNotification),
+    Response(JSONRPCResponse),
+    Error(JSONRPCError),
+}
+
+/// A request that expects a response.
+#[derive(Debug, Clone, PartialEq, Deserialize, Serialize, TS)]
+pub struct JSONRPCRequest {
+    pub id: RequestId,
+    pub method: String,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub params: Option<serde_json::Value>,
+}
+
+/// A notification which does not expect a response.
+#[derive(Debug, Clone, PartialEq, Deserialize, Serialize, TS)]
+pub struct JSONRPCNotification {
+    pub method: String,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub params: Option<serde_json::Value>,
+}
+
+/// A successful (non-error) response to a request.
+#[derive(Debug, Clone, PartialEq, Deserialize, Serialize, TS)]
+pub struct JSONRPCResponse {
+    pub id: RequestId,
+    pub result: Result,
+}
+
+/// A response to a request that indicates an error occurred.
+#[derive(Debug, Clone, PartialEq, Deserialize, Serialize, TS)]
+pub struct JSONRPCError {
+    pub error: JSONRPCErrorError,
+    pub id: RequestId,
+}
+
+#[derive(Debug, Clone, PartialEq, Deserialize, Serialize, TS)]
+pub struct JSONRPCErrorError {
+    pub code: i64,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub data: Option<serde_json::Value>,
+    pub message: String,
+}

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

@@ -0,0 +1,5 @@
+mod jsonrpc_lite;
+mod protocol;
+
+pub use jsonrpc_lite::*;
+pub use protocol::*;

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

@@ -1,77 +1,27 @@
 use std::collections::HashMap;
-use std::fmt::Display;
 use std::path::PathBuf;
 
-use crate::config_types::ReasoningEffort;
-use crate::config_types::ReasoningSummary;
-use crate::config_types::SandboxMode;
-use crate::config_types::Verbosity;
-use crate::protocol::AskForApproval;
-use crate::protocol::EventMsg;
-use crate::protocol::FileChange;
-use crate::protocol::ReviewDecision;
-use crate::protocol::SandboxPolicy;
-use crate::protocol::TurnAbortReason;
-use mcp_types::JSONRPCNotification;
-use mcp_types::RequestId;
+use crate::JSONRPCNotification;
+use crate::JSONRPCRequest;
+use crate::RequestId;
+use codex_protocol::ConversationId;
+use codex_protocol::config_types::ReasoningEffort;
+use codex_protocol::config_types::ReasoningSummary;
+use codex_protocol::config_types::SandboxMode;
+use codex_protocol::config_types::Verbosity;
+use codex_protocol::protocol::AskForApproval;
+use codex_protocol::protocol::EventMsg;
+use codex_protocol::protocol::FileChange;
+use codex_protocol::protocol::ReviewDecision;
+use codex_protocol::protocol::SandboxPolicy;
+use codex_protocol::protocol::TurnAbortReason;
+use paste::paste;
 use serde::Deserialize;
 use serde::Serialize;
 use strum_macros::Display;
 use ts_rs::TS;
 use uuid::Uuid;
 
-#[derive(Debug, Clone, Copy, PartialEq, Eq, TS, Hash)]
-#[ts(type = "string")]
-pub struct ConversationId {
-    uuid: Uuid,
-}
-
-impl ConversationId {
-    pub fn new() -> Self {
-        Self {
-            uuid: Uuid::now_v7(),
-        }
-    }
-
-    pub fn from_string(s: &str) -> Result<Self, uuid::Error> {
-        Ok(Self {
-            uuid: Uuid::parse_str(s)?,
-        })
-    }
-}
-
-impl Default for ConversationId {
-    fn default() -> Self {
-        Self::new()
-    }
-}
-
-impl Display for ConversationId {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        write!(f, "{}", self.uuid)
-    }
-}
-
-impl Serialize for ConversationId {
-    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
-    where
-        S: serde::Serializer,
-    {
-        serializer.collect_str(&self.uuid)
-    }
-}
-
-impl<'de> Deserialize<'de> for ConversationId {
-    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
-    where
-        D: serde::Deserializer<'de>,
-    {
-        let value = String::deserialize(deserializer)?;
-        let uuid = Uuid::parse_str(&value).map_err(serde::de::Error::custom)?;
-        Ok(Self { uuid })
-    }
-}
-
 #[derive(Serialize, Deserialize, Clone, Debug, PartialEq, TS)]
 #[ts(type = "string")]
 pub struct GitSha(pub String);
@@ -89,117 +39,137 @@ pub enum AuthMode {
     ChatGPT,
 }
 
-/// Request from the client to the server.
-#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, TS)]
-#[serde(tag = "method", rename_all = "camelCase")]
-pub enum ClientRequest {
+/// Generates an `enum ClientRequest` where each variant is a request that the
+/// client can send to the server. Each variant has associated `params` and
+/// `response` types. Also generates a `export_client_responses()` function to
+/// export all response types to TypeScript.
+macro_rules! client_request_definitions {
+    (
+        $(
+            $(#[$variant_meta:meta])*
+            $variant:ident {
+                params: $(#[$params_meta:meta])* $params:ty,
+                response: $response:ty,
+            }
+        ),* $(,)?
+    ) => {
+        /// Request from the client to the server.
+        #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, TS)]
+        #[serde(tag = "method", rename_all = "camelCase")]
+        pub enum ClientRequest {
+            $(
+                $(#[$variant_meta])*
+                $variant {
+                    #[serde(rename = "id")]
+                    request_id: RequestId,
+                    $(#[$params_meta])*
+                    params: $params,
+                },
+            )*
+        }
+
+        pub fn export_client_responses(
+            out_dir: &::std::path::Path,
+        ) -> ::std::result::Result<(), ::ts_rs::ExportError> {
+            $(
+                <$response as ::ts_rs::TS>::export_all_to(out_dir)?;
+            )*
+            Ok(())
+        }
+    };
+}
+
+client_request_definitions! {
     Initialize {
-        #[serde(rename = "id")]
-        request_id: RequestId,
         params: InitializeParams,
+        response: InitializeResponse,
     },
     NewConversation {
-        #[serde(rename = "id")]
-        request_id: RequestId,
         params: NewConversationParams,
+        response: NewConversationResponse,
     },
     /// List recorded Codex conversations (rollouts) with optional pagination and search.
     ListConversations {
-        #[serde(rename = "id")]
-        request_id: RequestId,
         params: ListConversationsParams,
+        response: ListConversationsResponse,
     },
     /// Resume a recorded Codex conversation from a rollout file.
     ResumeConversation {
-        #[serde(rename = "id")]
-        request_id: RequestId,
         params: ResumeConversationParams,
+        response: ResumeConversationResponse,
     },
     ArchiveConversation {
-        #[serde(rename = "id")]
-        request_id: RequestId,
         params: ArchiveConversationParams,
+        response: ArchiveConversationResponse,
     },
     SendUserMessage {
-        #[serde(rename = "id")]
-        request_id: RequestId,
         params: SendUserMessageParams,
+        response: SendUserMessageResponse,
     },
     SendUserTurn {
-        #[serde(rename = "id")]
-        request_id: RequestId,
         params: SendUserTurnParams,
+        response: SendUserTurnResponse,
     },
     InterruptConversation {
-        #[serde(rename = "id")]
-        request_id: RequestId,
         params: InterruptConversationParams,
+        response: InterruptConversationResponse,
     },
     AddConversationListener {
-        #[serde(rename = "id")]
-        request_id: RequestId,
         params: AddConversationListenerParams,
+        response: AddConversationSubscriptionResponse,
     },
     RemoveConversationListener {
-        #[serde(rename = "id")]
-        request_id: RequestId,
         params: RemoveConversationListenerParams,
+        response: RemoveConversationSubscriptionResponse,
     },
     GitDiffToRemote {
-        #[serde(rename = "id")]
-        request_id: RequestId,
         params: GitDiffToRemoteParams,
+        response: GitDiffToRemoteResponse,
     },
     LoginApiKey {
-        #[serde(rename = "id")]
-        request_id: RequestId,
         params: LoginApiKeyParams,
+        response: LoginApiKeyResponse,
     },
     LoginChatGpt {
-        #[serde(rename = "id")]
-        request_id: RequestId,
+        params: #[ts(type = "undefined")] #[serde(skip_serializing_if = "Option::is_none")] Option<()>,
+        response: LoginChatGptResponse,
     },
     CancelLoginChatGpt {
-        #[serde(rename = "id")]
-        request_id: RequestId,
         params: CancelLoginChatGptParams,
+        response: CancelLoginChatGptResponse,
     },
     LogoutChatGpt {
-        #[serde(rename = "id")]
-        request_id: RequestId,
+        params: #[ts(type = "undefined")] #[serde(skip_serializing_if = "Option::is_none")] Option<()>,
+        response: LogoutChatGptResponse,
     },
     GetAuthStatus {
-        #[serde(rename = "id")]
-        request_id: RequestId,
         params: GetAuthStatusParams,
+        response: GetAuthStatusResponse,
     },
     GetUserSavedConfig {
-        #[serde(rename = "id")]
-        request_id: RequestId,
+        params: #[ts(type = "undefined")] #[serde(skip_serializing_if = "Option::is_none")] Option<()>,
+        response: GetUserSavedConfigResponse,
     },
     SetDefaultModel {
-        #[serde(rename = "id")]
-        request_id: RequestId,
         params: SetDefaultModelParams,
+        response: SetDefaultModelResponse,
     },
     GetUserAgent {
-        #[serde(rename = "id")]
-        request_id: RequestId,
+        params: #[ts(type = "undefined")] #[serde(skip_serializing_if = "Option::is_none")] Option<()>,
+        response: GetUserAgentResponse,
     },
     UserInfo {
-        #[serde(rename = "id")]
-        request_id: RequestId,
+        params: #[ts(type = "undefined")] #[serde(skip_serializing_if = "Option::is_none")] Option<()>,
+        response: UserInfoResponse,
     },
     FuzzyFileSearch {
-        #[serde(rename = "id")]
-        request_id: RequestId,
         params: FuzzyFileSearchParams,
+        response: FuzzyFileSearchResponse,
     },
     /// Execute a command (argv vector) under the server's sandbox.
     ExecOneOffCommand {
-        #[serde(rename = "id")]
-        request_id: RequestId,
         params: ExecOneOffCommandParams,
+        response: ExecOneOffCommandResponse,
     },
 }
 
@@ -429,7 +399,7 @@ pub struct ExecOneOffCommandParams {
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, TS)]
 #[serde(rename_all = "camelCase")]
-pub struct ExecArbitraryCommandResponse {
+pub struct ExecOneOffCommandResponse {
     pub exit_code: i32,
     pub stdout: String,
     pub stderr: String,
@@ -633,30 +603,74 @@ pub enum InputItem {
     },
 }
 
-// TODO(mbolin): Need test to ensure these constants match the enum variants.
+/// Generates an `enum ServerRequest` where each variant is a request that the
+/// server can send to the client along with the corresponding params and
+/// response types. It also generates helper types used by the app/server
+/// infrastructure (payload enum, request constructor, and export helpers).
+macro_rules! server_request_definitions {
+    (
+        $(
+            $(#[$variant_meta:meta])*
+            $variant:ident
+        ),* $(,)?
+    ) => {
+        paste! {
+            /// Request initiated from the server and sent to the client.
+            #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, TS)]
+            #[serde(tag = "method", rename_all = "camelCase")]
+            pub enum ServerRequest {
+                $(
+                    $(#[$variant_meta])*
+                    $variant {
+                        #[serde(rename = "id")]
+                        request_id: RequestId,
+                        params: [<$variant Params>],
+                    },
+                )*
+            }
 
-pub const APPLY_PATCH_APPROVAL_METHOD: &str = "applyPatchApproval";
-pub const EXEC_COMMAND_APPROVAL_METHOD: &str = "execCommandApproval";
+            #[derive(Debug, Clone, PartialEq)]
+            pub enum ServerRequestPayload {
+                $( $variant([<$variant Params>]), )*
+            }
 
-/// Request initiated from the server and sent to the client.
-#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, TS)]
-#[serde(tag = "method", rename_all = "camelCase")]
-pub enum ServerRequest {
+            impl ServerRequestPayload {
+                pub fn request_with_id(self, request_id: RequestId) -> ServerRequest {
+                    match self {
+                        $(Self::$variant(params) => ServerRequest::$variant { request_id, params },)*
+                    }
+                }
+            }
+        }
+
+        pub fn export_server_responses(
+            out_dir: &::std::path::Path,
+        ) -> ::std::result::Result<(), ::ts_rs::ExportError> {
+            paste! {
+                $(<[<$variant Response>] as ::ts_rs::TS>::export_all_to(out_dir)?;)*
+            }
+            Ok(())
+        }
+    };
+}
+
+impl TryFrom<JSONRPCRequest> for ServerRequest {
+    type Error = serde_json::Error;
+
+    fn try_from(value: JSONRPCRequest) -> Result<Self, Self::Error> {
+        serde_json::from_value(serde_json::to_value(value)?)
+    }
+}
+
+server_request_definitions! {
     /// Request to approve a patch.
-    ApplyPatchApproval {
-        #[serde(rename = "id")]
-        request_id: RequestId,
-        params: ApplyPatchApprovalParams,
-    },
+    ApplyPatchApproval,
     /// Request to exec a command.
-    ExecCommandApproval {
-        #[serde(rename = "id")]
-        request_id: RequestId,
-        params: ExecCommandApprovalParams,
-    },
+    ExecCommandApproval,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, TS)]
+#[serde(rename_all = "camelCase")]
 pub struct ApplyPatchApprovalParams {
     pub conversation_id: ConversationId,
     /// Use to correlate this with [codex_core::protocol::PatchApplyBeginEvent]
@@ -673,6 +687,7 @@ pub struct ApplyPatchApprovalParams {
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, TS)]
+#[serde(rename_all = "camelCase")]
 pub struct ExecCommandApprovalParams {
     pub conversation_id: ConversationId,
     /// Use to correlate this with [codex_core::protocol::ExecCommandBeginEvent]
@@ -710,6 +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>>,
@@ -746,6 +762,7 @@ pub struct SessionConfiguredNotification {
     pub history_log_id: u64,
 
     /// Current number of entries in the history log.
+    #[ts(type = "number")]
     pub history_entry_count: usize,
 
     /// Optional initial messages (as events) for resumed sessions.
@@ -842,12 +859,6 @@ mod tests {
         Ok(())
     }
 
-    #[test]
-    fn test_conversation_id_default_is_not_zeroes() {
-        let id = ConversationId::default();
-        assert_ne!(id.uuid, Uuid::nil());
-    }
-
     #[test]
     fn conversation_id_serializes_as_plain_string() -> Result<()> {
         let id = ConversationId::from_string("67e55044-10b1-426f-9247-bb680e5fe0c8")?;
@@ -883,4 +894,39 @@ mod tests {
         );
         Ok(())
     }
+
+    #[test]
+    fn serialize_server_request() -> Result<()> {
+        let conversation_id = ConversationId::from_string("67e55044-10b1-426f-9247-bb680e5fe0c8")?;
+        let params = ExecCommandApprovalParams {
+            conversation_id,
+            call_id: "call-42".to_string(),
+            command: vec!["echo".to_string(), "hello".to_string()],
+            cwd: PathBuf::from("/tmp"),
+            reason: Some("because tests".to_string()),
+        };
+        let request = ServerRequest::ExecCommandApproval {
+            request_id: RequestId::Integer(7),
+            params: params.clone(),
+        };
+
+        assert_eq!(
+            json!({
+                "method": "execCommandApproval",
+                "id": 7,
+                "params": {
+                    "conversationId": "67e55044-10b1-426f-9247-bb680e5fe0c8",
+                    "callId": "call-42",
+                    "command": ["echo", "hello"],
+                    "cwd": "/tmp",
+                    "reason": "because tests",
+                }
+            }),
+            serde_json::to_value(&request)?,
+        );
+
+        let payload = ServerRequestPayload::ExecCommandApproval(params);
+        assert_eq!(payload.request_with_id(RequestId::Integer(7)), request);
+        Ok(())
+    }
 }

codex-rs/app-server/Cargo.toml

@@ -22,10 +22,8 @@ codex-core = { workspace = true }
 codex-file-search = { workspace = true }
 codex-login = { workspace = true }
 codex-protocol = { workspace = true }
+codex-app-server-protocol = { workspace = true }
 codex-utils-json-to-toml = { workspace = true }
-# We should only be using mcp-types for JSON-RPC types: it would be nice to
-# split this out into a separate crate at some point.
-mcp-types = { workspace = true }
 serde = { workspace = true, features = ["derive"] }
 serde_json = { workspace = true }
 tokio = { workspace = true, features = [

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

@@ -3,10 +3,57 @@ use crate::error_code::INVALID_REQUEST_ERROR_CODE;
 use crate::fuzzy_file_search::run_fuzzy_file_search;
 use crate::outgoing_message::OutgoingMessageSender;
 use crate::outgoing_message::OutgoingNotification;
+use codex_app_server_protocol::AddConversationListenerParams;
+use codex_app_server_protocol::AddConversationSubscriptionResponse;
+use codex_app_server_protocol::ApplyPatchApprovalParams;
+use codex_app_server_protocol::ApplyPatchApprovalResponse;
+use codex_app_server_protocol::ArchiveConversationParams;
+use codex_app_server_protocol::ArchiveConversationResponse;
+use codex_app_server_protocol::AuthStatusChangeNotification;
+use codex_app_server_protocol::ClientRequest;
+use codex_app_server_protocol::ConversationSummary;
+use codex_app_server_protocol::ExecCommandApprovalParams;
+use codex_app_server_protocol::ExecCommandApprovalResponse;
+use codex_app_server_protocol::ExecOneOffCommandParams;
+use codex_app_server_protocol::ExecOneOffCommandResponse;
+use codex_app_server_protocol::FuzzyFileSearchParams;
+use codex_app_server_protocol::FuzzyFileSearchResponse;
+use codex_app_server_protocol::GetUserAgentResponse;
+use codex_app_server_protocol::GetUserSavedConfigResponse;
+use codex_app_server_protocol::GitDiffToRemoteResponse;
+use codex_app_server_protocol::InputItem as WireInputItem;
+use codex_app_server_protocol::InterruptConversationParams;
+use codex_app_server_protocol::InterruptConversationResponse;
+use codex_app_server_protocol::JSONRPCErrorError;
+use codex_app_server_protocol::ListConversationsParams;
+use codex_app_server_protocol::ListConversationsResponse;
+use codex_app_server_protocol::LoginApiKeyParams;
+use codex_app_server_protocol::LoginApiKeyResponse;
+use codex_app_server_protocol::LoginChatGptCompleteNotification;
+use codex_app_server_protocol::LoginChatGptResponse;
+use codex_app_server_protocol::NewConversationParams;
+use codex_app_server_protocol::NewConversationResponse;
+use codex_app_server_protocol::RemoveConversationListenerParams;
+use codex_app_server_protocol::RemoveConversationSubscriptionResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::Result as JsonRpcResult;
+use codex_app_server_protocol::ResumeConversationParams;
+use codex_app_server_protocol::SendUserMessageParams;
+use codex_app_server_protocol::SendUserMessageResponse;
+use codex_app_server_protocol::SendUserTurnParams;
+use codex_app_server_protocol::SendUserTurnResponse;
+use codex_app_server_protocol::ServerNotification;
+use codex_app_server_protocol::ServerRequestPayload;
+use codex_app_server_protocol::SessionConfiguredNotification;
+use codex_app_server_protocol::SetDefaultModelParams;
+use codex_app_server_protocol::SetDefaultModelResponse;
+use codex_app_server_protocol::UserInfoResponse;
+use codex_app_server_protocol::UserSavedConfig;
 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;
@@ -36,58 +83,12 @@ use codex_core::protocol::ReviewDecision;
 use codex_login::ServerOptions as LoginServerOptions;
 use codex_login::ShutdownHandle;
 use codex_login::run_login_server;
-use codex_protocol::mcp_protocol::APPLY_PATCH_APPROVAL_METHOD;
-use codex_protocol::mcp_protocol::AddConversationListenerParams;
-use codex_protocol::mcp_protocol::AddConversationSubscriptionResponse;
-use codex_protocol::mcp_protocol::ApplyPatchApprovalParams;
-use codex_protocol::mcp_protocol::ApplyPatchApprovalResponse;
-use codex_protocol::mcp_protocol::ArchiveConversationParams;
-use codex_protocol::mcp_protocol::ArchiveConversationResponse;
-use codex_protocol::mcp_protocol::AuthStatusChangeNotification;
-use codex_protocol::mcp_protocol::ClientRequest;
-use codex_protocol::mcp_protocol::ConversationId;
-use codex_protocol::mcp_protocol::ConversationSummary;
-use codex_protocol::mcp_protocol::EXEC_COMMAND_APPROVAL_METHOD;
-use codex_protocol::mcp_protocol::ExecArbitraryCommandResponse;
-use codex_protocol::mcp_protocol::ExecCommandApprovalParams;
-use codex_protocol::mcp_protocol::ExecCommandApprovalResponse;
-use codex_protocol::mcp_protocol::ExecOneOffCommandParams;
-use codex_protocol::mcp_protocol::FuzzyFileSearchParams;
-use codex_protocol::mcp_protocol::FuzzyFileSearchResponse;
-use codex_protocol::mcp_protocol::GetUserAgentResponse;
-use codex_protocol::mcp_protocol::GetUserSavedConfigResponse;
-use codex_protocol::mcp_protocol::GitDiffToRemoteResponse;
-use codex_protocol::mcp_protocol::InputItem as WireInputItem;
-use codex_protocol::mcp_protocol::InterruptConversationParams;
-use codex_protocol::mcp_protocol::InterruptConversationResponse;
-use codex_protocol::mcp_protocol::ListConversationsParams;
-use codex_protocol::mcp_protocol::ListConversationsResponse;
-use codex_protocol::mcp_protocol::LoginApiKeyParams;
-use codex_protocol::mcp_protocol::LoginApiKeyResponse;
-use codex_protocol::mcp_protocol::LoginChatGptCompleteNotification;
-use codex_protocol::mcp_protocol::LoginChatGptResponse;
-use codex_protocol::mcp_protocol::NewConversationParams;
-use codex_protocol::mcp_protocol::NewConversationResponse;
-use codex_protocol::mcp_protocol::RemoveConversationListenerParams;
-use codex_protocol::mcp_protocol::RemoveConversationSubscriptionResponse;
-use codex_protocol::mcp_protocol::ResumeConversationParams;
-use codex_protocol::mcp_protocol::SendUserMessageParams;
-use codex_protocol::mcp_protocol::SendUserMessageResponse;
-use codex_protocol::mcp_protocol::SendUserTurnParams;
-use codex_protocol::mcp_protocol::SendUserTurnResponse;
-use codex_protocol::mcp_protocol::ServerNotification;
-use codex_protocol::mcp_protocol::SessionConfiguredNotification;
-use codex_protocol::mcp_protocol::SetDefaultModelParams;
-use codex_protocol::mcp_protocol::SetDefaultModelResponse;
-use codex_protocol::mcp_protocol::UserInfoResponse;
-use codex_protocol::mcp_protocol::UserSavedConfig;
+use codex_protocol::ConversationId;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::protocol::InputMessageKind;
 use codex_protocol::protocol::USER_MESSAGE_BEGIN;
 use codex_utils_json_to_toml::json_to_toml;
-use mcp_types::JSONRPCErrorError;
-use mcp_types::RequestId;
 use std::collections::HashMap;
 use std::ffi::OsStr;
 use std::path::PathBuf;
@@ -193,28 +194,43 @@ impl CodexMessageProcessor {
             ClientRequest::LoginApiKey { request_id, params } => {
                 self.login_api_key(request_id, params).await;
             }
-            ClientRequest::LoginChatGpt { request_id } => {
+            ClientRequest::LoginChatGpt {
+                request_id,
+                params: _,
+            } => {
                 self.login_chatgpt(request_id).await;
             }
             ClientRequest::CancelLoginChatGpt { request_id, params } => {
                 self.cancel_login_chatgpt(request_id, params.login_id).await;
             }
-            ClientRequest::LogoutChatGpt { request_id } => {
+            ClientRequest::LogoutChatGpt {
+                request_id,
+                params: _,
+            } => {
                 self.logout_chatgpt(request_id).await;
             }
             ClientRequest::GetAuthStatus { request_id, params } => {
                 self.get_auth_status(request_id, params).await;
             }
-            ClientRequest::GetUserSavedConfig { request_id } => {
+            ClientRequest::GetUserSavedConfig {
+                request_id,
+                params: _,
+            } => {
                 self.get_user_saved_config(request_id).await;
             }
             ClientRequest::SetDefaultModel { request_id, params } => {
                 self.set_default_model(request_id, params).await;
             }
-            ClientRequest::GetUserAgent { request_id } => {
+            ClientRequest::GetUserAgent {
+                request_id,
+                params: _,
+            } => {
                 self.get_user_agent(request_id).await;
             }
-            ClientRequest::UserInfo { request_id } => {
+            ClientRequest::UserInfo {
+                request_id,
+                params: _,
+            } => {
                 self.get_user_info(request_id).await;
             }
             ClientRequest::FuzzyFileSearch { request_id, params } => {
@@ -371,7 +387,7 @@ impl CodexMessageProcessor {
             self.outgoing
                 .send_response(
                     request_id,
-                    codex_protocol::mcp_protocol::CancelLoginChatGptResponse {},
+                    codex_app_server_protocol::CancelLoginChatGptResponse {},
                 )
                 .await;
         } else {
@@ -407,7 +423,7 @@ impl CodexMessageProcessor {
         self.outgoing
             .send_response(
                 request_id,
-                codex_protocol::mcp_protocol::LogoutChatGptResponse {},
+                codex_app_server_protocol::LogoutChatGptResponse {},
             )
             .await;
 
@@ -425,7 +441,7 @@ impl CodexMessageProcessor {
     async fn get_auth_status(
         &self,
         request_id: RequestId,
-        params: codex_protocol::mcp_protocol::GetAuthStatusParams,
+        params: codex_app_server_protocol::GetAuthStatusParams,
     ) {
         let include_token = params.include_token.unwrap_or(false);
         let do_refresh = params.refresh_token.unwrap_or(false);
@@ -440,7 +456,7 @@ impl CodexMessageProcessor {
         let requires_openai_auth = self.config.model_provider.requires_openai_auth;
 
         let response = if !requires_openai_auth {
-            codex_protocol::mcp_protocol::GetAuthStatusResponse {
+            codex_app_server_protocol::GetAuthStatusResponse {
                 auth_method: None,
                 auth_token: None,
                 requires_openai_auth: Some(false),
@@ -460,13 +476,13 @@ impl CodexMessageProcessor {
                             (None, None)
                         }
                     };
-                    codex_protocol::mcp_protocol::GetAuthStatusResponse {
+                    codex_app_server_protocol::GetAuthStatusResponse {
                         auth_method: reported_auth_method,
                         auth_token: token_opt,
                         requires_openai_auth: Some(true),
                     }
                 }
-                None => codex_protocol::mcp_protocol::GetAuthStatusResponse {
+                None => codex_app_server_protocol::GetAuthStatusResponse {
                     auth_method: None,
                     auth_token: None,
                     requires_openai_auth: Some(true),
@@ -484,7 +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 {
@@ -617,7 +633,7 @@ impl CodexMessageProcessor {
             .await
             {
                 Ok(output) => {
-                    let response = ExecArbitraryCommandResponse {
+                    let response = ExecOneOffCommandResponse {
                         exit_code: output.exit_code,
                         stdout: output.stdout.text,
                         stderr: output.stderr.text,
@@ -637,18 +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) => {
@@ -693,6 +710,7 @@ impl CodexMessageProcessor {
             &self.config.codex_home,
             page_size,
             cursor_ref,
+            INTERACTIVE_SESSION_SOURCES,
         )
         .await
         {
@@ -735,7 +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()),
         };
@@ -793,7 +811,7 @@ impl CodexMessageProcessor {
                 });
 
                 // Reply with conversation id + model and initial messages (when present)
-                let response = codex_protocol::mcp_protocol::ResumeConversationResponse {
+                let response = codex_app_server_protocol::ResumeConversationResponse {
                     conversation_id,
                     model: session_configured.model.clone(),
                     initial_messages,
@@ -1253,9 +1271,8 @@ async fn apply_bespoke_event_handling(
                 reason,
                 grant_root,
             };
-            let value = serde_json::to_value(&params).unwrap_or_default();
             let rx = outgoing
-                .send_request(APPLY_PATCH_APPROVAL_METHOD, Some(value))
+                .send_request(ServerRequestPayload::ApplyPatchApproval(params))
                 .await;
             // TODO(mbolin): Enforce a timeout so this task does not live indefinitely?
             tokio::spawn(async move {
@@ -1275,9 +1292,8 @@ async fn apply_bespoke_event_handling(
                 cwd,
                 reason,
             };
-            let value = serde_json::to_value(&params).unwrap_or_default();
             let rx = outgoing
-                .send_request(EXEC_COMMAND_APPROVAL_METHOD, Some(value))
+                .send_request(ServerRequestPayload::ExecCommandApproval(params))
                 .await;
 
             // TODO(mbolin): Enforce a timeout so this task does not live indefinitely?
@@ -1305,7 +1321,7 @@ 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> {
@@ -1343,12 +1359,12 @@ 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(
     event_id: String,
-    receiver: oneshot::Receiver<mcp_types::Result>,
+    receiver: oneshot::Receiver<JsonRpcResult>,
     codex: Arc<CodexConversation>,
 ) {
     let response = receiver.await;
@@ -1390,7 +1406,7 @@ async fn on_patch_approval_response(
 
 async fn on_exec_approval_response(
     event_id: String,
-    receiver: oneshot::Receiver<mcp_types::Result>,
+    receiver: oneshot::Receiver<JsonRpcResult>,
     conversation: Arc<CodexConversation>,
 ) {
     let response = receiver.await;

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

@@ -1,11 +1,12 @@
 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;
 
+use codex_app_server_protocol::FuzzyFileSearchResult;
 use codex_file_search as file_search;
-use codex_protocol::mcp_protocol::FuzzyFileSearchResult;
 use tokio::task::JoinSet;
 use tracing::warn;
 
@@ -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

@@ -8,7 +8,7 @@ use codex_common::CliConfigOverrides;
 use codex_core::config::Config;
 use codex_core::config::ConfigOverrides;
 
-use mcp_types::JSONRPCMessage;
+use codex_app_server_protocol::JSONRPCMessage;
 use tokio::io::AsyncBufReadExt;
 use tokio::io::AsyncWriteExt;
 use tokio::io::BufReader;
@@ -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}"))
         })?;
@@ -111,17 +112,17 @@ pub async fn run_main(
     let stdout_writer_handle = tokio::spawn(async move {
         let mut stdout = io::stdout();
         while let Some(outgoing_message) = outgoing_rx.recv().await {
-            let msg: JSONRPCMessage = outgoing_message.into();
-            match serde_json::to_string(&msg) {
-                Ok(json) => {
+            let Ok(value) = serde_json::to_value(outgoing_message) else {
+                error!("Failed to convert OutgoingMessage to JSON value");
+                continue;
+            };
+            match serde_json::to_string(&value) {
+                Ok(mut json) => {
+                    json.push('\n');
                     if let Err(e) = stdout.write_all(json.as_bytes()).await {
                         error!("Failed to write to stdout: {e}");
                         break;
                     }
-                    if let Err(e) = stdout.write_all(b"\n").await {
-                        error!("Failed to write newline to stdout: {e}");
-                        break;
-                    }
                 }
                 Err(e) => error!("Failed to serialize JSONRPCMessage: {e}"),
             }

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

@@ -3,20 +3,21 @@ use std::path::PathBuf;
 use crate::codex_message_processor::CodexMessageProcessor;
 use crate::error_code::INVALID_REQUEST_ERROR_CODE;
 use crate::outgoing_message::OutgoingMessageSender;
-use codex_protocol::mcp_protocol::ClientInfo;
-use codex_protocol::mcp_protocol::ClientRequest;
-use codex_protocol::mcp_protocol::InitializeResponse;
+use codex_app_server_protocol::ClientInfo;
+use codex_app_server_protocol::ClientRequest;
+use codex_app_server_protocol::InitializeResponse;
 
+use codex_app_server_protocol::JSONRPCError;
+use codex_app_server_protocol::JSONRPCErrorError;
+use codex_app_server_protocol::JSONRPCNotification;
+use codex_app_server_protocol::JSONRPCRequest;
+use codex_app_server_protocol::JSONRPCResponse;
 use codex_core::AuthManager;
 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 mcp_types::JSONRPCError;
-use mcp_types::JSONRPCErrorError;
-use mcp_types::JSONRPCNotification;
-use mcp_types::JSONRPCRequest;
-use mcp_types::JSONRPCResponse;
+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/src/outgoing_message.rs

@@ -2,16 +2,12 @@ use std::collections::HashMap;
 use std::sync::atomic::AtomicI64;
 use std::sync::atomic::Ordering;
 
-use codex_protocol::mcp_protocol::ServerNotification;
-use mcp_types::JSONRPC_VERSION;
-use mcp_types::JSONRPCError;
-use mcp_types::JSONRPCErrorError;
-use mcp_types::JSONRPCMessage;
-use mcp_types::JSONRPCNotification;
-use mcp_types::JSONRPCRequest;
-use mcp_types::JSONRPCResponse;
-use mcp_types::RequestId;
-use mcp_types::Result;
+use codex_app_server_protocol::JSONRPCErrorError;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::Result;
+use codex_app_server_protocol::ServerNotification;
+use codex_app_server_protocol::ServerRequest;
+use codex_app_server_protocol::ServerRequestPayload;
 use serde::Serialize;
 use tokio::sync::Mutex;
 use tokio::sync::mpsc;
@@ -38,8 +34,7 @@ impl OutgoingMessageSender {
 
     pub(crate) async fn send_request(
         &self,
-        method: &str,
-        params: Option<serde_json::Value>,
+        request: ServerRequestPayload,
     ) -> oneshot::Receiver<Result> {
         let id = RequestId::Integer(self.next_request_id.fetch_add(1, Ordering::Relaxed));
         let outgoing_message_id = id.clone();
@@ -49,11 +44,8 @@ impl OutgoingMessageSender {
             request_id_to_callback.insert(id, tx_approve);
         }
 
-        let outgoing_message = OutgoingMessage::Request(OutgoingRequest {
-            id: outgoing_message_id,
-            method: method.to_string(),
-            params,
-        });
+        let outgoing_message =
+            OutgoingMessage::Request(request.request_with_id(outgoing_message_id));
         let _ = self.sender.send(outgoing_message);
         rx_approve
     }
@@ -116,8 +108,10 @@ impl OutgoingMessageSender {
 }
 
 /// Outgoing message from the server to the client.
+#[derive(Debug, Clone, Serialize)]
+#[serde(untagged)]
 pub(crate) enum OutgoingMessage {
-    Request(OutgoingRequest),
+    Request(ServerRequest),
     Notification(OutgoingNotification),
     /// AppServerNotification is specific to the case where this is run as an
     /// "app server" as opposed to an MCP server.
@@ -126,64 +120,6 @@ pub(crate) enum OutgoingMessage {
     Error(OutgoingError),
 }
 
-impl From<OutgoingMessage> for JSONRPCMessage {
-    fn from(val: OutgoingMessage) -> Self {
-        use OutgoingMessage::*;
-        match val {
-            Request(OutgoingRequest { id, method, params }) => {
-                JSONRPCMessage::Request(JSONRPCRequest {
-                    jsonrpc: JSONRPC_VERSION.into(),
-                    id,
-                    method,
-                    params,
-                })
-            }
-            Notification(OutgoingNotification { method, params }) => {
-                JSONRPCMessage::Notification(JSONRPCNotification {
-                    jsonrpc: JSONRPC_VERSION.into(),
-                    method,
-                    params,
-                })
-            }
-            AppServerNotification(notification) => {
-                let method = notification.to_string();
-                let params = match notification.to_params() {
-                    Ok(params) => Some(params),
-                    Err(err) => {
-                        warn!("failed to serialize notification params: {err}");
-                        None
-                    }
-                };
-                JSONRPCMessage::Notification(JSONRPCNotification {
-                    jsonrpc: JSONRPC_VERSION.into(),
-                    method,
-                    params,
-                })
-            }
-            Response(OutgoingResponse { id, result }) => {
-                JSONRPCMessage::Response(JSONRPCResponse {
-                    jsonrpc: JSONRPC_VERSION.into(),
-                    id,
-                    result,
-                })
-            }
-            Error(OutgoingError { id, error }) => JSONRPCMessage::Error(JSONRPCError {
-                jsonrpc: JSONRPC_VERSION.into(),
-                id,
-                error,
-            }),
-        }
-    }
-}
-
-#[derive(Debug, Clone, PartialEq, Serialize)]
-pub(crate) struct OutgoingRequest {
-    pub id: RequestId,
-    pub method: String,
-    #[serde(default, skip_serializing_if = "Option::is_none")]
-    pub params: Option<serde_json::Value>,
-}
-
 #[derive(Debug, Clone, PartialEq, Serialize)]
 pub(crate) struct OutgoingNotification {
     pub method: String,
@@ -205,7 +141,7 @@ pub(crate) struct OutgoingError {
 
 #[cfg(test)]
 mod tests {
-    use codex_protocol::mcp_protocol::LoginChatGptCompleteNotification;
+    use codex_app_server_protocol::LoginChatGptCompleteNotification;
     use pretty_assertions::assert_eq;
     use serde_json::json;
     use uuid::Uuid;
@@ -221,18 +157,17 @@ mod tests {
                 error: None,
             });
 
-        let jsonrpc_notification: JSONRPCMessage =
-            OutgoingMessage::AppServerNotification(notification).into();
+        let jsonrpc_notification = OutgoingMessage::AppServerNotification(notification);
         assert_eq!(
-            JSONRPCMessage::Notification(JSONRPCNotification {
-                jsonrpc: "2.0".into(),
-                method: "loginChatGptComplete".into(),
-                params: Some(json!({
+            json!({
+                "method": "loginChatGptComplete",
+                "params": {
                     "loginId": Uuid::nil(),
                     "success": true,
-                })),
+                },
             }),
-            jsonrpc_notification,
+            serde_json::to_value(jsonrpc_notification)
+                .expect("ensure the strum macros serialize the method field correctly"),
             "ensure the strum macros serialize the method field correctly"
         );
     }

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

@@ -9,8 +9,7 @@ path = "lib.rs"
 [dependencies]
 anyhow = { workspace = true }
 assert_cmd = { workspace = true }
-codex-protocol = { workspace = true }
-mcp-types = { workspace = true }
+codex-app-server-protocol = { workspace = true }
 serde = { workspace = true }
 serde_json = { workspace = true }
 tokio = { workspace = true, features = [

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

@@ -2,8 +2,8 @@ mod mcp_process;
 mod mock_model_server;
 mod responses;
 
+use codex_app_server_protocol::JSONRPCResponse;
 pub use mcp_process::McpProcess;
-use mcp_types::JSONRPCResponse;
 pub use mock_model_server::create_mock_chat_completions_server;
 pub use responses::create_apply_patch_sse_response;
 pub use responses::create_final_assistant_message_sse_response;

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

@@ -1,3 +1,4 @@
+use std::collections::VecDeque;
 use std::path::Path;
 use std::process::Stdio;
 use std::sync::atomic::AtomicI64;
@@ -11,29 +12,30 @@ use tokio::process::ChildStdout;
 
 use anyhow::Context;
 use assert_cmd::prelude::*;
-use codex_protocol::mcp_protocol::AddConversationListenerParams;
-use codex_protocol::mcp_protocol::ArchiveConversationParams;
-use codex_protocol::mcp_protocol::CancelLoginChatGptParams;
-use codex_protocol::mcp_protocol::ClientInfo;
-use codex_protocol::mcp_protocol::ClientNotification;
-use codex_protocol::mcp_protocol::GetAuthStatusParams;
-use codex_protocol::mcp_protocol::InitializeParams;
-use codex_protocol::mcp_protocol::InterruptConversationParams;
-use codex_protocol::mcp_protocol::ListConversationsParams;
-use codex_protocol::mcp_protocol::LoginApiKeyParams;
-use codex_protocol::mcp_protocol::NewConversationParams;
-use codex_protocol::mcp_protocol::RemoveConversationListenerParams;
-use codex_protocol::mcp_protocol::ResumeConversationParams;
-use codex_protocol::mcp_protocol::SendUserMessageParams;
-use codex_protocol::mcp_protocol::SendUserTurnParams;
-use codex_protocol::mcp_protocol::SetDefaultModelParams;
+use codex_app_server_protocol::AddConversationListenerParams;
+use codex_app_server_protocol::ArchiveConversationParams;
+use codex_app_server_protocol::CancelLoginChatGptParams;
+use codex_app_server_protocol::ClientInfo;
+use codex_app_server_protocol::ClientNotification;
+use codex_app_server_protocol::GetAuthStatusParams;
+use codex_app_server_protocol::InitializeParams;
+use codex_app_server_protocol::InterruptConversationParams;
+use codex_app_server_protocol::ListConversationsParams;
+use codex_app_server_protocol::LoginApiKeyParams;
+use codex_app_server_protocol::NewConversationParams;
+use codex_app_server_protocol::RemoveConversationListenerParams;
+use codex_app_server_protocol::ResumeConversationParams;
+use codex_app_server_protocol::SendUserMessageParams;
+use codex_app_server_protocol::SendUserTurnParams;
+use codex_app_server_protocol::ServerRequest;
+use codex_app_server_protocol::SetDefaultModelParams;
 
-use mcp_types::JSONRPC_VERSION;
-use mcp_types::JSONRPCMessage;
-use mcp_types::JSONRPCNotification;
-use mcp_types::JSONRPCRequest;
-use mcp_types::JSONRPCResponse;
-use mcp_types::RequestId;
+use codex_app_server_protocol::JSONRPCError;
+use codex_app_server_protocol::JSONRPCMessage;
+use codex_app_server_protocol::JSONRPCNotification;
+use codex_app_server_protocol::JSONRPCRequest;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::RequestId;
 use std::process::Command as StdCommand;
 use tokio::process::Command;
 
@@ -46,6 +48,7 @@ pub struct McpProcess {
     process: Child,
     stdin: ChildStdin,
     stdout: BufReader<ChildStdout>,
+    pending_user_messages: VecDeque<JSONRPCNotification>,
 }
 
 impl McpProcess {
@@ -116,6 +119,7 @@ impl McpProcess {
             process,
             stdin,
             stdout,
+            pending_user_messages: VecDeque::new(),
         })
     }
 
@@ -317,7 +321,6 @@ impl McpProcess {
         let request_id = self.next_request_id.fetch_add(1, Ordering::Relaxed);
 
         let message = JSONRPCMessage::Request(JSONRPCRequest {
-            jsonrpc: JSONRPC_VERSION.into(),
             id: RequestId::Integer(request_id),
             method: method.to_string(),
             params,
@@ -331,12 +334,8 @@ impl McpProcess {
         id: RequestId,
         result: serde_json::Value,
     ) -> anyhow::Result<()> {
-        self.send_jsonrpc_message(JSONRPCMessage::Response(JSONRPCResponse {
-            jsonrpc: JSONRPC_VERSION.into(),
-            id,
-            result,
-        }))
-        .await
+        self.send_jsonrpc_message(JSONRPCMessage::Response(JSONRPCResponse { id, result }))
+            .await
     }
 
     pub async fn send_notification(
@@ -345,7 +344,6 @@ impl McpProcess {
     ) -> anyhow::Result<()> {
         let value = serde_json::to_value(notification)?;
         self.send_jsonrpc_message(JSONRPCMessage::Notification(JSONRPCNotification {
-            jsonrpc: JSONRPC_VERSION.into(),
             method: value
                 .get("method")
                 .and_then(|m| m.as_str())
@@ -373,18 +371,21 @@ impl McpProcess {
         Ok(message)
     }
 
-    pub async fn read_stream_until_request_message(&mut self) -> anyhow::Result<JSONRPCRequest> {
+    pub async fn read_stream_until_request_message(&mut self) -> anyhow::Result<ServerRequest> {
         eprintln!("in read_stream_until_request_message()");
 
         loop {
             let message = self.read_jsonrpc_message().await?;
 
             match message {
-                JSONRPCMessage::Notification(_) => {
-                    eprintln!("notification: {message:?}");
+                JSONRPCMessage::Notification(notification) => {
+                    eprintln!("notification: {notification:?}");
+                    self.enqueue_user_message(notification);
                 }
                 JSONRPCMessage::Request(jsonrpc_request) => {
-                    return Ok(jsonrpc_request);
+                    return jsonrpc_request.try_into().with_context(
+                        || "failed to deserialize ServerRequest from JSONRPCRequest",
+                    );
                 }
                 JSONRPCMessage::Error(_) => {
                     anyhow::bail!("unexpected JSONRPCMessage::Error: {message:?}");
@@ -405,8 +406,9 @@ impl McpProcess {
         loop {
             let message = self.read_jsonrpc_message().await?;
             match message {
-                JSONRPCMessage::Notification(_) => {
-                    eprintln!("notification: {message:?}");
+                JSONRPCMessage::Notification(notification) => {
+                    eprintln!("notification: {notification:?}");
+                    self.enqueue_user_message(notification);
                 }
                 JSONRPCMessage::Request(_) => {
                     anyhow::bail!("unexpected JSONRPCMessage::Request: {message:?}");
@@ -426,12 +428,13 @@ impl McpProcess {
     pub async fn read_stream_until_error_message(
         &mut self,
         request_id: RequestId,
-    ) -> anyhow::Result<mcp_types::JSONRPCError> {
+    ) -> anyhow::Result<JSONRPCError> {
         loop {
             let message = self.read_jsonrpc_message().await?;
             match message {
-                JSONRPCMessage::Notification(_) => {
-                    eprintln!("notification: {message:?}");
+                JSONRPCMessage::Notification(notification) => {
+                    eprintln!("notification: {notification:?}");
+                    self.enqueue_user_message(notification);
                 }
                 JSONRPCMessage::Request(_) => {
                     anyhow::bail!("unexpected JSONRPCMessage::Request: {message:?}");
@@ -454,6 +457,10 @@ impl McpProcess {
     ) -> anyhow::Result<JSONRPCNotification> {
         eprintln!("in read_stream_until_notification_message({method})");
 
+        if let Some(notification) = self.take_pending_notification_by_method(method) {
+            return Ok(notification);
+        }
+
         loop {
             let message = self.read_jsonrpc_message().await?;
             match message {
@@ -461,6 +468,7 @@ impl McpProcess {
                     if notification.method == method {
                         return Ok(notification);
                     }
+                    self.enqueue_user_message(notification);
                 }
                 JSONRPCMessage::Request(_) => {
                     anyhow::bail!("unexpected JSONRPCMessage::Request: {message:?}");
@@ -474,4 +482,21 @@ impl McpProcess {
             }
         }
     }
+
+    fn take_pending_notification_by_method(&mut self, method: &str) -> Option<JSONRPCNotification> {
+        if let Some(pos) = self
+            .pending_user_messages
+            .iter()
+            .position(|notification| notification.method == method)
+        {
+            return self.pending_user_messages.remove(pos);
+        }
+        None
+    }
+
+    fn enqueue_user_message(&mut self, notification: JSONRPCNotification) {
+        if notification.method == "codex/event/user_message" {
+            self.pending_user_messages.push_back(notification);
+        }
+    }
 }

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

@@ -2,13 +2,13 @@ use std::path::Path;
 
 use app_test_support::McpProcess;
 use app_test_support::to_response;
+use codex_app_server_protocol::ArchiveConversationParams;
+use codex_app_server_protocol::ArchiveConversationResponse;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::NewConversationParams;
+use codex_app_server_protocol::NewConversationResponse;
+use codex_app_server_protocol::RequestId;
 use codex_core::ARCHIVED_SESSIONS_SUBDIR;
-use codex_protocol::mcp_protocol::ArchiveConversationParams;
-use codex_protocol::mcp_protocol::ArchiveConversationResponse;
-use codex_protocol::mcp_protocol::NewConversationParams;
-use codex_protocol::mcp_protocol::NewConversationResponse;
-use mcp_types::JSONRPCResponse;
-use mcp_types::RequestId;
 use tempfile::TempDir;
 use tokio::time::timeout;
 

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

@@ -2,13 +2,13 @@ use std::path::Path;
 
 use app_test_support::McpProcess;
 use app_test_support::to_response;
-use codex_protocol::mcp_protocol::AuthMode;
-use codex_protocol::mcp_protocol::GetAuthStatusParams;
-use codex_protocol::mcp_protocol::GetAuthStatusResponse;
-use codex_protocol::mcp_protocol::LoginApiKeyParams;
-use codex_protocol::mcp_protocol::LoginApiKeyResponse;
-use mcp_types::JSONRPCResponse;
-use mcp_types::RequestId;
+use codex_app_server_protocol::AuthMode;
+use codex_app_server_protocol::GetAuthStatusParams;
+use codex_app_server_protocol::GetAuthStatusResponse;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::LoginApiKeyParams;
+use codex_app_server_protocol::LoginApiKeyResponse;
+use codex_app_server_protocol::RequestId;
 use pretty_assertions::assert_eq;
 use tempfile::TempDir;
 use tokio::time::timeout;

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

@@ -5,25 +5,31 @@ use app_test_support::create_final_assistant_message_sse_response;
 use app_test_support::create_mock_chat_completions_server;
 use app_test_support::create_shell_sse_response;
 use app_test_support::to_response;
+use codex_app_server_protocol::AddConversationListenerParams;
+use codex_app_server_protocol::AddConversationSubscriptionResponse;
+use codex_app_server_protocol::ExecCommandApprovalParams;
+use codex_app_server_protocol::InputItem;
+use codex_app_server_protocol::JSONRPCNotification;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::NewConversationParams;
+use codex_app_server_protocol::NewConversationResponse;
+use codex_app_server_protocol::RemoveConversationListenerParams;
+use codex_app_server_protocol::RemoveConversationSubscriptionResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::SendUserMessageParams;
+use codex_app_server_protocol::SendUserMessageResponse;
+use codex_app_server_protocol::SendUserTurnParams;
+use codex_app_server_protocol::SendUserTurnResponse;
+use codex_app_server_protocol::ServerRequest;
 use codex_core::protocol::AskForApproval;
 use codex_core::protocol::SandboxPolicy;
 use codex_core::protocol_config_types::ReasoningEffort;
 use codex_core::protocol_config_types::ReasoningSummary;
 use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
-use codex_protocol::mcp_protocol::AddConversationListenerParams;
-use codex_protocol::mcp_protocol::AddConversationSubscriptionResponse;
-use codex_protocol::mcp_protocol::EXEC_COMMAND_APPROVAL_METHOD;
-use codex_protocol::mcp_protocol::NewConversationParams;
-use codex_protocol::mcp_protocol::NewConversationResponse;
-use codex_protocol::mcp_protocol::RemoveConversationListenerParams;
-use codex_protocol::mcp_protocol::RemoveConversationSubscriptionResponse;
-use codex_protocol::mcp_protocol::SendUserMessageParams;
-use codex_protocol::mcp_protocol::SendUserMessageResponse;
-use codex_protocol::mcp_protocol::SendUserTurnParams;
-use codex_protocol::mcp_protocol::SendUserTurnResponse;
-use mcp_types::JSONRPCNotification;
-use mcp_types::JSONRPCResponse;
-use mcp_types::RequestId;
+use codex_protocol::config_types::SandboxMode;
+use codex_protocol::protocol::Event;
+use codex_protocol::protocol::EventMsg;
+use codex_protocol::protocol::InputMessageKind;
 use pretty_assertions::assert_eq;
 use std::env;
 use tempfile::TempDir;
@@ -115,7 +121,7 @@ async fn test_codex_jsonrpc_conversation_flow() {
     let send_user_id = mcp
         .send_send_user_message_request(SendUserMessageParams {
             conversation_id,
-            items: vec![codex_protocol::mcp_protocol::InputItem::Text {
+            items: vec![codex_app_server_protocol::InputItem::Text {
                 text: "text".to_string(),
             }],
         })
@@ -265,7 +271,7 @@ async fn test_send_user_turn_changes_approval_policy_behavior() {
     let send_user_id = mcp
         .send_send_user_message_request(SendUserMessageParams {
             conversation_id,
-            items: vec![codex_protocol::mcp_protocol::InputItem::Text {
+            items: vec![codex_app_server_protocol::InputItem::Text {
                 text: "run python".to_string(),
             }],
         })
@@ -290,11 +296,28 @@ async fn test_send_user_turn_changes_approval_policy_behavior() {
     .await
     .expect("waiting for exec approval request timeout")
     .expect("exec approval request");
-    assert_eq!(request.method, EXEC_COMMAND_APPROVAL_METHOD);
+    let ServerRequest::ExecCommandApproval { request_id, params } = request else {
+        panic!("expected ExecCommandApproval request, got: {request:?}");
+    };
+
+    assert_eq!(
+        ExecCommandApprovalParams {
+            conversation_id,
+            call_id: "call1".to_string(),
+            command: vec![
+                "python3".to_string(),
+                "-c".to_string(),
+                "print(42)".to_string(),
+            ],
+            cwd: working_directory.clone(),
+            reason: None,
+        },
+        params
+    );
 
     // Approve so the first turn can complete
     mcp.send_response(
-        request.id,
+        request_id,
         serde_json::json!({ "decision": codex_core::protocol::ReviewDecision::Approved }),
     )
     .await
@@ -313,7 +336,7 @@ async fn test_send_user_turn_changes_approval_policy_behavior() {
     let send_turn_id = mcp
         .send_send_user_turn_request(SendUserTurnParams {
             conversation_id,
-            items: vec![codex_protocol::mcp_protocol::InputItem::Text {
+            items: vec![codex_app_server_protocol::InputItem::Text {
                 text: "run python again".to_string(),
             }],
             cwd: working_directory.clone(),
@@ -349,6 +372,234 @@ async fn test_send_user_turn_changes_approval_policy_behavior() {
 }
 
 // Helper: minimal config.toml pointing at mock provider.
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 4)]
+async fn test_send_user_turn_updates_sandbox_and_cwd_between_turns() {
+    if env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    let tmp = TempDir::new().expect("tmp dir");
+    let codex_home = tmp.path().join("codex_home");
+    std::fs::create_dir(&codex_home).expect("create codex home dir");
+    let workspace_root = tmp.path().join("workspace");
+    std::fs::create_dir(&workspace_root).expect("create workspace root");
+    let first_cwd = workspace_root.join("turn1");
+    let second_cwd = workspace_root.join("turn2");
+    std::fs::create_dir(&first_cwd).expect("create first cwd");
+    std::fs::create_dir(&second_cwd).expect("create second cwd");
+
+    let responses = vec![
+        create_shell_sse_response(
+            vec![
+                "bash".to_string(),
+                "-lc".to_string(),
+                "echo first turn".to_string(),
+            ],
+            None,
+            Some(5000),
+            "call-first",
+        )
+        .expect("create first shell response"),
+        create_final_assistant_message_sse_response("done first")
+            .expect("create first final assistant message"),
+        create_shell_sse_response(
+            vec![
+                "bash".to_string(),
+                "-lc".to_string(),
+                "echo second turn".to_string(),
+            ],
+            None,
+            Some(5000),
+            "call-second",
+        )
+        .expect("create second shell response"),
+        create_final_assistant_message_sse_response("done second")
+            .expect("create second final assistant message"),
+    ];
+    let server = create_mock_chat_completions_server(responses).await;
+    create_config_toml(&codex_home, &server.uri()).expect("write config");
+
+    let mut mcp = McpProcess::new(&codex_home)
+        .await
+        .expect("spawn mcp process");
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
+        .await
+        .expect("init timeout")
+        .expect("init failed");
+
+    let new_conv_id = mcp
+        .send_new_conversation_request(NewConversationParams {
+            cwd: Some(first_cwd.to_string_lossy().into_owned()),
+            approval_policy: Some(AskForApproval::Never),
+            sandbox: Some(SandboxMode::WorkspaceWrite),
+            ..Default::default()
+        })
+        .await
+        .expect("send newConversation");
+    let new_conv_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(new_conv_id)),
+    )
+    .await
+    .expect("newConversation timeout")
+    .expect("newConversation resp");
+    let NewConversationResponse {
+        conversation_id,
+        model,
+        ..
+    } = to_response::<NewConversationResponse>(new_conv_resp)
+        .expect("deserialize newConversation response");
+
+    let add_listener_id = mcp
+        .send_add_conversation_listener_request(AddConversationListenerParams { conversation_id })
+        .await
+        .expect("send addConversationListener");
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(add_listener_id)),
+    )
+    .await
+    .expect("addConversationListener timeout")
+    .expect("addConversationListener resp");
+
+    let first_turn_id = mcp
+        .send_send_user_turn_request(SendUserTurnParams {
+            conversation_id,
+            items: vec![InputItem::Text {
+                text: "first turn".to_string(),
+            }],
+            cwd: first_cwd.clone(),
+            approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::WorkspaceWrite {
+                writable_roots: vec![first_cwd.clone()],
+                network_access: false,
+                exclude_tmpdir_env_var: false,
+                exclude_slash_tmp: false,
+            },
+            model: model.clone(),
+            effort: Some(ReasoningEffort::Medium),
+            summary: ReasoningSummary::Auto,
+        })
+        .await
+        .expect("send first sendUserTurn");
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(first_turn_id)),
+    )
+    .await
+    .expect("sendUserTurn 1 timeout")
+    .expect("sendUserTurn 1 resp");
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("codex/event/task_complete"),
+    )
+    .await
+    .expect("task_complete 1 timeout")
+    .expect("task_complete 1 notification");
+
+    let second_turn_id = mcp
+        .send_send_user_turn_request(SendUserTurnParams {
+            conversation_id,
+            items: vec![InputItem::Text {
+                text: "second turn".to_string(),
+            }],
+            cwd: second_cwd.clone(),
+            approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::DangerFullAccess,
+            model: model.clone(),
+            effort: Some(ReasoningEffort::Medium),
+            summary: ReasoningSummary::Auto,
+        })
+        .await
+        .expect("send second sendUserTurn");
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(second_turn_id)),
+    )
+    .await
+    .expect("sendUserTurn 2 timeout")
+    .expect("sendUserTurn 2 resp");
+
+    let mut env_message: Option<String> = None;
+    let second_cwd_str = second_cwd.to_string_lossy().into_owned();
+    for _ in 0..10 {
+        let notification = timeout(
+            DEFAULT_READ_TIMEOUT,
+            mcp.read_stream_until_notification_message("codex/event/user_message"),
+        )
+        .await
+        .expect("user_message timeout")
+        .expect("user_message notification");
+        let params = notification
+            .params
+            .clone()
+            .expect("user_message should include params");
+        let event: Event = serde_json::from_value(params).expect("deserialize user_message event");
+        if let EventMsg::UserMessage(user) = event.msg
+            && matches!(user.kind, Some(InputMessageKind::EnvironmentContext))
+            && user.message.contains(&second_cwd_str)
+        {
+            env_message = Some(user.message);
+            break;
+        }
+    }
+    let env_message = env_message.expect("expected environment context update");
+    assert!(
+        env_message.contains("<sandbox_mode>danger-full-access</sandbox_mode>"),
+        "env context should reflect new sandbox mode: {env_message}"
+    );
+    assert!(
+        env_message.contains("<network_access>enabled</network_access>"),
+        "env context should enable network access for danger-full-access policy: {env_message}"
+    );
+    assert!(
+        env_message.contains(&second_cwd_str),
+        "env context should include updated cwd: {env_message}"
+    );
+
+    let exec_begin_notification = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("codex/event/exec_command_begin"),
+    )
+    .await
+    .expect("exec_command_begin timeout")
+    .expect("exec_command_begin notification");
+    let params = exec_begin_notification
+        .params
+        .clone()
+        .expect("exec_command_begin params");
+    let event: Event = serde_json::from_value(params).expect("deserialize exec begin event");
+    let exec_begin = match event.msg {
+        EventMsg::ExecCommandBegin(exec_begin) => exec_begin,
+        other => panic!("expected ExecCommandBegin event, got {other:?}"),
+    };
+    assert_eq!(
+        exec_begin.cwd, second_cwd,
+        "exec turn should run from updated cwd"
+    );
+    assert_eq!(
+        exec_begin.command,
+        vec![
+            "bash".to_string(),
+            "-lc".to_string(),
+            "echo second turn".to_string()
+        ],
+        "exec turn should run expected command"
+    );
+
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("codex/event/task_complete"),
+    )
+    .await
+    .expect("task_complete 2 timeout")
+    .expect("task_complete 2 notification");
+}
+
 fn create_config_toml(codex_home: &Path, server_uri: &str) -> std::io::Result<()> {
     let config_toml = codex_home.join("config.toml");
     std::fs::write(

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

@@ -3,18 +3,18 @@ use std::path::Path;
 
 use app_test_support::McpProcess;
 use app_test_support::to_response;
+use codex_app_server_protocol::GetUserSavedConfigResponse;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::Profile;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::SandboxSettings;
+use codex_app_server_protocol::Tools;
+use codex_app_server_protocol::UserSavedConfig;
 use codex_core::protocol::AskForApproval;
 use codex_protocol::config_types::ReasoningEffort;
 use codex_protocol::config_types::ReasoningSummary;
 use codex_protocol::config_types::SandboxMode;
 use codex_protocol::config_types::Verbosity;
-use codex_protocol::mcp_protocol::GetUserSavedConfigResponse;
-use codex_protocol::mcp_protocol::Profile;
-use codex_protocol::mcp_protocol::SandboxSettings;
-use codex_protocol::mcp_protocol::Tools;
-use codex_protocol::mcp_protocol::UserSavedConfig;
-use mcp_types::JSONRPCResponse;
-use mcp_types::RequestId;
 use pretty_assertions::assert_eq;
 use tempfile::TempDir;
 use tokio::time::timeout;

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

@@ -4,15 +4,15 @@ use app_test_support::McpProcess;
 use app_test_support::create_final_assistant_message_sse_response;
 use app_test_support::create_mock_chat_completions_server;
 use app_test_support::to_response;
-use codex_protocol::mcp_protocol::AddConversationListenerParams;
-use codex_protocol::mcp_protocol::AddConversationSubscriptionResponse;
-use codex_protocol::mcp_protocol::InputItem;
-use codex_protocol::mcp_protocol::NewConversationParams;
-use codex_protocol::mcp_protocol::NewConversationResponse;
-use codex_protocol::mcp_protocol::SendUserMessageParams;
-use codex_protocol::mcp_protocol::SendUserMessageResponse;
-use mcp_types::JSONRPCResponse;
-use mcp_types::RequestId;
+use codex_app_server_protocol::AddConversationListenerParams;
+use codex_app_server_protocol::AddConversationSubscriptionResponse;
+use codex_app_server_protocol::InputItem;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::NewConversationParams;
+use codex_app_server_protocol::NewConversationResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::SendUserMessageParams;
+use codex_app_server_protocol::SendUserMessageResponse;
 use pretty_assertions::assert_eq;
 use serde_json::json;
 use tempfile::TempDir;

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

@@ -1,6 +1,8 @@
+use anyhow::Context;
+use anyhow::Result;
 use app_test_support::McpProcess;
-use mcp_types::JSONRPCResponse;
-use mcp_types::RequestId;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::RequestId;
 use pretty_assertions::assert_eq;
 use serde_json::json;
 use tempfile::TempDir;
@@ -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/app-server/tests/suite/interrupt.rs

@@ -3,17 +3,17 @@
 
 use std::path::Path;
 
+use codex_app_server_protocol::AddConversationListenerParams;
+use codex_app_server_protocol::InterruptConversationParams;
+use codex_app_server_protocol::InterruptConversationResponse;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::NewConversationParams;
+use codex_app_server_protocol::NewConversationResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::SendUserMessageParams;
+use codex_app_server_protocol::SendUserMessageResponse;
 use codex_core::protocol::TurnAbortReason;
-use codex_protocol::mcp_protocol::AddConversationListenerParams;
-use codex_protocol::mcp_protocol::InterruptConversationParams;
-use codex_protocol::mcp_protocol::InterruptConversationResponse;
-use codex_protocol::mcp_protocol::NewConversationParams;
-use codex_protocol::mcp_protocol::NewConversationResponse;
-use codex_protocol::mcp_protocol::SendUserMessageParams;
-use codex_protocol::mcp_protocol::SendUserMessageResponse;
 use core_test_support::skip_if_no_network;
-use mcp_types::JSONRPCResponse;
-use mcp_types::RequestId;
 use tempfile::TempDir;
 use tokio::time::timeout;
 
@@ -100,7 +100,7 @@ async fn shell_command_interruption() -> anyhow::Result<()> {
     let send_user_id = mcp
         .send_send_user_message_request(SendUserMessageParams {
             conversation_id,
-            items: vec![codex_protocol::mcp_protocol::InputItem::Text {
+            items: vec![codex_app_server_protocol::InputItem::Text {
                 text: "run first sleep command".to_string(),
             }],
         })

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

@@ -3,16 +3,16 @@ use std::path::Path;
 
 use app_test_support::McpProcess;
 use app_test_support::to_response;
-use codex_protocol::mcp_protocol::ListConversationsParams;
-use codex_protocol::mcp_protocol::ListConversationsResponse;
-use codex_protocol::mcp_protocol::NewConversationParams; // reused for overrides shape
-use codex_protocol::mcp_protocol::ResumeConversationParams;
-use codex_protocol::mcp_protocol::ResumeConversationResponse;
-use codex_protocol::mcp_protocol::ServerNotification;
-use codex_protocol::mcp_protocol::SessionConfiguredNotification;
-use mcp_types::JSONRPCNotification;
-use mcp_types::JSONRPCResponse;
-use mcp_types::RequestId;
+use codex_app_server_protocol::JSONRPCNotification;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::ListConversationsParams;
+use codex_app_server_protocol::ListConversationsResponse;
+use codex_app_server_protocol::NewConversationParams; // reused for overrides shape
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::ResumeConversationParams;
+use codex_app_server_protocol::ResumeConversationResponse;
+use codex_app_server_protocol::ServerNotification;
+use codex_app_server_protocol::SessionConfiguredNotification;
 use pretty_assertions::assert_eq;
 use serde_json::json;
 use tempfile::TempDir;

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

@@ -3,15 +3,15 @@ use std::time::Duration;
 
 use app_test_support::McpProcess;
 use app_test_support::to_response;
+use codex_app_server_protocol::CancelLoginChatGptParams;
+use codex_app_server_protocol::CancelLoginChatGptResponse;
+use codex_app_server_protocol::GetAuthStatusParams;
+use codex_app_server_protocol::GetAuthStatusResponse;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::LoginChatGptResponse;
+use codex_app_server_protocol::LogoutChatGptResponse;
+use codex_app_server_protocol::RequestId;
 use codex_login::login_with_api_key;
-use codex_protocol::mcp_protocol::CancelLoginChatGptParams;
-use codex_protocol::mcp_protocol::CancelLoginChatGptResponse;
-use codex_protocol::mcp_protocol::GetAuthStatusParams;
-use codex_protocol::mcp_protocol::GetAuthStatusResponse;
-use codex_protocol::mcp_protocol::LoginChatGptResponse;
-use codex_protocol::mcp_protocol::LogoutChatGptResponse;
-use mcp_types::JSONRPCResponse;
-use mcp_types::RequestId;
 use tempfile::TempDir;
 use tokio::time::timeout;
 

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

@@ -4,17 +4,17 @@ use app_test_support::McpProcess;
 use app_test_support::create_final_assistant_message_sse_response;
 use app_test_support::create_mock_chat_completions_server;
 use app_test_support::to_response;
-use codex_protocol::mcp_protocol::AddConversationListenerParams;
-use codex_protocol::mcp_protocol::AddConversationSubscriptionResponse;
-use codex_protocol::mcp_protocol::ConversationId;
-use codex_protocol::mcp_protocol::InputItem;
-use codex_protocol::mcp_protocol::NewConversationParams;
-use codex_protocol::mcp_protocol::NewConversationResponse;
-use codex_protocol::mcp_protocol::SendUserMessageParams;
-use codex_protocol::mcp_protocol::SendUserMessageResponse;
-use mcp_types::JSONRPCNotification;
-use mcp_types::JSONRPCResponse;
-use mcp_types::RequestId;
+use codex_app_server_protocol::AddConversationListenerParams;
+use codex_app_server_protocol::AddConversationSubscriptionResponse;
+use codex_app_server_protocol::InputItem;
+use codex_app_server_protocol::JSONRPCNotification;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::NewConversationParams;
+use codex_app_server_protocol::NewConversationResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::SendUserMessageParams;
+use codex_app_server_protocol::SendUserMessageResponse;
+use codex_protocol::ConversationId;
 use pretty_assertions::assert_eq;
 use tempfile::TempDir;
 use tokio::time::timeout;

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

@@ -2,11 +2,11 @@ use std::path::Path;
 
 use app_test_support::McpProcess;
 use app_test_support::to_response;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::SetDefaultModelParams;
+use codex_app_server_protocol::SetDefaultModelResponse;
 use codex_core::config::ConfigToml;
-use codex_protocol::mcp_protocol::SetDefaultModelParams;
-use codex_protocol::mcp_protocol::SetDefaultModelResponse;
-use mcp_types::JSONRPCResponse;
-use mcp_types::RequestId;
 use pretty_assertions::assert_eq;
 use tempfile::TempDir;
 use tokio::time::timeout;

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

@@ -1,8 +1,8 @@
 use app_test_support::McpProcess;
 use app_test_support::to_response;
-use codex_protocol::mcp_protocol::GetUserAgentResponse;
-use mcp_types::JSONRPCResponse;
-use mcp_types::RequestId;
+use codex_app_server_protocol::GetUserAgentResponse;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::RequestId;
 use pretty_assertions::assert_eq;
 use tempfile::TempDir;
 use tokio::time::timeout;

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

@@ -5,14 +5,14 @@ use app_test_support::McpProcess;
 use app_test_support::to_response;
 use base64::Engine;
 use base64::engine::general_purpose::URL_SAFE_NO_PAD;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::UserInfoResponse;
 use codex_core::auth::AuthDotJson;
 use codex_core::auth::get_auth_file;
 use codex_core::auth::write_auth_json;
 use codex_core::token_data::IdTokenInfo;
 use codex_core::token_data::TokenData;
-use codex_protocol::mcp_protocol::UserInfoResponse;
-use mcp_types::JSONRPCResponse;
-use mcp_types::RequestId;
 use pretty_assertions::assert_eq;
 use serde_json::json;
 use tempfile::TempDir;

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

@@ -28,9 +28,11 @@ codex-login = { workspace = true }
 codex-mcp-server = { workspace = true }
 codex-process-hardening = { workspace = true }
 codex-protocol = { workspace = true }
+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 }
@@ -43,10 +45,9 @@ tokio = { workspace = true, features = [
     "rt-multi-thread",
     "signal",
 ] }
-tracing = { workspace = true }
-tracing-subscriber = { workspace = true }
 
 [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/lib.rs

@@ -1,7 +1,6 @@
 pub mod debug_sandbox;
 mod exit_status;
 pub mod login;
-pub mod proto;
 
 use clap::Parser;
 use codex_common::CliConfigOverrides;

codex-rs/cli/src/login.rs

@@ -1,3 +1,4 @@
+use codex_app_server_protocol::AuthMode;
 use codex_common::CliConfigOverrides;
 use codex_core::CodexAuth;
 use codex_core::auth::CLIENT_ID;
@@ -8,7 +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 codex_protocol::mcp_protocol::AuthMode;
+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

@@ -7,12 +7,12 @@ 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;
 use codex_cli::login::run_login_with_device_code;
 use codex_cli::login::run_logout;
-use codex_cli::proto;
 use codex_cloud_tasks::Cli as CloudTasksCli;
 use codex_common::CliConfigOverrides;
 use codex_exec::Cli as ExecCli;
@@ -26,7 +26,6 @@ use supports_color::Stream;
 mod mcp_cmd;
 
 use crate::mcp_cmd::McpCli;
-use crate::proto::ProtoCli;
 
 /// Codex CLI
 ///
@@ -74,15 +73,12 @@ enum Subcommand {
     /// [experimental] Run the app server.
     AppServer,
 
-    /// Run the Protocol stream via stdin/stdout
-    #[clap(visible_alias = "p")]
-    Proto(ProtoCli),
-
     /// 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")]
@@ -126,18 +122,20 @@ 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)]
@@ -145,7 +143,18 @@ 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)
@@ -224,25 +233,12 @@ fn print_exit_messages(exit_info: AppExitInfo) {
     }
 }
 
-pub(crate) const CODEX_SECURE_MODE_ENV_VAR: &str = "CODEX_SECURE_MODE";
-
-/// As early as possible in the process lifecycle, apply hardening measures
-/// if the CODEX_SECURE_MODE environment variable is set to "1".
+/// As early as possible in the process lifecycle, apply hardening measures. We
+/// skip this in debug builds to avoid interfering with debugging.
 #[ctor::ctor]
+#[cfg(not(debug_assertions))]
 fn pre_main_hardening() {
-    let secure_mode = match std::env::var(CODEX_SECURE_MODE_ENV_VAR) {
-        Ok(value) => value,
-        Err(_) => return,
-    };
-
-    if secure_mode == "1" {
-        codex_process_hardening::pre_main_hardening();
-    }
-
-    // Always clear this env var so child processes don't inherit it.
-    unsafe {
-        std::env::remove_var(CODEX_SECURE_MODE_ENV_VAR);
-    }
+    codex_process_hardening::pre_main_hardening();
 }
 
 fn main() -> anyhow::Result<()> {
@@ -298,7 +294,8 @@ 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(
@@ -317,7 +314,13 @@ 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;
@@ -332,13 +335,6 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
             );
             run_logout(logout_cli.config_overrides).await;
         }
-        Some(Subcommand::Proto(mut proto_cli)) => {
-            prepend_config_flags(
-                &mut proto_cli.config_overrides,
-                root_config_overrides.clone(),
-            );
-            proto::run_main(proto_cli).await?;
-        }
         Some(Subcommand::Completion(completion_cli)) => {
             print_completion(completion_cli);
         }
@@ -349,8 +345,8 @@ 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(),
@@ -361,7 +357,7 @@ 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(),
@@ -480,8 +476,9 @@ fn print_completion(cmd: CompletionCommand) {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use assert_matches::assert_matches;
     use codex_core::protocol::TokenUsage;
-    use codex_protocol::mcp_protocol::ConversationId;
+    use codex_protocol::ConversationId;
 
     fn finalize_from_args(args: &[&str]) -> TuiCli {
         let cli = MultitoolCli::try_parse_from(args).expect("parse");
@@ -612,14 +609,14 @@ 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

@@ -12,6 +12,8 @@ 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_rmcp_client::delete_oauth_tokens;
+use codex_rmcp_client::perform_oauth_login;
 
 /// [experimental] Launch Codex as an MCP server or manage configured MCP servers.
 ///
@@ -43,6 +45,14 @@ 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)]
@@ -82,6 +92,18 @@ 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 +113,22 @@ 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?;
             }
         }
 
@@ -108,7 +136,7 @@ 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))?;
 
@@ -134,6 +162,7 @@ fn run_add(config_overrides: &CliConfigOverrides, add_args: AddArgs) -> Result<(
 
     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 {
@@ -156,7 +185,7 @@ fn run_add(config_overrides: &CliConfigOverrides, add_args: AddArgs) -> Result<(
     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 +194,7 @@ 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();
@@ -183,9 +213,65 @@ 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.use_experimental_use_rmcp_client {
+        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).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) {
+        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();
@@ -343,9 +429,10 @@ 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 {

codex-rs/cli/src/proto.rs

@@ -1,133 +0,0 @@
-use std::io::IsTerminal;
-
-use clap::Parser;
-use codex_common::CliConfigOverrides;
-use codex_core::AuthManager;
-use codex_core::ConversationManager;
-use codex_core::NewConversation;
-use codex_core::config::Config;
-use codex_core::config::ConfigOverrides;
-use codex_core::protocol::Event;
-use codex_core::protocol::EventMsg;
-use codex_core::protocol::Submission;
-use tokio::io::AsyncBufReadExt;
-use tokio::io::BufReader;
-use tracing::error;
-use tracing::info;
-
-#[derive(Debug, Parser)]
-pub struct ProtoCli {
-    #[clap(skip)]
-    pub config_overrides: CliConfigOverrides,
-}
-
-pub async fn run_main(opts: ProtoCli) -> anyhow::Result<()> {
-    if std::io::stdin().is_terminal() {
-        anyhow::bail!("Protocol mode expects stdin to be a pipe, not a terminal");
-    }
-
-    tracing_subscriber::fmt()
-        .with_writer(std::io::stderr)
-        .init();
-
-    let ProtoCli { config_overrides } = opts;
-    let overrides_vec = config_overrides
-        .parse_overrides()
-        .map_err(anyhow::Error::msg)?;
-
-    let config = Config::load_with_cli_overrides(overrides_vec, ConfigOverrides::default())?;
-    // Use conversation_manager API to start a conversation
-    let conversation_manager =
-        ConversationManager::new(AuthManager::shared(config.codex_home.clone()));
-    let NewConversation {
-        conversation_id: _,
-        conversation,
-        session_configured,
-    } = conversation_manager.new_conversation(config).await?;
-
-    // Simulate streaming the session_configured event.
-    let synthetic_event = Event {
-        // Fake id value.
-        id: "".to_string(),
-        msg: EventMsg::SessionConfigured(session_configured),
-    };
-    let session_configured_event = match serde_json::to_string(&synthetic_event) {
-        Ok(s) => s,
-        Err(e) => {
-            error!("Failed to serialize session_configured: {e}");
-            return Err(anyhow::Error::from(e));
-        }
-    };
-    println!("{session_configured_event}");
-
-    // Task that reads JSON lines from stdin and forwards to Submission Queue
-    let sq_fut = {
-        let conversation = conversation.clone();
-        async move {
-            let stdin = BufReader::new(tokio::io::stdin());
-            let mut lines = stdin.lines();
-            loop {
-                let result = tokio::select! {
-                    _ = tokio::signal::ctrl_c() => {
-                        break
-                    },
-                    res = lines.next_line() => res,
-                };
-
-                match result {
-                    Ok(Some(line)) => {
-                        let line = line.trim();
-                        if line.is_empty() {
-                            continue;
-                        }
-                        match serde_json::from_str::<Submission>(line) {
-                            Ok(sub) => {
-                                if let Err(e) = conversation.submit_with_id(sub).await {
-                                    error!("{e:#}");
-                                    break;
-                                }
-                            }
-                            Err(e) => {
-                                error!("invalid submission: {e}");
-                            }
-                        }
-                    }
-                    _ => {
-                        info!("Submission queue closed");
-                        break;
-                    }
-                }
-            }
-        }
-    };
-
-    // Task that reads events from the agent and prints them as JSON lines to stdout
-    let eq_fut = async move {
-        loop {
-            let event = tokio::select! {
-                _ = tokio::signal::ctrl_c() => break,
-                event = conversation.next_event() => event,
-            };
-            match event {
-                Ok(event) => {
-                    let event_str = match serde_json::to_string(&event) {
-                        Ok(s) => s,
-                        Err(e) => {
-                            error!("Failed to serialize event: {e}");
-                            continue;
-                        }
-                    };
-                    println!("{event_str}");
-                }
-                Err(e) => {
-                    error!("{e:#}");
-                    break;
-                }
-            }
-        }
-        info!("Event queue closed");
-    };
-
-    tokio::join!(sq_fut, eq_fut);
-    Ok(())
-}

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 {
@@ -43,7 +43,7 @@ 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 +53,14 @@ 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 +80,7 @@ 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,

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/lib.rs

@@ -190,7 +190,7 @@ pub async fn run_main(_cli: Cli, _codex_linux_sandbox_exe: Option<PathBuf>) -> a
         // 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)
+            .map(|home| codex_login::AuthManager::new(home, false))
             .and_then(|am| am.auth())
         {
             Some(auth) => {
@@ -400,16 +400,20 @@ 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)?;
             }
@@ -839,6 +843,9 @@ pub async fn run_main(_cli: Cli, _codex_linux_sandbox_exe: Option<PathBuf>) -> a
                             && matches!(key.code, KeyCode::Char('n') | KeyCode::Char('N'))
                             || matches!(key.code, KeyCode::Char('\u{000E}'));
                         if is_ctrl_n {
+                            if app.new_task.is_none() {
+                                continue;
+                            }
                             if app.best_of_modal.is_some() {
                                 app.best_of_modal = None;
                                 needs_redraw = true;

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…");
     }
 }
 
@@ -262,9 +263,9 @@ fn draw_footer(frame: &mut Frame, area: Rect, app: &mut App) {
         help.push(": Apply  ".dim());
     }
     help.push("o : Set Env  ".dim());
-    help.push("Ctrl+N".dim());
-    help.push(format!(": Attempts {}x  ", app.best_of_n).dim());
     if app.new_task.is_some() {
+        help.push("Ctrl+N".dim());
+        help.push(format!(": Attempts {}x  ", app.best_of_n).dim());
         help.push("(editing new task)  ".dim());
     } else {
         help.push("n : New Task  ".dim());
@@ -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;
     }
 
@@ -1004,32 +1015,40 @@ pub fn draw_best_of_modal(frame: &mut Frame, area: Rect, app: &mut App) {
     use ratatui::widgets::Wrap;
 
     let inner = overlay_outer(area);
+    const MAX_WIDTH: u16 = 40;
+    const MIN_WIDTH: u16 = 20;
+    const MAX_HEIGHT: u16 = 12;
+    const MIN_HEIGHT: u16 = 6;
+    let modal_width = inner.width.min(MAX_WIDTH).max(inner.width.min(MIN_WIDTH));
+    let modal_height = inner
+        .height
+        .min(MAX_HEIGHT)
+        .max(inner.height.min(MIN_HEIGHT));
+    let modal_x = inner.x + (inner.width.saturating_sub(modal_width)) / 2;
+    let modal_y = inner.y + (inner.height.saturating_sub(modal_height)) / 2;
+    let modal_area = Rect::new(modal_x, modal_y, modal_width, modal_height);
     let title = Line::from(vec!["Parallel Attempts".magenta().bold()]);
     let block = overlay_block().title(title);
 
-    frame.render_widget(Clear, inner);
-    frame.render_widget(block.clone(), inner);
-    let content = overlay_content(inner);
+    frame.render_widget(Clear, modal_area);
+    frame.render_widget(block.clone(), modal_area);
+    let content = overlay_content(modal_area);
 
     let rows = Layout::default()
         .direction(Direction::Vertical)
         .constraints([Constraint::Length(2), Constraint::Min(1)])
         .split(content);
 
-    let hint = Paragraph::new(Line::from(
-        "Use ↑/↓ to choose, 1-4 jump; Enter confirm, Esc cancel"
-            .cyan()
-            .dim(),
-    ))
-    .wrap(Wrap { trim: true });
+    let hint = Paragraph::new(Line::from("Use ↑/↓ to choose, 1-4 jump".cyan().dim()))
+        .wrap(Wrap { trim: true });
     frame.render_widget(hint, rows[0]);
 
     let selected = app.best_of_modal.as_ref().map(|m| m.selected).unwrap_or(0);
     let options = [1usize, 2, 3, 4];
     let mut items: Vec<ListItem> = Vec::new();
     for &attempts in &options {
-        let mut spans: Vec<ratatui::text::Span> =
-            vec![format!("{attempts} attempt{}", if attempts == 1 { "" } else { "s" }).into()];
+        let noun = if attempts == 1 { "attempt" } else { "attempts" };
+        let mut spans: Vec<ratatui::text::Span> = vec![format!("{attempts} {noun:<8}").into()];
         spans.push("  ".into());
         spans.push(format!("{attempts}x parallel").dim());
         if attempts == app.best_of_n {

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()

codex-rs/common/Cargo.toml

@@ -10,6 +10,7 @@ workspace = true
 clap = { workspace = true, features = ["derive", "wrap_help"], optional = true }
 codex-core = { workspace = true }
 codex-protocol = { workspace = true }
+codex-app-server-protocol = { workspace = true }
 serde = { workspace = true, optional = true }
 toml = { workspace = true, optional = true }
 

codex-rs/common/src/model_presets.rs

@@ -1,5 +1,5 @@
+use codex_app_server_protocol::AuthMode;
 use codex_core::protocol_config_types::ReasoningEffort;
-use codex_protocol::mcp_protocol::AuthMode;
 
 /// A simple preset pairing a model slug with a reasoning effort.
 #[derive(Debug, Clone, Copy)]
@@ -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 +19,16 @@ 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-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 }
 eventsource-stream = { workspace = true }
 futures = { workspace = true }
@@ -58,7 +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"] }
@@ -73,6 +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"] }
@@ -83,16 +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

@@ -5,18 +5,19 @@ You are Codex, based on GPT-5. You are running as a coding agent in the Codex CL
 - The arguments to `shell` will be passed to execvp(). Most terminal commands should be prefixed with ["bash", "-lc"].
 - Always set the `workdir` param when using the shell function. Do not use `cd` unless absolutely necessary.
 - When searching for text or files, prefer using `rg` or `rg --files` respectively because `rg` is much faster than alternatives like `grep`. (If the `rg` command is not found, then use alternatives.)
-- When editing or creating files, you MUST use apply_patch as a standalone tool without going through ["bash", "-lc"], `Python`, `cat`, `sed`, ... Example: functions.shell({"command":["apply_patch","*** Begin Patch\nAdd File: hello.txt\n+Hello, world!\n*** End Patch"]}).
 
 ## Editing constraints
 
 - Default to ASCII when editing or creating files. Only introduce non-ASCII or other Unicode characters when there is a clear justification and the file already uses them.
 - Add succinct code comments that explain what is going on if code is not self-explanatory. You should not add comments like "Assigns the value to the variable", but a brief comment might be useful ahead of a complex code block that the user would otherwise have to spend time parsing out. Usage of these comments should be rare.
+- Try to use apply_patch for single file edits, but it is fine to explore other options to make the edit if it does not work well. Do not use apply_patch for changes that are auto-generated (i.e. generating package.json or running a lint or format command like gofmt) or when scripting is more efficient (such as search and replacing a string across a codebase).
 - You may be in a dirty git worktree.
     * NEVER revert existing changes you did not make unless explicitly requested, since these changes were made by the user.
     * If asked to make a commit or code edits and there are unrelated changes to your work or changes that you didn't make in those files, don't revert those changes.
     * If the changes are in files you've touched recently, you should read carefully and understand how you can work with the changes rather than reverting them.
     * If the changes are in unrelated files, just ignore them and don't revert them.
 - 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
 
@@ -90,7 +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/apply_patch.rs

@@ -27,6 +27,7 @@ pub(crate) enum InternalApplyPatchInvocation {
     DelegateToExec(ApplyPatchExec),
 }
 
+#[derive(Debug)]
 pub(crate) struct ApplyPatchExec {
     pub(crate) action: ApplyPatchAction,
     pub(crate) user_explicitly_approved_this_action: bool,
@@ -109,3 +110,28 @@ pub(crate) fn convert_apply_patch_to_protocol(
     }
     result
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use pretty_assertions::assert_eq;
+
+    use tempfile::tempdir;
+
+    #[test]
+    fn convert_apply_patch_maps_add_variant() {
+        let tmp = tempdir().expect("tmp");
+        let p = tmp.path().join("a.txt");
+        // Create an action with a single Add change
+        let action = ApplyPatchAction::new_add_for_test(&p, "hello".to_string());
+
+        let got = convert_apply_patch_to_protocol(&action);
+
+        assert_eq!(
+            got.get(&p),
+            Some(&FileChange::Add {
+                content: "hello".to_string()
+            })
+        );
+    }
+}

codex-rs/core/src/auth.rs

@@ -15,7 +15,7 @@ use std::sync::Arc;
 use std::sync::Mutex;
 use std::time::Duration;
 
-use codex_protocol::mcp_protocol::AuthMode;
+use codex_app_server_protocol::AuthMode;
 
 use crate::token_data::PlanType;
 use crate::token_data::TokenData;
@@ -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

@@ -6,6 +6,8 @@ use crate::client_common::ResponseEvent;
 use crate::client_common::ResponseStream;
 use crate::error::CodexErr;
 use crate::error::Result;
+use crate::error::RetryLimitReachedError;
+use crate::error::UnexpectedResponseError;
 use crate::model_family::ModelFamily;
 use crate::openai_tools::create_tools_json_for_chat_completions_api;
 use crate::util::backoff;
@@ -320,11 +322,18 @@ pub(crate) async fn stream_chat_completions(
                 let status = res.status();
                 if !(status == StatusCode::TOO_MANY_REQUESTS || status.is_server_error()) {
                     let body = (res.text().await).unwrap_or_default();
-                    return Err(CodexErr::UnexpectedStatus(status, body));
+                    return Err(CodexErr::UnexpectedStatus(UnexpectedResponseError {
+                        status,
+                        body,
+                        request_id: None,
+                    }));
                 }
 
                 if attempt > max_retries {
-                    return Err(CodexErr::RetryLimit(status));
+                    return Err(CodexErr::RetryLimit(RetryLimitReachedError {
+                        status,
+                        request_id: None,
+                    }));
                 }
 
                 let retry_after_secs = res

codex-rs/core/src/client.rs

@@ -5,9 +5,11 @@ use std::time::Duration;
 
 use crate::AuthManager;
 use crate::auth::CodexAuth;
+use crate::error::RetryLimitReachedError;
+use crate::error::UnexpectedResponseError;
 use bytes::Bytes;
-use codex_protocol::mcp_protocol::AuthMode;
-use codex_protocol::mcp_protocol::ConversationId;
+use codex_app_server_protocol::AuthMode;
+use codex_protocol::ConversationId;
 use eventsource_stream::Eventsource;
 use futures::prelude::*;
 use regex_lite::Regex;
@@ -29,6 +31,7 @@ use crate::client_common::Prompt;
 use crate::client_common::ResponseEvent;
 use crate::client_common::ResponseStream;
 use crate::client_common::ResponsesApiRequest;
+use crate::client_common::TurnType;
 use crate::client_common::create_reasoning_param_for_request;
 use crate::client_common::create_text_param_for_request;
 use crate::config::Config;
@@ -61,7 +64,6 @@ struct ErrorResponse {
 #[derive(Debug, Deserialize)]
 struct Error {
     r#type: Option<String>,
-    #[allow(dead_code)]
     code: Option<String>,
     message: Option<String>,
 
@@ -226,7 +228,7 @@ impl ModelClient {
             input: &input_with_instructions,
             tools: &tools_json,
             tool_choice: "auto",
-            parallel_tool_calls: false,
+            parallel_tool_calls: prompt.parallel_tool_calls,
             reasoning,
             store: azure_workaround,
             stream: true,
@@ -243,7 +245,7 @@ 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, prompt.turn_type)
                 .await
             {
                 Ok(stream) => {
@@ -271,6 +273,7 @@ impl ModelClient {
         attempt: u64,
         payload_json: &Value,
         auth_manager: &Option<Arc<AuthManager>>,
+        turn_type: TurnType,
     ) -> 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());
@@ -292,6 +295,13 @@ impl ModelClient {
             // Send session_id for compatibility.
             .header("conversation_id", self.conversation_id.to_string())
             .header("session_id", self.conversation_id.to_string())
+            .header(
+                "action_kind",
+                match turn_type {
+                    TurnType::Review => "review",
+                    TurnType::Regular => "turn",
+                },
+            )
             .header(reqwest::header::ACCEPT, "text/event-stream")
             .json(payload_json);
 
@@ -307,14 +317,17 @@ impl ModelClient {
             .log_request(attempt, || req_builder.send())
             .await;
 
+        let mut request_id = None;
         if let Ok(resp) = &res {
+            request_id = resp
+                .headers()
+                .get("cf-ray")
+                .map(|v| v.to_str().unwrap_or_default().to_string());
+
             trace!(
-                "Response status: {}, cf-ray: {}",
+                "Response status: {}, cf-ray: {:?}",
                 resp.status(),
-                resp.headers()
-                    .get("cf-ray")
-                    .map(|v| v.to_str().unwrap_or_default())
-                    .unwrap_or_default()
+                request_id
             );
         }
 
@@ -374,7 +387,11 @@ impl ModelClient {
                     // Surface the error body to callers. Use `unwrap_or_default` per Clippy.
                     let body = res.text().await.unwrap_or_default();
                     return Err(StreamAttemptError::Fatal(CodexErr::UnexpectedStatus(
-                        status, body,
+                        UnexpectedResponseError {
+                            status,
+                            body,
+                            request_id: None,
+                        },
                     )));
                 }
 
@@ -405,6 +422,7 @@ impl ModelClient {
                 Err(StreamAttemptError::RetryableHttpError {
                     status,
                     retry_after,
+                    request_id,
                 })
             }
             Err(e) => Err(StreamAttemptError::RetryableTransportError(e.into())),
@@ -448,6 +466,7 @@ enum StreamAttemptError {
     RetryableHttpError {
         status: StatusCode,
         retry_after: Option<Duration>,
+        request_id: Option<String>,
     },
     RetryableTransportError(CodexErr),
     Fatal(CodexErr),
@@ -472,11 +491,13 @@ impl StreamAttemptError {
 
     fn into_error(self) -> CodexErr {
         match self {
-            Self::RetryableHttpError { status, .. } => {
+            Self::RetryableHttpError {
+                status, request_id, ..
+            } => {
                 if status == StatusCode::INTERNAL_SERVER_ERROR {
                     CodexErr::InternalServerError
                 } else {
-                    CodexErr::RetryLimit(status)
+                    CodexErr::RetryLimit(RetryLimitReachedError { status, request_id })
                 }
             }
             Self::RetryableTransportError(error) => error,
@@ -781,9 +802,13 @@ 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}");
@@ -909,9 +934,14 @@ 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;
@@ -1166,6 +1196,74 @@ 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`
     // ────────────────────────────
@@ -1303,10 +1401,7 @@ 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\"");
@@ -1321,7 +1416,7 @@ 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;
@@ -21,6 +23,13 @@ use tokio::sync::mpsc;
 /// Review thread system prompt. Edit `core/src/review_prompt.md` to customize.
 pub const REVIEW_PROMPT: &str = include_str!("../review_prompt.md");
 
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
+pub enum TurnType {
+    #[default]
+    Regular,
+    Review,
+}
+
 /// API request payload for a single model turn
 #[derive(Default, Debug, Clone)]
 pub struct Prompt {
@@ -29,13 +38,19 @@ 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>,
+
+    /// The type of turn being executed
+    pub turn_type: TurnType,
 }
 
 impl Prompt {
@@ -49,8 +64,8 @@ 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 +79,125 @@ 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,
@@ -160,6 +290,65 @@ 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>,
@@ -279,7 +468,7 @@ mod tests {
             input: &input,
             tools: &tools,
             tool_choice: "auto",
-            parallel_tool_calls: false,
+            parallel_tool_calls: true,
             reasoning: None,
             store: false,
             stream: true,
@@ -320,7 +509,7 @@ mod tests {
             input: &input,
             tools: &tools,
             tool_choice: "auto",
-            parallel_tool_calls: false,
+            parallel_tool_calls: true,
             reasoning: None,
             store: false,
             stream: true,
@@ -356,7 +545,7 @@ mod tests {
             input: &input,
             tools: &tools,
             tool_choice: "auto",
-            parallel_tool_calls: false,
+            parallel_tool_calls: true,
             reasoning: None,
             store: false,
             stream: true,

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

@@ -103,6 +103,18 @@ async fn run_compact_task_inner(
             Err(CodexErr::Interrupted) => {
                 return;
             }
+            Err(e @ CodexErr::ContextWindowExceeded) => {
+                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;

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

@@ -1,25 +1,431 @@
-// This is a WIP. This will eventually contain a real list of common safe Windows commands.
-pub fn is_safe_command_windows(_command: &[String]) -> bool {
+use shlex::split as shlex_split;
+
+/// On Windows, we conservatively allow only clearly read-only PowerShell invocations
+/// that match a small safelist. Anything else (including direct CMD commands) is unsafe.
+pub fn is_safe_command_windows(command: &[String]) -> bool {
+    if let Some(commands) = try_parse_powershell_command_sequence(command) {
+        return commands
+            .iter()
+            .all(|cmd| is_safe_powershell_command(cmd.as_slice()));
+    }
+    // Only PowerShell invocations are allowed on Windows for now; anything else is unsafe.
+    false
+}
+
+/// Returns each command sequence if the invocation starts with a PowerShell binary.
+/// For example, the tokens from `pwsh Get-ChildItem | Measure-Object` become two sequences.
+fn try_parse_powershell_command_sequence(command: &[String]) -> Option<Vec<Vec<String>>> {
+    let (exe, rest) = command.split_first()?;
+    if !is_powershell_executable(exe) {
+        return None;
+    }
+    parse_powershell_invocation(rest)
+}
+
+/// Parses a PowerShell invocation into discrete command vectors, rejecting unsafe patterns.
+fn parse_powershell_invocation(args: &[String]) -> Option<Vec<Vec<String>>> {
+    if args.is_empty() {
+        // Examples rejected here: "pwsh" and "powershell.exe" with no additional arguments.
+        return None;
+    }
+
+    let mut idx = 0;
+    while idx < args.len() {
+        let arg = &args[idx];
+        let lower = arg.to_ascii_lowercase();
+        match lower.as_str() {
+            "-command" | "/command" | "-c" => {
+                let script = args.get(idx + 1)?;
+                if idx + 2 != args.len() {
+                    // Reject if there is more than one token representing the actual command.
+                    // Examples rejected here: "pwsh -Command foo bar" and "powershell -c ls extra".
+                    return None;
+                }
+                return parse_powershell_script(script);
+            }
+            _ if lower.starts_with("-command:") || lower.starts_with("/command:") => {
+                if idx + 1 != args.len() {
+                    // Reject if there are more tokens after the command itself.
+                    // Examples rejected here: "pwsh -Command:dir C:\\" and "powershell /Command:dir C:\\" with trailing args.
+                    return None;
+                }
+                let script = arg.split_once(':')?.1;
+                return parse_powershell_script(script);
+            }
+
+            // Benign, no-arg flags we tolerate.
+            "-nologo" | "-noprofile" | "-noninteractive" | "-mta" | "-sta" => {
+                idx += 1;
+                continue;
+            }
+
+            // Explicitly forbidden/opaque or unnecessary for read-only operations.
+            "-encodedcommand" | "-ec" | "-file" | "/file" | "-windowstyle" | "-executionpolicy"
+            | "-workingdirectory" => {
+                // Examples rejected here: "pwsh -EncodedCommand ..." and "powershell -File script.ps1".
+                return None;
+            }
+
+            // Unknown switch → bail conservatively.
+            _ if lower.starts_with('-') => {
+                // Examples rejected here: "pwsh -UnknownFlag" and "powershell -foo bar".
+                return None;
+            }
+
+            // If we hit non-flag tokens, treat the remainder as a command sequence.
+            // This happens if powershell is invoked without -Command, e.g.
+            // ["pwsh", "-NoLogo", "git", "-c", "core.pager=cat", "status"]
+            _ => {
+                return split_into_commands(args[idx..].to_vec());
+            }
+        }
+    }
+
+    // Examples rejected here: "pwsh" and "powershell.exe -NoLogo" without a script.
+    None
+}
+
+/// Tokenizes an inline PowerShell script and delegates to the command splitter.
+/// Examples of when this is called: pwsh.exe -Command '<script>' or pwsh.exe -Command:<script>
+fn parse_powershell_script(script: &str) -> Option<Vec<Vec<String>>> {
+    let tokens = shlex_split(script)?;
+    split_into_commands(tokens)
+}
+
+/// Splits tokens into pipeline segments while ensuring no unsafe separators slip through.
+/// e.g. Get-ChildItem | Measure-Object -> [['Get-ChildItem'], ['Measure-Object']]
+fn split_into_commands(tokens: Vec<String>) -> Option<Vec<Vec<String>>> {
+    if tokens.is_empty() {
+        // Examples rejected here: "pwsh -Command ''" and "powershell -Command \"\"".
+        return None;
+    }
+
+    let mut commands = Vec::new();
+    let mut current = Vec::new();
+    for token in tokens.into_iter() {
+        match token.as_str() {
+            "|" | "||" | "&&" | ";" => {
+                if current.is_empty() {
+                    // Examples rejected here: "pwsh -Command '| Get-ChildItem'" and "pwsh -Command '; dir'".
+                    return None;
+                }
+                commands.push(current);
+                current = Vec::new();
+            }
+            // Reject if any token embeds separators, redirection, or call operator characters.
+            _ if token.contains(['|', ';', '>', '<', '&']) || token.contains("$(") => {
+                // Examples rejected here: "pwsh -Command 'dir|select'" and "pwsh -Command 'echo hi > out.txt'".
+                return None;
+            }
+            _ => current.push(token),
+        }
+    }
+
+    if current.is_empty() {
+        // Examples rejected here: "pwsh -Command 'dir |'" and "pwsh -Command 'Get-ChildItem ;'".
+        return None;
+    }
+    commands.push(current);
+    Some(commands)
+}
+
+/// Returns true when the executable name is one of the supported PowerShell binaries.
+fn is_powershell_executable(exe: &str) -> bool {
+    matches!(
+        exe.to_ascii_lowercase().as_str(),
+        "powershell" | "powershell.exe" | "pwsh" | "pwsh.exe"
+    )
+}
+
+/// Validates that a parsed PowerShell command stays within our read-only safelist.
+/// Everything before this is parsing, and rejecting things that make us feel uncomfortable.
+fn is_safe_powershell_command(words: &[String]) -> bool {
+    if words.is_empty() {
+        // Examples rejected here: "pwsh -Command ''" and "pwsh -Command \"\"".
+        return false;
+    }
+
+    // Reject nested unsafe cmdlets inside parentheses or arguments
+    for w in words.iter() {
+        let inner = w
+            .trim_matches(|c| c == '(' || c == ')')
+            .trim_start_matches('-')
+            .to_ascii_lowercase();
+        if matches!(
+            inner.as_str(),
+            "set-content"
+                | "add-content"
+                | "out-file"
+                | "new-item"
+                | "remove-item"
+                | "move-item"
+                | "copy-item"
+                | "rename-item"
+                | "start-process"
+                | "stop-process"
+        ) {
+            // Examples rejected here: "Write-Output (Set-Content foo6.txt 'abc')" and "Get-Content (New-Item bar.txt)".
+            return false;
+        }
+    }
+
+    // Block PowerShell call operator or any redirection explicitly.
+    if words.iter().any(|w| {
+        matches!(
+            w.as_str(),
+            "&" | ">" | ">>" | "1>" | "2>" | "2>&1" | "*>" | "<" | "<<"
+        )
+    }) {
+        // Examples rejected here: "pwsh -Command '& Remove-Item foo'" and "pwsh -Command 'Get-Content foo > bar'".
+        return false;
+    }
+
+    let command = words[0]
+        .trim_matches(|c| c == '(' || c == ')')
+        .trim_start_matches('-')
+        .to_ascii_lowercase();
+    match command.as_str() {
+        "echo" | "write-output" | "write-host" => true, // (no redirection allowed)
+        "dir" | "ls" | "get-childitem" | "gci" => true,
+        "cat" | "type" | "gc" | "get-content" => true,
+        "select-string" | "sls" | "findstr" => true,
+        "measure-object" | "measure" => true,
+        "get-location" | "gl" | "pwd" => true,
+        "test-path" | "tp" => true,
+        "resolve-path" | "rvpa" => true,
+        "select-object" | "select" => true,
+        "get-item" => true,
+
+        "git" => is_safe_git_command(words),
+
+        "rg" => is_safe_ripgrep(words),
+
+        // Extra safety: explicitly prohibit common side-effecting cmdlets regardless of args.
+        "set-content" | "add-content" | "out-file" | "new-item" | "remove-item" | "move-item"
+        | "copy-item" | "rename-item" | "start-process" | "stop-process" => {
+            // Examples rejected here: "pwsh -Command 'Set-Content notes.txt data'" and "pwsh -Command 'Remove-Item temp.log'".
+            false
+        }
+
+        _ => {
+            // Examples rejected here: "pwsh -Command 'Invoke-WebRequest https://example.com'" and "pwsh -Command 'Start-Service Spooler'".
+            false
+        }
+    }
+}
+
+/// Checks that an `rg` invocation avoids options that can spawn arbitrary executables.
+fn is_safe_ripgrep(words: &[String]) -> bool {
+    const UNSAFE_RIPGREP_OPTIONS_WITH_ARGS: &[&str] = &["--pre", "--hostname-bin"];
+    const UNSAFE_RIPGREP_OPTIONS_WITHOUT_ARGS: &[&str] = &["--search-zip", "-z"];
+
+    !words.iter().skip(1).any(|arg| {
+        let arg_lc = arg.to_ascii_lowercase();
+        // Examples rejected here: "pwsh -Command 'rg --pre cat pattern'" and "pwsh -Command 'rg --search-zip pattern'".
+        UNSAFE_RIPGREP_OPTIONS_WITHOUT_ARGS.contains(&arg_lc.as_str())
+            || UNSAFE_RIPGREP_OPTIONS_WITH_ARGS
+                .iter()
+                .any(|opt| arg_lc == *opt || arg_lc.starts_with(&format!("{opt}=")))
+    })
+}
+
+/// Ensures a Git command sticks to whitelisted read-only subcommands and flags.
+fn is_safe_git_command(words: &[String]) -> bool {
+    const SAFE_SUBCOMMANDS: &[&str] = &["status", "log", "show", "diff", "cat-file"];
+
+    let mut iter = words.iter().skip(1);
+    while let Some(arg) = iter.next() {
+        let arg_lc = arg.to_ascii_lowercase();
+
+        if arg.starts_with('-') {
+            if arg.eq_ignore_ascii_case("-c") || arg.eq_ignore_ascii_case("--config") {
+                if iter.next().is_none() {
+                    // Examples rejected here: "pwsh -Command 'git -c'" and "pwsh -Command 'git --config'".
+                    return false;
+                }
+                continue;
+            }
+
+            if arg_lc.starts_with("-c=")
+                || arg_lc.starts_with("--config=")
+                || arg_lc.starts_with("--git-dir=")
+                || arg_lc.starts_with("--work-tree=")
+            {
+                continue;
+            }
+
+            if arg.eq_ignore_ascii_case("--git-dir") || arg.eq_ignore_ascii_case("--work-tree") {
+                if iter.next().is_none() {
+                    // Examples rejected here: "pwsh -Command 'git --git-dir'" and "pwsh -Command 'git --work-tree'".
+                    return false;
+                }
+                continue;
+            }
+
+            continue;
+        }
+
+        return SAFE_SUBCOMMANDS.contains(&arg_lc.as_str());
+    }
+
+    // Examples rejected here: "pwsh -Command 'git'" and "pwsh -Command 'git status --short | Remove-Item foo'".
     false
 }
 
 #[cfg(test)]
 mod tests {
     use super::is_safe_command_windows;
+    use std::string::ToString;
 
+    /// Converts a slice of string literals into owned `String`s for the tests.
     fn vec_str(args: &[&str]) -> Vec<String> {
         args.iter().map(ToString::to_string).collect()
     }
 
     #[test]
-    fn everything_is_unsafe() {
-        for cmd in [
-            vec_str(&["powershell.exe", "-NoLogo", "-Command", "echo hello"]),
-            vec_str(&["copy", "foo", "bar"]),
-            vec_str(&["del", "file.txt"]),
-            vec_str(&["powershell.exe", "Get-ChildItem"]),
-        ] {
-            assert!(!is_safe_command_windows(&cmd));
-        }
+    fn recognizes_safe_powershell_wrappers() {
+        assert!(is_safe_command_windows(&vec_str(&[
+            "powershell.exe",
+            "-NoLogo",
+            "-Command",
+            "Get-ChildItem -Path .",
+        ])));
+
+        assert!(is_safe_command_windows(&vec_str(&[
+            "powershell.exe",
+            "-NoProfile",
+            "-Command",
+            "git status",
+        ])));
+
+        assert!(is_safe_command_windows(&vec_str(&[
+            "powershell.exe",
+            "Get-Content",
+            "Cargo.toml",
+        ])));
+
+        // pwsh parity
+        assert!(is_safe_command_windows(&vec_str(&[
+            "pwsh.exe",
+            "-NoProfile",
+            "-Command",
+            "Get-ChildItem",
+        ])));
+    }
+
+    #[test]
+    fn allows_read_only_pipelines_and_git_usage() {
+        assert!(is_safe_command_windows(&vec_str(&[
+            "pwsh",
+            "-NoLogo",
+            "-NoProfile",
+            "-Command",
+            "rg --files-with-matches foo | Measure-Object | Select-Object -ExpandProperty Count",
+        ])));
+
+        assert!(is_safe_command_windows(&vec_str(&[
+            "pwsh",
+            "-NoLogo",
+            "-NoProfile",
+            "-Command",
+            "Get-Content foo.rs | Select-Object -Skip 200",
+        ])));
+
+        assert!(is_safe_command_windows(&vec_str(&[
+            "pwsh",
+            "-NoLogo",
+            "-NoProfile",
+            "-Command",
+            "git -c core.pager=cat show HEAD:foo.rs",
+        ])));
+
+        assert!(is_safe_command_windows(&vec_str(&[
+            "pwsh",
+            "-Command",
+            "-git cat-file -p HEAD:foo.rs",
+        ])));
+
+        assert!(is_safe_command_windows(&vec_str(&[
+            "pwsh",
+            "-Command",
+            "(Get-Content foo.rs -Raw)",
+        ])));
+
+        assert!(is_safe_command_windows(&vec_str(&[
+            "pwsh",
+            "-Command",
+            "Get-Item foo.rs | Select-Object Length",
+        ])));
+    }
+
+    #[test]
+    fn rejects_powershell_commands_with_side_effects() {
+        assert!(!is_safe_command_windows(&vec_str(&[
+            "powershell.exe",
+            "-NoLogo",
+            "-Command",
+            "Remove-Item foo.txt",
+        ])));
+
+        assert!(!is_safe_command_windows(&vec_str(&[
+            "powershell.exe",
+            "-NoProfile",
+            "-Command",
+            "rg --pre cat",
+        ])));
+
+        assert!(!is_safe_command_windows(&vec_str(&[
+            "powershell.exe",
+            "-Command",
+            "Set-Content foo.txt 'hello'",
+        ])));
+
+        // Redirections are blocked
+        assert!(!is_safe_command_windows(&vec_str(&[
+            "powershell.exe",
+            "-Command",
+            "echo hi > out.txt",
+        ])));
+        assert!(!is_safe_command_windows(&vec_str(&[
+            "powershell.exe",
+            "-Command",
+            "Get-Content x | Out-File y",
+        ])));
+        assert!(!is_safe_command_windows(&vec_str(&[
+            "powershell.exe",
+            "-Command",
+            "Write-Output foo 2> err.txt",
+        ])));
+
+        // Call operator is blocked
+        assert!(!is_safe_command_windows(&vec_str(&[
+            "powershell.exe",
+            "-Command",
+            "& Remove-Item foo",
+        ])));
+
+        // Chained safe + unsafe must fail
+        assert!(!is_safe_command_windows(&vec_str(&[
+            "powershell.exe",
+            "-Command",
+            "Get-ChildItem; Remove-Item foo",
+        ])));
+        // Nested unsafe cmdlet inside safe command must fail
+        assert!(!is_safe_command_windows(&vec_str(&[
+            "powershell.exe",
+            "-Command",
+            "Write-Output (Set-Content foo6.txt 'abc')",
+        ])));
+        // Additional nested unsafe cmdlet examples must fail
+        assert!(!is_safe_command_windows(&vec_str(&[
+            "powershell.exe",
+            "-Command",
+            "Write-Host (Remove-Item foo.txt)",
+        ])));
+        assert!(!is_safe_command_windows(&vec_str(&[
+            "powershell.exe",
+            "-Command",
+            "Get-Content (New-Item bar.txt)",
+        ])));
     }
 }

codex-rs/core/src/config.rs

@@ -1,3 +1,7 @@
+use crate::config_loader::LoadedConfigLayers;
+pub use crate::config_loader::load_config_as_toml;
+use crate::config_loader::load_config_layers_with_overrides;
+use crate::config_loader::merge_toml_values;
 use crate::config_profile::ConfigProfile;
 use crate::config_types::DEFAULT_OTEL_ENVIRONMENT;
 use crate::config_types::History;
@@ -23,12 +27,12 @@ use crate::openai_model_info::get_model_info;
 use crate::protocol::AskForApproval;
 use crate::protocol::SandboxPolicy;
 use anyhow::Context;
+use codex_app_server_protocol::Tools;
+use codex_app_server_protocol::UserSavedConfig;
 use codex_protocol::config_types::ReasoningEffort;
 use codex_protocol::config_types::ReasoningSummary;
 use codex_protocol::config_types::SandboxMode;
 use codex_protocol::config_types::Verbosity;
-use codex_protocol::mcp_protocol::Tools;
-use codex_protocol::mcp_protocol::UserSavedConfig;
 use dirs::home_dir;
 use serde::Deserialize;
 use std::collections::BTreeMap;
@@ -42,7 +46,10 @@ use toml_edit::DocumentMut;
 use toml_edit::Item as TomlItem;
 use toml_edit::Table as TomlTable;
 
-const OPENAI_DEFAULT_MODEL: &str = "gpt-5-codex";
+#[cfg(target_os = "windows")]
+pub const OPENAI_DEFAULT_MODEL: &str = "gpt-5";
+#[cfg(not(target_os = "windows"))]
+pub const OPENAI_DEFAULT_MODEL: &str = "gpt-5-codex";
 const OPENAI_DEFAULT_REVIEW_MODEL: &str = "gpt-5-codex";
 pub const GPT_5_CODEX_MEDIUM_MODEL: &str = "gpt-5-codex";
 
@@ -141,6 +148,9 @@ pub struct Config {
     /// Maximum number of bytes to include from an AGENTS.md project doc file.
     pub project_doc_max_bytes: usize,
 
+    /// Additional filenames to try when looking for project-level docs.
+    pub project_doc_fallback_filenames: Vec<String>,
+
     /// Directory containing all Codex state (defaults to `~/.codex` but can be
     /// overridden by the `CODEX_HOME` environment variable).
     pub codex_home: PathBuf,
@@ -199,6 +209,9 @@ pub struct Config {
     /// The active profile name used to derive this `Config` (if any).
     pub active_profile: Option<String>,
 
+    /// Tracks whether the Windows onboarding screen has been acknowledged.
+    pub windows_wsl_setup_acknowledged: bool,
+
     /// When true, disables burst-paste detection for typed input entirely.
     /// All characters are inserted as they are received, and no buffering
     /// or placeholder replacement will occur for fast keypress bursts.
@@ -209,50 +222,38 @@ pub struct Config {
 }
 
 impl Config {
-    /// Load configuration with *generic* CLI overrides (`-c key=value`) applied
-    /// **in between** the values parsed from `config.toml` and the
-    /// strongly-typed overrides specified via [`ConfigOverrides`].
-    ///
-    /// The precedence order is therefore: `config.toml` < `-c` overrides <
-    /// `ConfigOverrides`.
-    pub fn load_with_cli_overrides(
+    pub async fn load_with_cli_overrides(
         cli_overrides: Vec<(String, TomlValue)>,
         overrides: ConfigOverrides,
     ) -> std::io::Result<Self> {
-        // Resolve the directory that stores Codex state (e.g. ~/.codex or the
-        // value of $CODEX_HOME) so we can embed it into the resulting
-        // `Config` instance.
         let codex_home = find_codex_home()?;
 
-        // Step 1: parse `config.toml` into a generic JSON value.
-        let mut root_value = load_config_as_toml(&codex_home)?;
+        let root_value = load_resolved_config(
+            &codex_home,
+            cli_overrides,
+            crate::config_loader::LoaderOverrides::default(),
+        )
+        .await?;
 
-        // Step 2: apply the `-c` overrides.
-        for (path, value) in cli_overrides.into_iter() {
-            apply_toml_override(&mut root_value, &path, value);
-        }
-
-        // Step 3: deserialize into `ConfigToml` so that Serde can enforce the
-        // correct types.
         let cfg: ConfigToml = root_value.try_into().map_err(|e| {
             tracing::error!("Failed to deserialize overridden config: {e}");
             std::io::Error::new(std::io::ErrorKind::InvalidData, e)
         })?;
 
-        // Step 4: merge with the strongly-typed overrides.
         Self::load_from_base_config_with_overrides(cfg, overrides, codex_home)
     }
 }
 
-pub fn load_config_as_toml_with_cli_overrides(
+pub async fn load_config_as_toml_with_cli_overrides(
     codex_home: &Path,
     cli_overrides: Vec<(String, TomlValue)>,
 ) -> std::io::Result<ConfigToml> {
-    let mut root_value = load_config_as_toml(codex_home)?;
-
-    for (path, value) in cli_overrides.into_iter() {
-        apply_toml_override(&mut root_value, &path, value);
-    }
+    let root_value = load_resolved_config(
+        codex_home,
+        cli_overrides,
+        crate::config_loader::LoaderOverrides::default(),
+    )
+    .await?;
 
     let cfg: ConfigToml = root_value.try_into().map_err(|e| {
         tracing::error!("Failed to deserialize overridden config: {e}");
@@ -262,33 +263,40 @@ pub fn load_config_as_toml_with_cli_overrides(
     Ok(cfg)
 }
 
-/// Read `CODEX_HOME/config.toml` and return it as a generic TOML value. Returns
-/// an empty TOML table when the file does not exist.
-pub fn load_config_as_toml(codex_home: &Path) -> std::io::Result<TomlValue> {
-    let config_path = codex_home.join(CONFIG_TOML_FILE);
-    match std::fs::read_to_string(&config_path) {
-        Ok(contents) => match toml::from_str::<TomlValue>(&contents) {
-            Ok(val) => Ok(val),
-            Err(e) => {
-                tracing::error!("Failed to parse config.toml: {e}");
-                Err(std::io::Error::new(std::io::ErrorKind::InvalidData, e))
-            }
-        },
-        Err(e) if e.kind() == std::io::ErrorKind::NotFound => {
-            tracing::info!("config.toml not found, using defaults");
-            Ok(TomlValue::Table(Default::default()))
-        }
-        Err(e) => {
-            tracing::error!("Failed to read config.toml: {e}");
-            Err(e)
-        }
-    }
+async fn load_resolved_config(
+    codex_home: &Path,
+    cli_overrides: Vec<(String, TomlValue)>,
+    overrides: crate::config_loader::LoaderOverrides,
+) -> std::io::Result<TomlValue> {
+    let layers = load_config_layers_with_overrides(codex_home, overrides).await?;
+    Ok(apply_overlays(layers, cli_overrides))
 }
 
-pub fn load_global_mcp_servers(
+fn apply_overlays(
+    layers: LoadedConfigLayers,
+    cli_overrides: Vec<(String, TomlValue)>,
+) -> TomlValue {
+    let LoadedConfigLayers {
+        mut base,
+        managed_config,
+        managed_preferences,
+    } = layers;
+
+    for (path, value) in cli_overrides.into_iter() {
+        apply_toml_override(&mut base, &path, value);
+    }
+
+    for overlay in [managed_config, managed_preferences].into_iter().flatten() {
+        merge_toml_values(&mut base, &overlay);
+    }
+
+    base
+}
+
+pub async fn load_global_mcp_servers(
     codex_home: &Path,
 ) -> std::io::Result<BTreeMap<String, McpServerConfig>> {
-    let root_value = load_config_as_toml(codex_home)?;
+    let root_value = load_config_as_toml(codex_home).await?;
     let Some(servers_value) = root_value.get("mcp_servers") else {
         return Ok(BTreeMap::new());
     };
@@ -466,6 +474,29 @@ pub fn set_project_trusted(codex_home: &Path, project_path: &Path) -> anyhow::Re
     Ok(())
 }
 
+/// Persist the acknowledgement flag for the Windows onboarding screen.
+pub fn set_windows_wsl_setup_acknowledged(
+    codex_home: &Path,
+    acknowledged: bool,
+) -> anyhow::Result<()> {
+    let config_path = codex_home.join(CONFIG_TOML_FILE);
+    let mut doc = match std::fs::read_to_string(config_path.clone()) {
+        Ok(s) => s.parse::<DocumentMut>()?,
+        Err(e) if e.kind() == std::io::ErrorKind::NotFound => DocumentMut::new(),
+        Err(e) => return Err(e.into()),
+    };
+
+    doc["windows_wsl_setup_acknowledged"] = toml_edit::value(acknowledged);
+
+    std::fs::create_dir_all(codex_home)?;
+
+    let tmp_file = NamedTempFile::new_in(codex_home)?;
+    std::fs::write(tmp_file.path(), doc.to_string())?;
+    tmp_file.persist(config_path)?;
+
+    Ok(())
+}
+
 fn ensure_profile_table<'a>(
     doc: &'a mut DocumentMut,
     profile_name: &str,
@@ -670,6 +701,9 @@ pub struct ConfigToml {
     /// Maximum number of bytes to include from an AGENTS.md project doc file.
     pub project_doc_max_bytes: Option<usize>,
 
+    /// Ordered list of fallback filenames to look for when AGENTS.md is missing.
+    pub project_doc_fallback_filenames: Option<Vec<String>>,
+
     /// Profile to use from the `profiles` map.
     pub profile: Option<String>,
 
@@ -716,6 +750,7 @@ pub struct ConfigToml {
     pub experimental_use_exec_command_tool: Option<bool>,
     pub experimental_use_unified_exec_tool: Option<bool>,
     pub experimental_use_rmcp_client: Option<bool>,
+    pub experimental_use_freeform_apply_patch: Option<bool>,
 
     pub projects: Option<HashMap<String, ProjectConfig>>,
 
@@ -729,6 +764,9 @@ pub struct ConfigToml {
 
     /// OTEL configuration.
     pub otel: Option<crate::config_types::OtelConfigToml>,
+
+    /// Tracks whether the Windows onboarding screen has been acknowledged.
+    pub windows_wsl_setup_acknowledged: Option<bool>,
 }
 
 impl From<ConfigToml> for UserSavedConfig {
@@ -1038,6 +1076,19 @@ impl Config {
             mcp_servers: cfg.mcp_servers,
             model_providers,
             project_doc_max_bytes: cfg.project_doc_max_bytes.unwrap_or(PROJECT_DOC_MAX_BYTES),
+            project_doc_fallback_filenames: cfg
+                .project_doc_fallback_filenames
+                .unwrap_or_default()
+                .into_iter()
+                .filter_map(|name| {
+                    let trimmed = name.trim();
+                    if trimmed.is_empty() {
+                        None
+                    } else {
+                        Some(trimmed.to_string())
+                    }
+                })
+                .collect(),
             codex_home,
             history,
             file_opener: cfg.file_opener.unwrap_or(UriBasedFileOpener::VsCode),
@@ -1061,7 +1112,9 @@ impl Config {
                 .or(cfg.chatgpt_base_url)
                 .unwrap_or("https://chatgpt.com/backend-api/".to_string()),
             include_plan_tool: include_plan_tool.unwrap_or(false),
-            include_apply_patch_tool: include_apply_patch_tool.unwrap_or(false),
+            include_apply_patch_tool: include_apply_patch_tool
+                .or(cfg.experimental_use_freeform_apply_patch)
+                .unwrap_or(false),
             tools_web_search_request,
             use_experimental_streamable_shell_tool: cfg
                 .experimental_use_exec_command_tool
@@ -1072,6 +1125,7 @@ impl Config {
             use_experimental_use_rmcp_client: cfg.experimental_use_rmcp_client.unwrap_or(false),
             include_view_image_tool,
             active_profile: active_profile_name,
+            windows_wsl_setup_acknowledged: cfg.windows_wsl_setup_acknowledged.unwrap_or(false),
             disable_paste_burst: cfg.disable_paste_burst.unwrap_or(false),
             tui_notifications: cfg
                 .tui
@@ -1310,18 +1364,18 @@ exclude_slash_tmp = true
         );
     }
 
-    #[test]
-    fn load_global_mcp_servers_returns_empty_if_missing() -> anyhow::Result<()> {
+    #[tokio::test]
+    async fn load_global_mcp_servers_returns_empty_if_missing() -> anyhow::Result<()> {
         let codex_home = TempDir::new()?;
 
-        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 write_global_mcp_servers_round_trips_entries() -> anyhow::Result<()> {
+    #[tokio::test]
+    async fn write_global_mcp_servers_round_trips_entries() -> anyhow::Result<()> {
         let codex_home = TempDir::new()?;
 
         let mut servers = BTreeMap::new();
@@ -1340,7 +1394,7 @@ exclude_slash_tmp = true
 
         write_global_mcp_servers(codex_home.path(), &servers)?;
 
-        let loaded = load_global_mcp_servers(codex_home.path())?;
+        let loaded = load_global_mcp_servers(codex_home.path()).await?;
         assert_eq!(loaded.len(), 1);
         let docs = loaded.get("docs").expect("docs entry");
         match &docs.transport {
@@ -1356,14 +1410,47 @@ exclude_slash_tmp = true
 
         let empty = BTreeMap::new();
         write_global_mcp_servers(codex_home.path(), &empty)?;
-        let loaded = load_global_mcp_servers(codex_home.path())?;
+        let loaded = load_global_mcp_servers(codex_home.path()).await?;
         assert!(loaded.is_empty());
 
         Ok(())
     }
 
-    #[test]
-    fn load_global_mcp_servers_accepts_legacy_ms_field() -> anyhow::Result<()> {
+    #[tokio::test]
+    async fn managed_config_wins_over_cli_overrides() -> anyhow::Result<()> {
+        let codex_home = TempDir::new()?;
+        let managed_path = codex_home.path().join("managed_config.toml");
+
+        std::fs::write(
+            codex_home.path().join(CONFIG_TOML_FILE),
+            "model = \"base\"\n",
+        )?;
+        std::fs::write(&managed_path, "model = \"managed_config\"\n")?;
+
+        let overrides = crate::config_loader::LoaderOverrides {
+            managed_config_path: Some(managed_path),
+            #[cfg(target_os = "macos")]
+            managed_preferences_base64: None,
+        };
+
+        let root_value = load_resolved_config(
+            codex_home.path(),
+            vec![("model".to_string(), TomlValue::String("cli".to_string()))],
+            overrides,
+        )
+        .await?;
+
+        let cfg: ConfigToml = root_value.try_into().map_err(|e| {
+            tracing::error!("Failed to deserialize overridden config: {e}");
+            std::io::Error::new(std::io::ErrorKind::InvalidData, e)
+        })?;
+
+        assert_eq!(cfg.model.as_deref(), Some("managed_config"));
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn load_global_mcp_servers_accepts_legacy_ms_field() -> anyhow::Result<()> {
         let codex_home = TempDir::new()?;
         let config_path = codex_home.path().join(CONFIG_TOML_FILE);
 
@@ -1377,15 +1464,15 @@ startup_timeout_ms = 2500
 "#,
         )?;
 
-        let servers = load_global_mcp_servers(codex_home.path())?;
+        let servers = load_global_mcp_servers(codex_home.path()).await?;
         let docs = servers.get("docs").expect("docs entry");
         assert_eq!(docs.startup_timeout_sec, Some(Duration::from_millis(2500)));
 
         Ok(())
     }
 
-    #[test]
-    fn write_global_mcp_servers_serializes_env_sorted() -> anyhow::Result<()> {
+    #[tokio::test]
+    async fn write_global_mcp_servers_serializes_env_sorted() -> anyhow::Result<()> {
         let codex_home = TempDir::new()?;
 
         let servers = BTreeMap::from([(
@@ -1420,7 +1507,7 @@ ZIG_VAR = "3"
 "#
         );
 
-        let loaded = load_global_mcp_servers(codex_home.path())?;
+        let loaded = load_global_mcp_servers(codex_home.path()).await?;
         let docs = loaded.get("docs").expect("docs entry");
         match &docs.transport {
             McpServerTransportConfig::Stdio { command, args, env } => {
@@ -1438,8 +1525,8 @@ ZIG_VAR = "3"
         Ok(())
     }
 
-    #[test]
-    fn write_global_mcp_servers_serializes_streamable_http() -> anyhow::Result<()> {
+    #[tokio::test]
+    async fn write_global_mcp_servers_serializes_streamable_http() -> anyhow::Result<()> {
         let codex_home = TempDir::new()?;
 
         let mut servers = BTreeMap::from([(
@@ -1467,7 +1554,7 @@ startup_timeout_sec = 2.0
 "#
         );
 
-        let loaded = load_global_mcp_servers(codex_home.path())?;
+        let loaded = load_global_mcp_servers(codex_home.path()).await?;
         let docs = loaded.get("docs").expect("docs entry");
         match &docs.transport {
             McpServerTransportConfig::StreamableHttp { url, bearer_token } => {
@@ -1499,7 +1586,7 @@ url = "https://example.com/mcp"
 "#
         );
 
-        let loaded = load_global_mcp_servers(codex_home.path())?;
+        let loaded = load_global_mcp_servers(codex_home.path()).await?;
         let docs = loaded.get("docs").expect("docs entry");
         match &docs.transport {
             McpServerTransportConfig::StreamableHttp { url, bearer_token } => {
@@ -1811,6 +1898,7 @@ model_verbosity = "high"
                 mcp_servers: HashMap::new(),
                 model_providers: fixture.model_provider_map.clone(),
                 project_doc_max_bytes: PROJECT_DOC_MAX_BYTES,
+                project_doc_fallback_filenames: Vec::new(),
                 codex_home: fixture.codex_home(),
                 history: History::default(),
                 file_opener: UriBasedFileOpener::VsCode,
@@ -1830,6 +1918,7 @@ model_verbosity = "high"
                 use_experimental_use_rmcp_client: false,
                 include_view_image_tool: true,
                 active_profile: Some("o3".to_string()),
+                windows_wsl_setup_acknowledged: false,
                 disable_paste_burst: false,
                 tui_notifications: Default::default(),
                 otel: OtelConfig::default(),
@@ -1871,6 +1960,7 @@ model_verbosity = "high"
             mcp_servers: HashMap::new(),
             model_providers: fixture.model_provider_map.clone(),
             project_doc_max_bytes: PROJECT_DOC_MAX_BYTES,
+            project_doc_fallback_filenames: Vec::new(),
             codex_home: fixture.codex_home(),
             history: History::default(),
             file_opener: UriBasedFileOpener::VsCode,
@@ -1890,6 +1980,7 @@ model_verbosity = "high"
             use_experimental_use_rmcp_client: false,
             include_view_image_tool: true,
             active_profile: Some("gpt3".to_string()),
+            windows_wsl_setup_acknowledged: false,
             disable_paste_burst: false,
             tui_notifications: Default::default(),
             otel: OtelConfig::default(),
@@ -1946,6 +2037,7 @@ model_verbosity = "high"
             mcp_servers: HashMap::new(),
             model_providers: fixture.model_provider_map.clone(),
             project_doc_max_bytes: PROJECT_DOC_MAX_BYTES,
+            project_doc_fallback_filenames: Vec::new(),
             codex_home: fixture.codex_home(),
             history: History::default(),
             file_opener: UriBasedFileOpener::VsCode,
@@ -1965,6 +2057,7 @@ model_verbosity = "high"
             use_experimental_use_rmcp_client: false,
             include_view_image_tool: true,
             active_profile: Some("zdr".to_string()),
+            windows_wsl_setup_acknowledged: false,
             disable_paste_burst: false,
             tui_notifications: Default::default(),
             otel: OtelConfig::default(),
@@ -2007,6 +2100,7 @@ model_verbosity = "high"
             mcp_servers: HashMap::new(),
             model_providers: fixture.model_provider_map.clone(),
             project_doc_max_bytes: PROJECT_DOC_MAX_BYTES,
+            project_doc_fallback_filenames: Vec::new(),
             codex_home: fixture.codex_home(),
             history: History::default(),
             file_opener: UriBasedFileOpener::VsCode,
@@ -2026,6 +2120,7 @@ model_verbosity = "high"
             use_experimental_use_rmcp_client: false,
             include_view_image_tool: true,
             active_profile: Some("gpt5".to_string()),
+            windows_wsl_setup_acknowledged: false,
             disable_paste_burst: false,
             tui_notifications: Default::default(),
             otel: OtelConfig::default(),
@@ -2136,6 +2231,7 @@ trust_level = "trusted"
 #[cfg(test)]
 mod notifications_tests {
     use crate::config_types::Notifications;
+    use assert_matches::assert_matches;
     use serde::Deserialize;
 
     #[derive(Deserialize, Debug, PartialEq)]
@@ -2155,10 +2251,7 @@ mod notifications_tests {
             notifications = true
         "#;
         let parsed: RootTomlTest = toml::from_str(toml).expect("deserialize notifications=true");
-        assert!(matches!(
-            parsed.tui.notifications,
-            Notifications::Enabled(true)
-        ));
+        assert_matches!(parsed.tui.notifications, Notifications::Enabled(true));
     }
 
     #[test]
@@ -2169,9 +2262,9 @@ mod notifications_tests {
         "#;
         let parsed: RootTomlTest =
             toml::from_str(toml).expect("deserialize notifications=[\"foo\"]");
-        assert!(matches!(
+        assert_matches!(
             parsed.tui.notifications,
             Notifications::Custom(ref v) if v == &vec!["foo".to_string()]
-        ));
+        );
     }
 }

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

@@ -22,7 +22,7 @@ pub struct ConfigProfile {
     pub experimental_instructions_file: Option<PathBuf>,
 }
 
-impl From<ConfigProfile> for codex_protocol::mcp_protocol::Profile {
+impl From<ConfigProfile> for codex_app_server_protocol::Profile {
     fn from(config_profile: ConfigProfile) -> Self {
         Self {
             model: config_profile.model,

codex-rs/core/src/config_types.rs

@@ -313,7 +313,7 @@ pub struct SandboxWorkspaceWrite {
     pub exclude_slash_tmp: bool,
 }
 
-impl From<SandboxWorkspaceWrite> for codex_protocol::mcp_protocol::SandboxSettings {
+impl From<SandboxWorkspaceWrite> for codex_app_server_protocol::SandboxSettings {
     fn from(sandbox_workspace_write: SandboxWorkspaceWrite) -> Self {
         Self {
             writable_roots: sandbox_workspace_write.writable_roots,

codex-rs/core/src/conversation_manager.rs

@@ -13,10 +13,11 @@ use crate::protocol::Event;
 use crate::protocol::EventMsg;
 use crate::protocol::SessionConfiguredEvent;
 use crate::rollout::RolloutRecorder;
-use codex_protocol::mcp_protocol::ConversationId;
+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

@@ -1,7 +1,7 @@
 use crate::exec::ExecToolCallOutput;
 use crate::token_data::KnownPlan;
 use crate::token_data::PlanType;
-use codex_protocol::mcp_protocol::ConversationId;
+use codex_protocol::ConversationId;
 use codex_protocol::protocol::RateLimitSnapshot;
 use reqwest::StatusCode;
 use serde_json;
@@ -55,6 +55,11 @@ 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),
 
@@ -76,8 +81,8 @@ pub enum CodexErr {
     Interrupted,
 
     /// Unexpected HTTP status code.
-    #[error("unexpected status {0}: {1}")]
-    UnexpectedStatus(StatusCode, String),
+    #[error("{0}")]
+    UnexpectedStatus(UnexpectedResponseError),
 
     #[error("{0}")]
     UsageLimitReached(UsageLimitReachedError),
@@ -91,8 +96,8 @@ pub enum CodexErr {
     InternalServerError,
 
     /// Retry limit exceeded.
-    #[error("exceeded retry limit, last status: {0}")]
-    RetryLimit(StatusCode),
+    #[error("{0}")]
+    RetryLimit(RetryLimitReachedError),
 
     /// Agent loop died unexpectedly
     #[error("internal error; agent loop died unexpectedly")]
@@ -108,6 +113,9 @@ pub enum CodexErr {
     #[error("unsupported operation: {0}")]
     UnsupportedOperation(String),
 
+    #[error("Fatal error: {0}")]
+    Fatal(String),
+
     // -----------------------------------------------------------------
     // Automatic conversions for common external error types
     // -----------------------------------------------------------------
@@ -135,6 +143,49 @@ pub enum CodexErr {
     EnvVar(EnvVarError),
 }
 
+#[derive(Debug)]
+pub struct UnexpectedResponseError {
+    pub status: StatusCode,
+    pub body: String,
+    pub request_id: Option<String>,
+}
+
+impl std::fmt::Display for UnexpectedResponseError {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(
+            f,
+            "unexpected status {}: {}{}",
+            self.status,
+            self.body,
+            self.request_id
+                .as_ref()
+                .map(|id| format!(", request id: {id}"))
+                .unwrap_or_default()
+        )
+    }
+}
+
+impl std::error::Error for UnexpectedResponseError {}
+#[derive(Debug)]
+pub struct RetryLimitReachedError {
+    pub status: StatusCode,
+    pub request_id: Option<String>,
+}
+
+impl std::fmt::Display for RetryLimitReachedError {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(
+            f,
+            "exceeded retry limit, last status: {}{}",
+            self.status,
+            self.request_id
+                .as_ref()
+                .map(|id| format!(", request id: {id}"))
+                .unwrap_or_default()
+        )
+    }
+}
+
 #[derive(Debug)]
 pub struct UsageLimitReachedError {
     pub(crate) plan_type: Option<PlanType>,

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_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

@@ -0,0 +1,101 @@
+use std::collections::HashMap;
+use std::env;
+
+use async_trait::async_trait;
+
+use crate::CODEX_APPLY_PATCH_ARG1;
+use crate::apply_patch::ApplyPatchExec;
+use crate::exec::ExecParams;
+use crate::function_tool::FunctionCallError;
+
+pub(crate) enum ExecutionMode {
+    Shell,
+    ApplyPatch(ApplyPatchExec),
+}
+
+#[async_trait]
+/// Backend-specific hooks that prepare and post-process execution requests for a
+/// given [`ExecutionMode`].
+pub(crate) trait ExecutionBackend: Send + Sync {
+    fn prepare(
+        &self,
+        params: ExecParams,
+        // Required for downcasting the apply_patch.
+        mode: &ExecutionMode,
+    ) -> Result<ExecParams, FunctionCallError>;
+
+    fn stream_stdout(&self, _mode: &ExecutionMode) -> bool {
+        true
+    }
+}
+
+static SHELL_BACKEND: ShellBackend = ShellBackend;
+static APPLY_PATCH_BACKEND: ApplyPatchBackend = ApplyPatchBackend;
+
+pub(crate) fn backend_for_mode(mode: &ExecutionMode) -> &'static dyn ExecutionBackend {
+    match mode {
+        ExecutionMode::Shell => &SHELL_BACKEND,
+        ExecutionMode::ApplyPatch(_) => &APPLY_PATCH_BACKEND,
+    }
+}
+
+struct ShellBackend;
+
+#[async_trait]
+impl ExecutionBackend for ShellBackend {
+    fn prepare(
+        &self,
+        params: ExecParams,
+        mode: &ExecutionMode,
+    ) -> Result<ExecParams, FunctionCallError> {
+        match mode {
+            ExecutionMode::Shell => Ok(params),
+            _ => Err(FunctionCallError::RespondToModel(
+                "shell backend invoked with non-shell mode".to_string(),
+            )),
+        }
+    }
+}
+
+struct ApplyPatchBackend;
+
+#[async_trait]
+impl ExecutionBackend for ApplyPatchBackend {
+    fn prepare(
+        &self,
+        params: ExecParams,
+        mode: &ExecutionMode,
+    ) -> 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 patch = exec.action.patch.clone();
+                Ok(ExecParams {
+                    command: vec![path_to_codex, CODEX_APPLY_PATCH_ARG1.to_string(), patch],
+                    cwd: exec.action.cwd.clone(),
+                    timeout_ms: params.timeout_ms,
+                    // Run apply_patch with a minimal environment for determinism and to
+                    // avoid leaking host environment variables into the patch process.
+                    env: HashMap::new(),
+                    with_escalated_permissions: params.with_escalated_permissions,
+                    justification: params.justification,
+                })
+            }
+            ExecutionMode::Shell => Err(FunctionCallError::RespondToModel(
+                "apply_patch backend invoked without patch context".to_string(),
+            )),
+        }
+    }
+
+    fn stream_stdout(&self, _mode: &ExecutionMode) -> bool {
+        false
+    }
+}

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

@@ -0,0 +1,51 @@
+use std::collections::HashSet;
+use std::sync::Arc;
+use std::sync::Mutex;
+
+#[derive(Clone, Debug, Default)]
+/// Thread-safe store of user approvals so repeated commands can reuse
+/// previously granted trust.
+pub(crate) struct ApprovalCache {
+    inner: Arc<Mutex<HashSet<Vec<String>>>>,
+}
+
+impl ApprovalCache {
+    pub(crate) fn insert(&self, command: Vec<String>) {
+        if command.is_empty() {
+            return;
+        }
+        if let Ok(mut guard) = self.inner.lock() {
+            guard.insert(command);
+        }
+    }
+
+    pub(crate) fn snapshot(&self) -> HashSet<Vec<String>> {
+        self.inner.lock().map(|g| g.clone()).unwrap_or_default()
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use pretty_assertions::assert_eq;
+
+    #[test]
+    fn insert_ignores_empty_and_dedupes() {
+        let cache = ApprovalCache::default();
+
+        // Empty should be ignored
+        cache.insert(vec![]);
+        assert!(cache.snapshot().is_empty());
+
+        // Insert a command and verify snapshot contains it
+        let cmd = vec!["foo".to_string(), "bar".to_string()];
+        cache.insert(cmd.clone());
+        let snap1 = cache.snapshot();
+        assert!(snap1.contains(&cmd));
+
+        // Reinserting should not create duplicates
+        cache.insert(cmd);
+        let snap2 = cache.snapshot();
+        assert_eq!(snap1, snap2);
+    }
+}

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

@@ -0,0 +1,64 @@
+mod backends;
+mod cache;
+mod runner;
+mod sandbox;
+
+pub(crate) use backends::ExecutionMode;
+pub(crate) use runner::ExecutionRequest;
+pub(crate) use runner::Executor;
+pub(crate) use runner::ExecutorConfig;
+pub(crate) use runner::normalize_exec_result;
+
+pub(crate) mod linkers {
+    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,
+        pub(crate) request: ExecutionRequest,
+    }
+
+    impl PreparedExec {
+        pub fn new(
+            context: ExecCommandContext,
+            params: ExecParams,
+            approval_command: Vec<String>,
+            mode: ExecutionMode,
+            stdout_stream: Option<StdoutStream>,
+            use_shell_profile: bool,
+        ) -> Self {
+            let request = ExecutionRequest {
+                params,
+                approval_command,
+                mode,
+                stdout_stream,
+                use_shell_profile,
+            };
+
+            Self { context, request }
+        }
+    }
+}
+
+pub mod errors {
+    use crate::error::CodexErr;
+    use crate::function_tool::FunctionCallError;
+    use thiserror::Error;
+
+    #[derive(Debug, Error)]
+    pub enum ExecError {
+        #[error(transparent)]
+        Function(#[from] FunctionCallError),
+        #[error(transparent)]
+        Codex(#[from] CodexErr),
+    }
+
+    impl ExecError {
+        pub(crate) fn rejection(msg: impl Into<String>) -> Self {
+            FunctionCallError::RespondToModel(msg.into()).into()
+        }
+    }
+}

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

@@ -0,0 +1,409 @@
+use std::path::PathBuf;
+use std::sync::Arc;
+use std::sync::RwLock;
+use std::time::Duration;
+
+use super::backends::ExecutionMode;
+use super::backends::backend_for_mode;
+use super::cache::ApprovalCache;
+use crate::codex::Session;
+use crate::error::CodexErr;
+use crate::error::SandboxErr;
+use crate::error::get_error_message_ui;
+use crate::exec::ExecParams;
+use crate::exec::ExecToolCallOutput;
+use crate::exec::SandboxType;
+use crate::exec::StdoutStream;
+use crate::exec::StreamOutput;
+use crate::exec::process_exec_tool_call;
+use crate::executor::errors::ExecError;
+use crate::executor::sandbox::select_sandbox;
+use crate::function_tool::FunctionCallError;
+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>,
+}
+
+impl ExecutorConfig {
+    pub(crate) fn new(
+        sandbox_policy: SandboxPolicy,
+        sandbox_cwd: PathBuf,
+        codex_linux_sandbox_exe: Option<PathBuf>,
+    ) -> Self {
+        Self {
+            sandbox_policy,
+            sandbox_cwd,
+            codex_linux_sandbox_exe,
+        }
+    }
+}
+
+/// Coordinates sandbox selection, backend-specific preparation, and command
+/// execution for tool calls requested by the model.
+pub(crate) struct Executor {
+    approval_cache: ApprovalCache,
+    config: Arc<RwLock<ExecutorConfig>>,
+}
+
+impl Executor {
+    pub(crate) fn new(config: ExecutorConfig) -> Self {
+        Self {
+            approval_cache: ApprovalCache::default(),
+            config: Arc::new(RwLock::new(config)),
+        }
+    }
+
+    /// Updates the sandbox policy and working directory used for future
+    /// executions without recreating the executor.
+    pub(crate) fn update_environment(&self, sandbox_policy: SandboxPolicy, sandbox_cwd: PathBuf) {
+        if let Ok(mut cfg) = self.config.write() {
+            cfg.sandbox_policy = sandbox_policy;
+            cfg.sandbox_cwd = sandbox_cwd;
+        }
+    }
+
+    /// Runs a prepared execution request end-to-end: prepares parameters, decides on
+    /// sandbox placement (prompting the user when necessary), launches the command,
+    /// and lets the backend post-process the final output.
+    pub(crate) async fn run(
+        &self,
+        mut request: ExecutionRequest,
+        session: &Session,
+        approval_policy: AskForApproval,
+        context: &ExecCommandContext,
+    ) -> Result<ExecToolCallOutput, ExecError> {
+        if matches!(request.mode, ExecutionMode::Shell) {
+            request.params =
+                maybe_translate_shell_command(request.params, session, request.use_shell_profile);
+        }
+
+        // Step 1: 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()
+        } else {
+            None
+        };
+        request.params = backend
+            .prepare(request.params, &request.mode)
+            .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,
+            approval_policy,
+            self.approval_cache.snapshot(),
+            &config,
+            session,
+            &context.sub_id,
+            &context.call_id,
+            &context.otel_event_manager,
+        )
+        .await?;
+        if sandbox_decision.record_session_approval {
+            self.approval_cache.insert(request.approval_command.clone());
+        }
+
+        // Step 4: Launch the command within the chosen sandbox.
+        let first_attempt = self
+            .spawn(
+                request.params.clone(),
+                sandbox_decision.initial_sandbox,
+                &config,
+                stdout_stream.clone(),
+            )
+            .await;
+
+        // Step 5: Handle sandbox outcomes, optionally escalating to an unsandboxed retry.
+        match first_attempt {
+            Ok(output) => Ok(output),
+            Err(CodexErr::Sandbox(SandboxErr::Timeout { output })) => {
+                Err(CodexErr::Sandbox(SandboxErr::Timeout { output }).into())
+            }
+            Err(CodexErr::Sandbox(error)) => {
+                if sandbox_decision.escalate_on_failure {
+                    self.retry_without_sandbox(
+                        &request,
+                        &config,
+                        session,
+                        context,
+                        stdout_stream,
+                        error,
+                    )
+                    .await
+                } else {
+                    let message = sandbox_failure_message(error);
+                    Err(ExecError::rejection(message))
+                }
+            }
+            Err(err) => Err(err.into()),
+        }
+    }
+
+    /// Fallback path invoked when a sandboxed run is denied so the user can
+    /// approve rerunning without isolation.
+    async fn retry_without_sandbox(
+        &self,
+        request: &ExecutionRequest,
+        config: &ExecutorConfig,
+        session: &Session,
+        context: &ExecCommandContext,
+        stdout_stream: Option<StdoutStream>,
+        sandbox_error: SandboxErr,
+    ) -> Result<ExecToolCallOutput, ExecError> {
+        session
+            .notify_background_event(
+                &context.sub_id,
+                format!("Execution failed: {sandbox_error}"),
+            )
+            .await;
+        let decision = session
+            .request_command_approval(
+                context.sub_id.to_string(),
+                context.call_id.to_string(),
+                request.approval_command.clone(),
+                request.params.cwd.clone(),
+                Some("command failed; retry without sandbox?".to_string()),
+            )
+            .await;
+
+        context.otel_event_manager.tool_decision(
+            &context.tool_name,
+            &context.call_id,
+            decision,
+            ToolDecisionSource::User,
+        );
+        match decision {
+            ReviewDecision::Approved | ReviewDecision::ApprovedForSession => {
+                if matches!(decision, ReviewDecision::ApprovedForSession) {
+                    self.approval_cache.insert(request.approval_command.clone());
+                }
+                session
+                    .notify_background_event(&context.sub_id, "retrying command without sandbox")
+                    .await;
+
+                let retry_output = self
+                    .spawn(
+                        request.params.clone(),
+                        SandboxType::None,
+                        config,
+                        stdout_stream,
+                    )
+                    .await?;
+
+                Ok(retry_output)
+            }
+            ReviewDecision::Denied | ReviewDecision::Abort => {
+                Err(ExecError::rejection("exec command rejected by user"))
+            }
+        }
+    }
+
+    async fn spawn(
+        &self,
+        params: ExecParams,
+        sandbox: SandboxType,
+        config: &ExecutorConfig,
+        stdout_stream: Option<StdoutStream>,
+    ) -> Result<ExecToolCallOutput, CodexErr> {
+        process_exec_tool_call(
+            params,
+            sandbox,
+            &config.sandbox_policy,
+            &config.sandbox_cwd,
+            &config.codex_linux_sandbox_exe,
+            stdout_stream,
+        )
+        .await
+    }
+}
+
+fn maybe_translate_shell_command(
+    params: ExecParams,
+    session: &Session,
+    use_shell_profile: bool,
+) -> ExecParams {
+    let should_translate =
+        matches!(session.user_shell(), shell::Shell::PowerShell(_)) || use_shell_profile;
+
+    if should_translate
+        && let Some(command) = session
+            .user_shell()
+            .format_default_shell_invocation(params.command.clone())
+    {
+        return ExecParams { command, ..params };
+    }
+
+    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>,
+    pub mode: ExecutionMode,
+    pub stdout_stream: Option<StdoutStream>,
+    pub use_shell_profile: bool,
+}
+
+pub(crate) struct NormalizedExecOutput<'a> {
+    borrowed: Option<&'a ExecToolCallOutput>,
+    synthetic: Option<ExecToolCallOutput>,
+}
+
+impl<'a> NormalizedExecOutput<'a> {
+    pub(crate) fn event_output(&'a self) -> &'a ExecToolCallOutput {
+        match (self.borrowed, self.synthetic.as_ref()) {
+            (Some(output), _) => output,
+            (None, Some(output)) => output,
+            (None, None) => unreachable!("normalized exec output missing data"),
+        }
+    }
+}
+
+/// Converts a raw execution result into a uniform view that always exposes an
+/// [`ExecToolCallOutput`], synthesizing error output when the command fails
+/// before producing a response.
+pub(crate) fn normalize_exec_result(
+    result: &Result<ExecToolCallOutput, ExecError>,
+) -> NormalizedExecOutput<'_> {
+    match result {
+        Ok(output) => NormalizedExecOutput {
+            borrowed: Some(output),
+            synthetic: None,
+        },
+        Err(ExecError::Codex(CodexErr::Sandbox(SandboxErr::Timeout { output }))) => {
+            NormalizedExecOutput {
+                borrowed: Some(output.as_ref()),
+                synthetic: None,
+            }
+        }
+        Err(err) => {
+            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,
+                stdout: StreamOutput::new(String::new()),
+                stderr: StreamOutput::new(message.clone()),
+                aggregated_output: StreamOutput::new(message),
+                duration: Duration::default(),
+                timed_out: false,
+            };
+            NormalizedExecOutput {
+                borrowed: None,
+                synthetic: Some(synthetic),
+            }
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::error::CodexErr;
+    use crate::error::EnvVarError;
+    use crate::error::SandboxErr;
+    use crate::exec::StreamOutput;
+    use pretty_assertions::assert_eq;
+
+    fn make_output(text: &str) -> ExecToolCallOutput {
+        ExecToolCallOutput {
+            exit_code: 1,
+            stdout: StreamOutput::new(String::new()),
+            stderr: StreamOutput::new(String::new()),
+            aggregated_output: StreamOutput::new(text.to_string()),
+            duration: Duration::from_millis(123),
+            timed_out: false,
+        }
+    }
+
+    #[test]
+    fn normalize_success_borrows() {
+        let out = make_output("ok");
+        let result: Result<ExecToolCallOutput, ExecError> = Ok(out);
+        let normalized = normalize_exec_result(&result);
+        assert_eq!(normalized.event_output().aggregated_output.text, "ok");
+    }
+
+    #[test]
+    fn normalize_timeout_borrows_embedded_output() {
+        let out = make_output("timed out payload");
+        let err = CodexErr::Sandbox(SandboxErr::Timeout {
+            output: Box::new(out),
+        });
+        let result: Result<ExecToolCallOutput, ExecError> = Err(ExecError::Codex(err));
+        let normalized = normalize_exec_result(&result);
+        assert_eq!(
+            normalized.event_output().aggregated_output.text,
+            "timed out payload"
+        );
+    }
+
+    #[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 normalize_function_error_synthesizes_payload() {
+        let err = FunctionCallError::RespondToModel("boom".to_string());
+        let result: Result<ExecToolCallOutput, ExecError> = Err(ExecError::Function(err));
+        let normalized = normalize_exec_result(&result);
+        assert_eq!(normalized.event_output().aggregated_output.text, "boom");
+    }
+
+    #[test]
+    fn normalize_codex_error_synthesizes_user_message() {
+        // Use a simple EnvVar error which formats to a clear message
+        let e = CodexErr::EnvVar(EnvVarError {
+            var: "FOO".to_string(),
+            instructions: Some("set it".to_string()),
+        });
+        let result: Result<ExecToolCallOutput, ExecError> = Err(ExecError::Codex(e));
+        let normalized = normalize_exec_result(&result);
+        assert!(
+            normalized
+                .event_output()
+                .aggregated_output
+                .text
+                .contains("Missing environment variable: `FOO`"),
+            "expected synthesized user-friendly message"
+        );
+    }
+}

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

@@ -0,0 +1,405 @@
+use crate::apply_patch::ApplyPatchExec;
+use crate::codex::Session;
+use crate::exec::SandboxType;
+use crate::executor::ExecutionMode;
+use crate::executor::ExecutionRequest;
+use crate::executor::ExecutorConfig;
+use crate::executor::errors::ExecError;
+use crate::safety::SafetyCheck;
+use crate::safety::assess_command_safety;
+use crate::safety::assess_patch_safety;
+use codex_otel::otel_event_manager::OtelEventManager;
+use codex_otel::otel_event_manager::ToolDecisionSource;
+use codex_protocol::protocol::AskForApproval;
+use codex_protocol::protocol::ReviewDecision;
+use std::collections::HashSet;
+
+/// Sandbox placement options selected for an execution run, including whether
+/// to escalate after failures and whether approvals should persist.
+pub(crate) struct SandboxDecision {
+    pub(crate) initial_sandbox: SandboxType,
+    pub(crate) escalate_on_failure: bool,
+    pub(crate) record_session_approval: bool,
+}
+
+impl SandboxDecision {
+    fn auto(sandbox: SandboxType, escalate_on_failure: bool) -> Self {
+        Self {
+            initial_sandbox: sandbox,
+            escalate_on_failure,
+            record_session_approval: false,
+        }
+    }
+
+    fn user_override(record_session_approval: bool) -> Self {
+        Self {
+            initial_sandbox: SandboxType::None,
+            escalate_on_failure: false,
+            record_session_approval,
+        }
+    }
+}
+
+fn should_escalate_on_failure(approval: AskForApproval, sandbox: SandboxType) -> bool {
+    matches!(
+        (approval, sandbox),
+        (
+            AskForApproval::UnlessTrusted | AskForApproval::OnFailure,
+            SandboxType::MacosSeatbelt | SandboxType::LinuxSeccomp
+        )
+    )
+}
+
+/// Determines how a command should be sandboxed, prompting the user when
+/// policy requires explicit approval.
+#[allow(clippy::too_many_arguments)]
+pub async fn select_sandbox(
+    request: &ExecutionRequest,
+    approval_policy: AskForApproval,
+    approval_cache: HashSet<Vec<String>>,
+    config: &ExecutorConfig,
+    session: &Session,
+    sub_id: &str,
+    call_id: &str,
+    otel_event_manager: &OtelEventManager,
+) -> Result<SandboxDecision, ExecError> {
+    match &request.mode {
+        ExecutionMode::Shell => {
+            select_shell_sandbox(
+                request,
+                approval_policy,
+                approval_cache,
+                config,
+                session,
+                sub_id,
+                call_id,
+                otel_event_manager,
+            )
+            .await
+        }
+        ExecutionMode::ApplyPatch(exec) => {
+            select_apply_patch_sandbox(exec, approval_policy, config)
+        }
+    }
+}
+
+#[allow(clippy::too_many_arguments)]
+async fn select_shell_sandbox(
+    request: &ExecutionRequest,
+    approval_policy: AskForApproval,
+    approved_snapshot: HashSet<Vec<String>>,
+    config: &ExecutorConfig,
+    session: &Session,
+    sub_id: &str,
+    call_id: &str,
+    otel_event_manager: &OtelEventManager,
+) -> Result<SandboxDecision, ExecError> {
+    let command_for_safety = if request.approval_command.is_empty() {
+        request.params.command.clone()
+    } else {
+        request.approval_command.clone()
+    };
+
+    let safety = assess_command_safety(
+        &command_for_safety,
+        approval_policy,
+        &config.sandbox_policy,
+        &approved_snapshot,
+        request.params.with_escalated_permissions.unwrap_or(false),
+    );
+
+    match safety {
+        SafetyCheck::AutoApprove {
+            sandbox_type,
+            user_explicitly_approved,
+        } => {
+            let mut decision = SandboxDecision::auto(
+                sandbox_type,
+                should_escalate_on_failure(approval_policy, sandbox_type),
+            );
+            if user_explicitly_approved {
+                decision.record_session_approval = true;
+            }
+            let (decision_for_event, source) = if user_explicitly_approved {
+                (ReviewDecision::ApprovedForSession, ToolDecisionSource::User)
+            } else {
+                (ReviewDecision::Approved, ToolDecisionSource::Config)
+            };
+            otel_event_manager.tool_decision("local_shell", call_id, decision_for_event, source);
+            Ok(decision)
+        }
+        SafetyCheck::AskUser => {
+            let decision = session
+                .request_command_approval(
+                    sub_id.to_string(),
+                    call_id.to_string(),
+                    request.approval_command.clone(),
+                    request.params.cwd.clone(),
+                    request.params.justification.clone(),
+                )
+                .await;
+
+            otel_event_manager.tool_decision(
+                "local_shell",
+                call_id,
+                decision,
+                ToolDecisionSource::User,
+            );
+            match decision {
+                ReviewDecision::Approved => Ok(SandboxDecision::user_override(false)),
+                ReviewDecision::ApprovedForSession => Ok(SandboxDecision::user_override(true)),
+                ReviewDecision::Denied | ReviewDecision::Abort => {
+                    Err(ExecError::rejection("exec command rejected by user"))
+                }
+            }
+        }
+        SafetyCheck::Reject { reason } => Err(ExecError::rejection(format!(
+            "exec command rejected: {reason}"
+        ))),
+    }
+}
+
+fn select_apply_patch_sandbox(
+    exec: &ApplyPatchExec,
+    approval_policy: AskForApproval,
+    config: &ExecutorConfig,
+) -> Result<SandboxDecision, ExecError> {
+    if exec.user_explicitly_approved_this_action {
+        return Ok(SandboxDecision::user_override(false));
+    }
+
+    match assess_patch_safety(
+        &exec.action,
+        approval_policy,
+        &config.sandbox_policy,
+        &config.sandbox_cwd,
+    ) {
+        SafetyCheck::AutoApprove { sandbox_type, .. } => Ok(SandboxDecision::auto(
+            sandbox_type,
+            should_escalate_on_failure(approval_policy, sandbox_type),
+        )),
+        SafetyCheck::AskUser => Err(ExecError::rejection(
+            "patch requires approval but none was recorded",
+        )),
+        SafetyCheck::Reject { reason } => {
+            Err(ExecError::rejection(format!("patch rejected: {reason}")))
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::codex::make_session_and_context;
+    use crate::exec::ExecParams;
+    use crate::function_tool::FunctionCallError;
+    use crate::protocol::SandboxPolicy;
+    use codex_apply_patch::ApplyPatchAction;
+    use pretty_assertions::assert_eq;
+
+    #[tokio::test]
+    async fn select_apply_patch_user_override_when_explicit() {
+        let (session, ctx) = make_session_and_context();
+        let tmp = tempfile::tempdir().expect("tmp");
+        let p = tmp.path().join("a.txt");
+        let action = ApplyPatchAction::new_add_for_test(&p, "hello".to_string());
+        let exec = ApplyPatchExec {
+            action,
+            user_explicitly_approved_this_action: true,
+        };
+        let cfg = ExecutorConfig::new(SandboxPolicy::ReadOnly, std::env::temp_dir(), None);
+        let request = ExecutionRequest {
+            params: ExecParams {
+                command: vec!["apply_patch".into()],
+                cwd: std::env::temp_dir(),
+                timeout_ms: None,
+                env: std::collections::HashMap::new(),
+                with_escalated_permissions: None,
+                justification: None,
+            },
+            approval_command: vec!["apply_patch".into()],
+            mode: ExecutionMode::ApplyPatch(exec),
+            stdout_stream: None,
+            use_shell_profile: false,
+        };
+        let otel_event_manager = ctx.client.get_otel_event_manager();
+        let decision = select_sandbox(
+            &request,
+            AskForApproval::OnRequest,
+            Default::default(),
+            &cfg,
+            &session,
+            "sub",
+            "call",
+            &otel_event_manager,
+        )
+        .await
+        .expect("ok");
+        // Explicit user override runs without sandbox
+        assert_eq!(decision.initial_sandbox, SandboxType::None);
+        assert_eq!(decision.escalate_on_failure, false);
+    }
+
+    #[tokio::test]
+    async fn select_apply_patch_autoapprove_in_danger() {
+        let (session, ctx) = make_session_and_context();
+        let tmp = tempfile::tempdir().expect("tmp");
+        let p = tmp.path().join("a.txt");
+        let action = ApplyPatchAction::new_add_for_test(&p, "hello".to_string());
+        let exec = ApplyPatchExec {
+            action,
+            user_explicitly_approved_this_action: false,
+        };
+        let cfg = ExecutorConfig::new(SandboxPolicy::DangerFullAccess, std::env::temp_dir(), None);
+        let request = ExecutionRequest {
+            params: ExecParams {
+                command: vec!["apply_patch".into()],
+                cwd: std::env::temp_dir(),
+                timeout_ms: None,
+                env: std::collections::HashMap::new(),
+                with_escalated_permissions: None,
+                justification: None,
+            },
+            approval_command: vec!["apply_patch".into()],
+            mode: ExecutionMode::ApplyPatch(exec),
+            stdout_stream: None,
+            use_shell_profile: false,
+        };
+        let otel_event_manager = ctx.client.get_otel_event_manager();
+        let decision = select_sandbox(
+            &request,
+            AskForApproval::OnRequest,
+            Default::default(),
+            &cfg,
+            &session,
+            "sub",
+            "call",
+            &otel_event_manager,
+        )
+        .await
+        .expect("ok");
+        // On platforms with a sandbox, DangerFullAccess still prefers it
+        let expected = crate::safety::get_platform_sandbox().unwrap_or(SandboxType::None);
+        assert_eq!(decision.initial_sandbox, expected);
+        assert_eq!(decision.escalate_on_failure, false);
+    }
+
+    #[tokio::test]
+    async fn select_apply_patch_requires_approval_on_unless_trusted() {
+        let (session, ctx) = make_session_and_context();
+        let tempdir = tempfile::tempdir().expect("tmpdir");
+        let p = tempdir.path().join("a.txt");
+        let action = ApplyPatchAction::new_add_for_test(&p, "hello".to_string());
+        let exec = ApplyPatchExec {
+            action,
+            user_explicitly_approved_this_action: false,
+        };
+        let cfg = ExecutorConfig::new(SandboxPolicy::ReadOnly, std::env::temp_dir(), None);
+        let request = ExecutionRequest {
+            params: ExecParams {
+                command: vec!["apply_patch".into()],
+                cwd: std::env::temp_dir(),
+                timeout_ms: None,
+                env: std::collections::HashMap::new(),
+                with_escalated_permissions: None,
+                justification: None,
+            },
+            approval_command: vec!["apply_patch".into()],
+            mode: ExecutionMode::ApplyPatch(exec),
+            stdout_stream: None,
+            use_shell_profile: false,
+        };
+        let otel_event_manager = ctx.client.get_otel_event_manager();
+        let result = select_sandbox(
+            &request,
+            AskForApproval::UnlessTrusted,
+            Default::default(),
+            &cfg,
+            &session,
+            "sub",
+            "call",
+            &otel_event_manager,
+        )
+        .await;
+        match result {
+            Ok(_) => panic!("expected error"),
+            Err(ExecError::Function(FunctionCallError::RespondToModel(msg))) => {
+                assert!(msg.contains("requires approval"))
+            }
+            Err(other) => panic!("unexpected error: {other:?}"),
+        }
+    }
+
+    #[tokio::test]
+    async fn select_shell_autoapprove_in_danger_mode() {
+        let (session, ctx) = make_session_and_context();
+        let cfg = ExecutorConfig::new(SandboxPolicy::DangerFullAccess, std::env::temp_dir(), None);
+        let request = ExecutionRequest {
+            params: ExecParams {
+                command: vec!["some-unknown".into()],
+                cwd: std::env::temp_dir(),
+                timeout_ms: None,
+                env: std::collections::HashMap::new(),
+                with_escalated_permissions: None,
+                justification: None,
+            },
+            approval_command: vec!["some-unknown".into()],
+            mode: ExecutionMode::Shell,
+            stdout_stream: None,
+            use_shell_profile: false,
+        };
+        let otel_event_manager = ctx.client.get_otel_event_manager();
+        let decision = select_sandbox(
+            &request,
+            AskForApproval::OnRequest,
+            Default::default(),
+            &cfg,
+            &session,
+            "sub",
+            "call",
+            &otel_event_manager,
+        )
+        .await
+        .expect("ok");
+        assert_eq!(decision.initial_sandbox, SandboxType::None);
+        assert_eq!(decision.escalate_on_failure, false);
+    }
+
+    #[cfg(any(target_os = "macos", target_os = "linux"))]
+    #[tokio::test]
+    async fn select_shell_escalates_on_failure_with_platform_sandbox() {
+        let (session, ctx) = make_session_and_context();
+        let cfg = ExecutorConfig::new(SandboxPolicy::ReadOnly, std::env::temp_dir(), None);
+        let request = ExecutionRequest {
+            params: ExecParams {
+                // Unknown command => untrusted but not flagged dangerous
+                command: vec!["some-unknown".into()],
+                cwd: std::env::temp_dir(),
+                timeout_ms: None,
+                env: std::collections::HashMap::new(),
+                with_escalated_permissions: None,
+                justification: None,
+            },
+            approval_command: vec!["some-unknown".into()],
+            mode: ExecutionMode::Shell,
+            stdout_stream: None,
+            use_shell_profile: false,
+        };
+        let otel_event_manager = ctx.client.get_otel_event_manager();
+        let decision = select_sandbox(
+            &request,
+            AskForApproval::OnFailure,
+            Default::default(),
+            &cfg,
+            &session,
+            "sub",
+            "call",
+            &otel_event_manager,
+        )
+        .await
+        .expect("ok");
+        // On macOS/Linux we should have a platform sandbox and escalate on failure
+        assert_ne!(decision.initial_sandbox, SandboxType::None);
+        assert_eq!(decision.escalate_on_failure, true);
+    }
+}

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/git_info.rs

@@ -2,7 +2,7 @@ use std::collections::HashSet;
 use std::path::Path;
 use std::path::PathBuf;
 
-use codex_protocol::mcp_protocol::GitSha;
+use codex_app_server_protocol::GitSha;
 use codex_protocol::protocol::GitInfo;
 use futures::future::join_all;
 use serde::Deserialize;

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;
@@ -27,6 +28,7 @@ pub mod error;
 pub mod exec;
 mod exec_command;
 pub mod exec_env;
+pub mod executor;
 mod flags;
 pub mod git_info;
 pub mod landlock;
@@ -56,7 +58,6 @@ 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;
@@ -64,9 +65,10 @@ 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_connection_manager.rs

@@ -108,9 +108,6 @@ 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 +120,15 @@ impl McpClientAdapter {
     }
 
     async fn new_streamable_http_client(
+        server_name: String,
         url: String,
         bearer_token: Option<String>,
         params: mcp_types::InitializeRequestParams,
         startup_timeout: Duration,
     ) -> 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).await?,
+        );
         client.initialize(params, Some(startup_timeout)).await?;
         Ok(McpClientAdapter::Rmcp(client))
     }
@@ -202,22 +202,9 @@ 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
-                );
-                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;
             join_set.spawn(async move {
                 let McpServerConfig { transport, .. } = cfg;
                 let params = mcp_types::InitializeRequestParams {
@@ -246,17 +233,18 @@ 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 } => {
                         McpClientAdapter::new_streamable_http_client(
+                            server_name.clone(),
                             url,
                             bearer_token,
                             params,

codex-rs/core/src/message_history.rs

@@ -30,7 +30,7 @@ use tokio::io::AsyncReadExt;
 use crate::config::Config;
 use crate::config_types::HistoryPersistence;
 
-use codex_protocol::mcp_protocol::ConversationId;
+use codex_protocol::ConversationId;
 #[cfg(unix)]
 use std::os::unix::fs::OpenOptionsExt;
 #[cfg(unix)]

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 +112,40 @@ 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![
+                "read_file".to_string(),
+                "list_dir".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!["read_file".to_string(), "list_dir".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 +166,9 @@ 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/model_provider_info.rs

@@ -6,7 +6,7 @@
 //!      key. These override or extend the defaults at runtime.
 
 use crate::CodexAuth;
-use codex_protocol::mcp_protocol::AuthMode;
+use codex_app_server_protocol::AuthMode;
 use serde::Deserialize;
 use serde::Serialize;
 use std::collections::HashMap;

codex-rs/core/src/project_doc.rs

@@ -1,6 +1,7 @@
 //! Project-level documentation discovery.
 //!
-//! Project-level documentation can be stored in files named `AGENTS.md`.
+//! Project-level documentation is primarily stored in files named `AGENTS.md`.
+//! Additional fallback filenames can be configured via `project_doc_fallback_filenames`.
 //! We include the concatenation of all files found along the path from the
 //! repository root to the current working directory as follows:
 //!
@@ -13,12 +14,13 @@
 //! 3.  We do **not** walk past the Git root.
 
 use crate::config::Config;
+use dunce::canonicalize as normalize_path;
 use std::path::PathBuf;
 use tokio::io::AsyncReadExt;
 use tracing::error;
 
-/// Currently, we only match the filename `AGENTS.md` exactly.
-const CANDIDATE_FILENAMES: &[&str] = &["AGENTS.md"];
+/// Default filename scanned for project-level docs.
+pub const DEFAULT_PROJECT_DOC_FILENAME: &str = "AGENTS.md";
 
 /// When both `Config::instructions` and the project doc are present, they will
 /// be concatenated with the following separator.
@@ -108,7 +110,7 @@ pub async fn read_project_docs(config: &Config) -> std::io::Result<Option<String
 /// is zero, returns an empty list.
 pub fn discover_project_doc_paths(config: &Config) -> std::io::Result<Vec<PathBuf>> {
     let mut dir = config.cwd.clone();
-    if let Ok(canon) = dir.canonicalize() {
+    if let Ok(canon) = normalize_path(&dir) {
         dir = canon;
     }
 
@@ -152,8 +154,9 @@ pub fn discover_project_doc_paths(config: &Config) -> std::io::Result<Vec<PathBu
     };
 
     let mut found: Vec<PathBuf> = Vec::new();
+    let candidate_filenames = candidate_filenames(config);
     for d in search_dirs {
-        for name in CANDIDATE_FILENAMES {
+        for name in &candidate_filenames {
             let candidate = d.join(name);
             match std::fs::symlink_metadata(&candidate) {
                 Ok(md) => {
@@ -173,6 +176,22 @@ pub fn discover_project_doc_paths(config: &Config) -> std::io::Result<Vec<PathBu
     Ok(found)
 }
 
+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());
+    names.push(DEFAULT_PROJECT_DOC_FILENAME);
+    for candidate in &config.project_doc_fallback_filenames {
+        let candidate = candidate.as_str();
+        if candidate.is_empty() {
+            continue;
+        }
+        if !names.contains(&candidate) {
+            names.push(candidate);
+        }
+    }
+    names
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -202,6 +221,20 @@ mod tests {
         config
     }
 
+    fn make_config_with_fallback(
+        root: &TempDir,
+        limit: usize,
+        instructions: Option<&str>,
+        fallbacks: &[&str],
+    ) -> Config {
+        let mut config = make_config(root, limit, instructions);
+        config.project_doc_fallback_filenames = fallbacks
+            .iter()
+            .map(std::string::ToString::to_string)
+            .collect();
+        config
+    }
+
     /// AGENTS.md missing – should yield `None`.
     #[tokio::test]
     async fn no_doc_file_returns_none() {
@@ -347,4 +380,45 @@ mod tests {
         let res = get_user_instructions(&cfg).await.expect("doc expected");
         assert_eq!(res, "root doc\n\ncrate doc");
     }
+
+    /// When AGENTS.md is absent but a configured fallback exists, the fallback is used.
+    #[tokio::test]
+    async fn uses_configured_fallback_when_agents_missing() {
+        let tmp = tempfile::tempdir().expect("tempdir");
+        fs::write(tmp.path().join("EXAMPLE.md"), "example instructions").unwrap();
+
+        let cfg = make_config_with_fallback(&tmp, 4096, None, &["EXAMPLE.md"]);
+
+        let res = get_user_instructions(&cfg)
+            .await
+            .expect("fallback doc expected");
+
+        assert_eq!(res, "example instructions");
+    }
+
+    /// AGENTS.md remains preferred when both AGENTS.md and fallbacks are present.
+    #[tokio::test]
+    async fn agents_md_preferred_over_fallbacks() {
+        let tmp = tempfile::tempdir().expect("tempdir");
+        fs::write(tmp.path().join("AGENTS.md"), "primary").unwrap();
+        fs::write(tmp.path().join("EXAMPLE.md"), "secondary").unwrap();
+
+        let cfg = make_config_with_fallback(&tmp, 4096, None, &["EXAMPLE.md", ".example.md"]);
+
+        let res = get_user_instructions(&cfg)
+            .await
+            .expect("AGENTS.md should win");
+
+        assert_eq!(res, "primary");
+
+        let discovery = discover_project_doc_paths(&cfg).expect("discover paths");
+        assert_eq!(discovery.len(), 1);
+        assert!(
+            discovery[0]
+                .file_name()
+                .unwrap()
+                .to_string_lossy()
+                .eq(DEFAULT_PROJECT_DOC_FILENAME)
+        );
+    }
 }

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)]
@@ -40,10 +41,25 @@ pub struct ConversationItem {
     pub head: Vec<serde_json::Value>,
     /// Last up to `TAIL_RECORD_LIMIT` JSONL response records parsed as JSON.
     pub tail: Vec<serde_json::Value>,
+    /// RFC3339 timestamp string for when the session was created, if available.
+    pub created_at: Option<String>,
+    /// RFC3339 timestamp string for the most recent response in the tail, if available.
+    pub updated_at: Option<String>,
+}
+
+#[derive(Default)]
+struct HeadTailSummary {
+    head: Vec<serde_json::Value>,
+    tail: Vec<serde_json::Value>,
+    saw_session_meta: bool,
+    saw_user_event: bool,
+    source: Option<SessionSource>,
+    created_at: Option<String>,
+    updated_at: Option<String>,
 }
 
 /// Hard cap to bound worst‑case work per request.
-const MAX_SCAN_FILES: usize = 100;
+const MAX_SCAN_FILES: usize = 10000;
 const HEAD_RECORD_LIMIT: usize = 10;
 const TAIL_RECORD_LIMIT: usize = 10;
 
@@ -92,6 +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);
@@ -107,7 +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)
 }
 
@@ -126,6 +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;
@@ -179,13 +198,33 @@ async fn traverse_directories_for_paths(
                     }
                     // Read head and simultaneously detect message events within the same
                     // first N JSONL records to avoid a second file read.
-                    let (head, tail, saw_session_meta, saw_user_event) =
-                        read_head_and_tail(&path, HEAD_RECORD_LIMIT, TAIL_RECORD_LIMIT)
-                            .await
-                            .unwrap_or((Vec::new(), Vec::new(), false, false));
+                    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 saw_session_meta && saw_user_event {
-                        items.push(ConversationItem { path, head, tail });
+                    if summary.saw_session_meta && summary.saw_user_event {
+                        let HeadTailSummary {
+                            head,
+                            tail,
+                            created_at,
+                            mut updated_at,
+                            ..
+                        } = summary;
+                        updated_at = updated_at.or_else(|| created_at.clone());
+                        items.push(ConversationItem {
+                            path,
+                            head,
+                            tail,
+                            created_at,
+                            updated_at,
+                        });
                     }
                 }
             }
@@ -293,17 +332,15 @@ async fn read_head_and_tail(
     path: &Path,
     head_limit: usize,
     tail_limit: usize,
-) -> io::Result<(Vec<serde_json::Value>, Vec<serde_json::Value>, bool, bool)> {
+) -> io::Result<HeadTailSummary> {
     use tokio::io::AsyncBufReadExt;
 
     let file = tokio::fs::File::open(path).await?;
     let reader = tokio::io::BufReader::new(file);
     let mut lines = reader.lines();
-    let mut head: Vec<serde_json::Value> = Vec::new();
-    let mut saw_session_meta = false;
-    let mut saw_user_event = false;
+    let mut summary = HeadTailSummary::default();
 
-    while head.len() < head_limit {
+    while summary.head.len() < head_limit {
         let line_opt = lines.next_line().await?;
         let Some(line) = line_opt else { break };
         let trimmed = line.trim();
@@ -316,14 +353,23 @@ 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()
+                    .or_else(|| Some(rollout_line.timestamp.clone()));
                 if let Ok(val) = serde_json::to_value(session_meta_line) {
-                    head.push(val);
-                    saw_session_meta = true;
+                    summary.head.push(val);
+                    summary.saw_session_meta = true;
                 }
             }
             RolloutItem::ResponseItem(item) => {
+                summary.created_at = summary
+                    .created_at
+                    .clone()
+                    .or_else(|| Some(rollout_line.timestamp.clone()));
                 if let Ok(val) = serde_json::to_value(item) {
-                    head.push(val);
+                    summary.head.push(val);
                 }
             }
             RolloutItem::TurnContext(_) => {
@@ -334,28 +380,30 @@ async fn read_head_and_tail(
             }
             RolloutItem::EventMsg(ev) => {
                 if matches!(ev, EventMsg::UserMessage(_)) {
-                    saw_user_event = true;
+                    summary.saw_user_event = true;
                 }
             }
         }
     }
 
-    let tail = if tail_limit == 0 {
-        Vec::new()
-    } else {
-        read_tail_records(path, tail_limit).await?
-    };
-
-    Ok((head, tail, saw_session_meta, saw_user_event))
+    if tail_limit != 0 {
+        let (tail, updated_at) = read_tail_records(path, tail_limit).await?;
+        summary.tail = tail;
+        summary.updated_at = updated_at;
+    }
+    Ok(summary)
 }
 
-async fn read_tail_records(path: &Path, max_records: usize) -> io::Result<Vec<serde_json::Value>> {
+async fn read_tail_records(
+    path: &Path,
+    max_records: usize,
+) -> io::Result<(Vec<serde_json::Value>, Option<String>)> {
     use std::io::SeekFrom;
     use tokio::io::AsyncReadExt;
     use tokio::io::AsyncSeekExt;
 
     if max_records == 0 {
-        return Ok(Vec::new());
+        return Ok((Vec::new(), None));
     }
 
     const CHUNK_SIZE: usize = 8192;
@@ -363,24 +411,28 @@ async fn read_tail_records(path: &Path, max_records: usize) -> io::Result<Vec<se
     let mut file = tokio::fs::File::open(path).await?;
     let mut pos = file.seek(SeekFrom::End(0)).await?;
     if pos == 0 {
-        return Ok(Vec::new());
+        return Ok((Vec::new(), None));
     }
 
     let mut buffer: Vec<u8> = Vec::new();
+    let mut latest_timestamp: Option<String> = None;
 
     loop {
         let slice_start = match (pos > 0, buffer.iter().position(|&b| b == b'\n')) {
             (true, Some(idx)) => idx + 1,
             _ => 0,
         };
-        let tail = collect_last_response_values(&buffer[slice_start..], max_records);
+        let (tail, newest_ts) = collect_last_response_values(&buffer[slice_start..], max_records);
+        if latest_timestamp.is_none() {
+            latest_timestamp = newest_ts.clone();
+        }
         if tail.len() >= max_records || pos == 0 {
-            return Ok(tail);
+            return Ok((tail, latest_timestamp.or(newest_ts)));
         }
 
         let read_size = CHUNK_SIZE.min(pos as usize);
         if read_size == 0 {
-            return Ok(tail);
+            return Ok((tail, latest_timestamp.or(newest_ts)));
         }
         pos -= read_size as u64;
         file.seek(SeekFrom::Start(pos)).await?;
@@ -391,15 +443,19 @@ async fn read_tail_records(path: &Path, max_records: usize) -> io::Result<Vec<se
     }
 }
 
-fn collect_last_response_values(buffer: &[u8], max_records: usize) -> Vec<serde_json::Value> {
+fn collect_last_response_values(
+    buffer: &[u8],
+    max_records: usize,
+) -> (Vec<serde_json::Value>, Option<String>) {
     use std::borrow::Cow;
 
     if buffer.is_empty() || max_records == 0 {
-        return Vec::new();
+        return (Vec::new(), None);
     }
 
     let text: Cow<'_, str> = String::from_utf8_lossy(buffer);
     let mut collected_rev: Vec<serde_json::Value> = Vec::new();
+    let mut latest_timestamp: Option<String> = None;
     for line in text.lines().rev() {
         let trimmed = line.trim();
         if trimmed.is_empty() {
@@ -407,9 +463,13 @@ fn collect_last_response_values(buffer: &[u8], max_records: usize) -> Vec<serde_
         }
         let parsed: serde_json::Result<RolloutLine> = serde_json::from_str(trimmed);
         let Ok(rollout_line) = parsed else { continue };
-        if let RolloutItem::ResponseItem(item) = rollout_line.item
-            && let Ok(val) = serde_json::to_value(item)
+        let RolloutLine { timestamp, item } = rollout_line;
+        if let RolloutItem::ResponseItem(item) = item
+            && let Ok(val) = serde_json::to_value(&item)
         {
+            if latest_timestamp.is_none() {
+                latest_timestamp = Some(timestamp.clone());
+            }
             collected_rev.push(val);
             if collected_rev.len() == max_records {
                 break;
@@ -417,7 +477,7 @@ fn collect_last_response_values(buffer: &[u8], max_records: usize) -> Vec<serde_
         }
     }
     collected_rev.reverse();
-    collected_rev
+    (collected_rev, latest_timestamp)
 }
 
 /// Locate a recorded conversation rollout file by its UUID string using the existing

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

@@ -6,7 +6,7 @@ use std::io::Error as IoError;
 use std::path::Path;
 use std::path::PathBuf;
 
-use codex_protocol::mcp_protocol::ConversationId;
+use codex_protocol::ConversationId;
 use serde_json::Value;
 use time::OffsetDateTime;
 use time::format_description::FormatItem;
@@ -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

@@ -12,13 +12,14 @@ 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;
 use crate::rollout::list::get_conversation;
 use crate::rollout::list::get_conversations;
 use anyhow::Result;
-use codex_protocol::mcp_protocol::ConversationId;
+use codex_protocol::ConversationId;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::protocol::CompactedItem;
@@ -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 =
@@ -159,16 +196,22 @@ async fn test_list_conversations_latest_first() {
                 path: p1,
                 head: head_3,
                 tail: Vec::new(),
+                created_at: Some("2025-01-03T12-00-00".into()),
+                updated_at: Some("2025-01-03T12-00-00".into()),
             },
             ConversationItem {
                 path: p2,
                 head: head_2,
                 tail: Vec::new(),
+                created_at: Some("2025-01-02T12-00-00".into()),
+                updated_at: Some("2025-01-02T12-00-00".into()),
             },
             ConversationItem {
                 path: p3,
                 head: head_1,
                 tail: Vec::new(),
+                created_at: Some("2025-01-01T12-00-00".into()),
+                updated_at: Some("2025-01-01T12-00-00".into()),
             },
         ],
         next_cursor: Some(expected_cursor),
@@ -192,13 +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")
@@ -217,7 +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,
@@ -225,7 +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();
@@ -235,11 +317,15 @@ async fn test_pagination_cursor() {
                 path: p5,
                 head: head_5,
                 tail: Vec::new(),
+                created_at: Some("2025-03-05T09-00-00".into()),
+                updated_at: Some("2025-03-05T09-00-00".into()),
             },
             ConversationItem {
                 path: p4,
                 head: head_4,
                 tail: Vec::new(),
+                created_at: Some("2025-03-04T09-00-00".into()),
+                updated_at: Some("2025-03-04T09-00-00".into()),
             },
         ],
         next_cursor: Some(expected_cursor1.clone()),
@@ -248,9 +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")
@@ -269,7 +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,
@@ -277,7 +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();
@@ -287,11 +380,15 @@ async fn test_pagination_cursor() {
                 path: p3,
                 head: head_3,
                 tail: Vec::new(),
+                created_at: Some("2025-03-03T09-00-00".into()),
+                updated_at: Some("2025-03-03T09-00-00".into()),
             },
             ConversationItem {
                 path: p2,
                 head: head_2,
                 tail: Vec::new(),
+                created_at: Some("2025-03-02T09-00-00".into()),
+                updated_at: Some("2025-03-02T09-00-00".into()),
             },
         ],
         next_cursor: Some(expected_cursor2.clone()),
@@ -300,9 +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")
@@ -315,7 +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();
@@ -324,6 +427,8 @@ async fn test_pagination_cursor() {
             path: p1,
             head: head_1,
             tail: Vec::new(),
+            created_at: Some("2025-03-01T09-00-00".into()),
+            updated_at: Some("2025-03-01T09-00-00".into()),
         }],
         next_cursor: Some(expected_cursor3),
         num_scanned_files: 5, // scanned 05, 04 (anchor), 03, 02 (anchor), 01
@@ -339,9 +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();
@@ -359,7 +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 {
@@ -367,6 +475,8 @@ async fn test_get_conversation_contents() {
             path: expected_path,
             head: expected_head,
             tail: Vec::new(),
+            created_at: Some(ts.into()),
+            updated_at: Some(ts.into()),
         }],
         next_cursor: Some(expected_cursor),
         num_scanned_files: 1,
@@ -375,7 +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",
@@ -410,6 +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,
         }),
@@ -442,25 +565,30 @@ 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));
 
     let expected: Vec<serde_json::Value> = (total_messages - tail_len..total_messages)
         .map(|idx| {
-            serde_json::to_value(ResponseItem::Message {
-                id: None,
-                role: "assistant".into(),
-                content: vec![ContentItem::OutputText {
-                    text: format!("reply-{idx}"),
-                }],
+            serde_json::json!({
+                "type": "message",
+                "role": "assistant",
+                "content": [
+                    {
+                        "type": "output_text",
+                        "text": format!("reply-{idx}"),
+                    }
+                ],
             })
-            .expect("serialize response item")
         })
         .collect();
 
     assert_eq!(item.tail, expected);
+    assert_eq!(item.created_at.as_deref(), Some(ts));
+    let expected_updated = format!("{ts}-{last:02}", last = total_messages - 1);
+    assert_eq!(item.updated_at.as_deref(), Some(expected_updated.as_str()));
 
     Ok(())
 }
@@ -488,6 +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,
         }),
@@ -519,25 +648,32 @@ 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);
 
     let expected: Vec<serde_json::Value> = (0..3)
         .map(|idx| {
-            serde_json::to_value(ResponseItem::Message {
-                id: None,
-                role: "assistant".into(),
-                content: vec![ContentItem::OutputText {
-                    text: format!("short-{idx}"),
-                }],
+            serde_json::json!({
+                "type": "message",
+                "role": "assistant",
+                "content": [
+                    {
+                        "type": "output_text",
+                        "text": format!("short-{idx}"),
+                    }
+                ],
             })
-            .expect("serialize response item")
         })
         .collect();
 
     assert_eq!(tail, &expected);
+    let expected_updated = format!("{ts}-{last:02}", last = 2);
+    assert_eq!(
+        page.items[0].updated_at.as_deref(),
+        Some(expected_updated.as_str())
+    );
 
     Ok(())
 }
@@ -565,6 +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,
         }),
@@ -610,23 +747,30 @@ 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)
         .map(|idx| {
-            serde_json::to_value(ResponseItem::Message {
-                id: None,
-                role: "assistant".into(),
-                content: vec![ContentItem::OutputText {
-                    text: format!("response-{idx}"),
-                }],
+            serde_json::json!({
+                "type": "message",
+                "role": "assistant",
+                "content": [
+                    {
+                        "type": "output_text",
+                        "text": format!("response-{idx}"),
+                    }
+                ],
             })
-            .expect("serialize response item")
         })
         .collect();
 
     assert_eq!(tail, &expected);
+    let expected_updated = format!("{ts}-{last:02}", last = 3);
+    assert_eq!(
+        page.items[0].updated_at.as_deref(),
+        Some(expected_updated.as_str())
+    );
 
     Ok(())
 }
@@ -641,11 +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")
@@ -666,7 +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();
@@ -676,11 +823,15 @@ async fn test_stable_ordering_same_second_pagination() {
                 path: p3,
                 head: head(u3),
                 tail: Vec::new(),
+                created_at: Some(ts.to_string()),
+                updated_at: Some(ts.to_string()),
             },
             ConversationItem {
                 path: p2,
                 head: head(u2),
                 tail: Vec::new(),
+                created_at: Some(ts.to_string()),
+                updated_at: Some(ts.to_string()),
             },
         ],
         next_cursor: Some(expected_cursor1.clone()),
@@ -689,9 +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")
@@ -704,6 +860,8 @@ async fn test_stable_ordering_same_second_pagination() {
             path: p1,
             head: head(u1),
             tail: Vec::new(),
+            created_at: Some(ts.to_string()),
+            updated_at: Some(ts.to_string()),
         }],
         next_cursor: Some(expected_cursor2),
         num_scanned_files: 3, // scanned u3, u2 (anchor), u1
@@ -711,3 +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/safety.rs

@@ -125,9 +125,10 @@ pub fn assess_command_safety(
     // the session _because_ they know it needs to run outside a sandbox.
 
     if is_known_safe_command(command) || approved.contains(command) {
+        let user_explicitly_approved = approved.contains(command);
         return SafetyCheck::AutoApprove {
             sandbox_type: SandboxType::None,
-            user_explicitly_approved: false,
+            user_explicitly_approved,
         };
     }
 
@@ -380,7 +381,7 @@ mod tests {
             safety_check,
             SafetyCheck::AutoApprove {
                 sandbox_type: SandboxType::None,
-                user_explicitly_approved: false,
+                user_explicitly_approved: true,
             }
         );
     }