fix: stop using actions/cache@v4 for release builds

Because we are writing to a pipe, these `flush()` calls are unnecessary,
so removing these saves us one syscall per write in these two cases.

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

@@ -0,0 +1,50 @@
+name: 🧑‍💻 VS Code Extension
+description: Report an issue with the VS Code extension
+labels:
+  - extension
+  - needs triage
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Before submitting a new issue, please search for existing issues to see if your issue has already been reported.
+        If it has, please add a 👍 reaction (no need to leave a comment) to the existing issue instead of creating a new one.
+
+  - type: input
+    id: version
+    attributes:
+      label: What version of the VS Code extension are you using?
+  - type: input
+    id: ide
+    attributes:
+      label: Which IDE are you using?
+      description: Like `VS Code`, `Cursor`, `Windsurf`, etc.
+  - type: input
+    id: platform
+    attributes:
+      label: What platform is your computer?
+      description: |
+        For MacOS and Linux: copy the output of `uname -mprs`
+        For Windows: copy the output of `"$([Environment]::OSVersion | ForEach-Object VersionString) $(if ([Environment]::Is64BitOperatingSystem) { "x64" } else { "x86" })"` in the PowerShell console
+  - type: textarea
+    id: steps
+    attributes:
+      label: What steps can reproduce the bug?
+      description: Explain the bug and provide a code snippet that can reproduce it.
+    validations:
+      required: true
+  - type: textarea
+    id: expected
+    attributes:
+      label: What is the expected behavior?
+      description: If possible, please provide text instead of a screenshot.
+  - type: textarea
+    id: actual
+    attributes:
+      label: What do you see instead?
+      description: If possible, please provide text instead of a screenshot.
+  - type: textarea
+    id: notes
+    attributes:
+      label: Additional information
+      description: Is there anything else you think we should know?

.github/actions/codex/.gitignore

@@ -1 +0,0 @@
-/node_modules/

.github/actions/codex/.prettierrc.toml

@@ -1,8 +0,0 @@
-printWidth = 80
-quoteProps = "consistent"
-semi = true
-tabWidth = 2
-trailingComma = "all"
-
-# Preserve existing behavior for markdown/text wrapping.
-proseWrap = "preserve"

.github/actions/codex/README.md

@@ -1,140 +0,0 @@
-# openai/codex-action
-
-`openai/codex-action` is a GitHub Action that facilitates the use of [Codex](https://github.com/openai/codex) on GitHub issues and pull requests. Using the action, associate **labels** to run Codex with the appropriate prompt for the given context. Codex will respond by posting comments or creating PRs, whichever you specify!
-
-Here is a sample workflow that uses `openai/codex-action`:
-
-```yaml
-name: Codex
-
-on:
-  issues:
-    types: [opened, labeled]
-  pull_request:
-    branches: [main]
-    types: [labeled]
-
-jobs:
-  codex:
-    if: ... # optional, but can be effective in conserving CI resources
-    runs-on: ubuntu-latest
-    # TODO(mbolin): Need to verify if/when `write` is necessary.
-    permissions:
-      contents: write
-      issues: write
-      pull-requests: write
-    steps:
-      # By default, Codex runs network disabled using --full-auto, so perform
-      # any setup that requires network (such as installing dependencies)
-      # before openai/codex-action.
-      - name: Checkout repository
-        uses: actions/checkout@v4
-
-      - name: Run Codex
-        uses: openai/codex-action@latest
-        with:
-          openai_api_key: ${{ secrets.CODEX_OPENAI_API_KEY }}
-          github_token: ${{ secrets.GITHUB_TOKEN }}
-```
-
-See sample usage in [`codex.yml`](../../workflows/codex.yml).
-
-## Triggering the Action
-
-Using the sample workflow above, we have:
-
-```yaml
-on:
-  issues:
-    types: [opened, labeled]
-  pull_request:
-    branches: [main]
-    types: [labeled]
-```
-
-which means our workflow will be triggered when any of the following events occur:
-
-- a label is added to an issue
-- a label is added to a pull request against the `main` branch
-
-### Label-Based Triggers
-
-To define a GitHub label that should trigger Codex, create a file named `.github/codex/labels/LABEL-NAME.md` in your repository where `LABEL-NAME` is the name of the label. The content of the file is the prompt template to use when the label is added (see more on [Prompt Template Variables](#prompt-template-variables) below).
-
-For example, if the file `.github/codex/labels/codex-review.md` exists, then:
-
-- Adding the `codex-review` label will trigger the workflow containing the `openai/codex-action` GitHub Action.
-- When `openai/codex-action` starts, it will replace the `codex-review` label with `codex-review-in-progress`.
-- When `openai/codex-action` is finished, it will replace the `codex-review-in-progress` label with `codex-review-completed`.
-
-If Codex sees that either `codex-review-in-progress` or `codex-review-completed` is already present, it will not perform the action.
-
-As determined by the [default config](./src/default-label-config.ts), Codex will act on the following labels by default:
-
-- Adding the `codex-review` label to a pull request will have Codex review the PR and add it to the PR as a comment.
-- Adding the `codex-triage` label to an issue will have Codex investigate the issue and report its findings as a comment.
-- Adding the `codex-issue-fix` label to an issue will have Codex attempt to fix the issue and create a PR wit the fix, if any.
-
-## Action Inputs
-
-The `openai/codex-action` GitHub Action takes the following inputs
-
-### `openai_api_key` (required)
-
-Set your `OPENAI_API_KEY` as a [repository secret](https://docs.github.com/en/actions/security-for-github-actions/security-guides/using-secrets-in-github-actions). See **Secrets and varaibles** then **Actions** in the settings for your GitHub repo.
-
-Note that the secret name does not have to be `OPENAI_API_KEY`. For example, you might want to name it `CODEX_OPENAI_API_KEY` and then configure it on `openai/codex-action` as follows:
-
-```yaml
-openai_api_key: ${{ secrets.CODEX_OPENAI_API_KEY }}
-```
-
-### `github_token` (required)
-
-This is required so that Codex can post a comment or create a PR. Set this value on the action as follows:
-
-```yaml
-github_token: ${{ secrets.GITHUB_TOKEN }}
-```
-
-### `codex_args`
-
-A whitespace-delimited list of arguments to pass to Codex. Defaults to `--full-auto`, but if you want to override the default model to use `o3`:
-
-```yaml
-codex_args: "--full-auto --model o3"
-```
-
-For more complex configurations, use the `codex_home` input.
-
-### `codex_home`
-
-If set, the value to use for the `$CODEX_HOME` environment variable when running Codex. As explained [in the docs](https://github.com/openai/codex/tree/main/codex-rs#readme), this folder can contain the `config.toml` to configure Codex, custom instructions, and log files.
-
-This should be a relative path within your repo.
-
-## Prompt Template Variables
-
-As shown above, `"prompt"` and `"promptPath"` are used to define prompt templates that will be populated and passed to Codex in response to certain events. All template variables are of the form `{CODEX_ACTION_...}` and the supported values are defined below.
-
-### `CODEX_ACTION_ISSUE_TITLE`
-
-If the action was triggered on a GitHub issue, this is the issue title.
-
-Specifically it is read as the `.issue.title` from the `$GITHUB_EVENT_PATH`.
-
-### `CODEX_ACTION_ISSUE_BODY`
-
-If the action was triggered on a GitHub issue, this is the issue body.
-
-Specifically it is read as the `.issue.body` from the `$GITHUB_EVENT_PATH`.
-
-### `CODEX_ACTION_GITHUB_EVENT_PATH`
-
-The value of the `$GITHUB_EVENT_PATH` environment variable, which is the path to the file that contains the JSON payload for the event that triggered the workflow. Codex can use `jq` to read only the fields of interest from this file.
-
-### `CODEX_ACTION_PR_DIFF`
-
-If the action was triggered on a pull request, this is the diff between the base and head commits of the PR. It is the output from `git diff`.
-
-Note that the content of the diff could be quite large, so is generally safer to point Codex at `CODEX_ACTION_GITHUB_EVENT_PATH` and let it decide how it wants to explore the change.

.github/actions/codex/action.yml

@@ -1,125 +0,0 @@
-name: "Codex [reusable action]"
-description: "A reusable action that runs a Codex model."
-
-inputs:
-  openai_api_key:
-    description: "The value to use as the OPENAI_API_KEY environment variable when running Codex."
-    required: true
-  trigger_phrase:
-    description: "Text to trigger Codex from a PR/issue body or comment."
-    required: false
-    default: ""
-  github_token:
-    description: "Token so Codex can comment on the PR or issue."
-    required: true
-  codex_args:
-    description: "A whitespace-delimited list of arguments to pass to Codex. Due to limitations in YAML, arguments with spaces are not supported. For more complex configurations, use the `codex_home` input."
-    required: false
-    default: "--config hide_agent_reasoning=true --full-auto"
-  codex_home:
-    description: "Value to use as the CODEX_HOME environment variable when running Codex."
-    required: false
-  codex_release_tag:
-    description: "The release tag of the Codex model to run, e.g., 'rust-v0.3.0'. Defaults to the latest release."
-    required: false
-    default: ""
-
-runs:
-  using: "composite"
-  steps:
-    # Do this in Bash so we do not even bother to install Bun if the sender does
-    # not have write access to the repo.
-    - name: Verify user has write access to the repo.
-      env:
-        GH_TOKEN: ${{ github.token }}
-      shell: bash
-      run: |
-        set -euo pipefail
-
-        PERMISSION=$(gh api \
-          "/repos/${GITHUB_REPOSITORY}/collaborators/${{ github.event.sender.login }}/permission" \
-          | jq -r '.permission')
-
-        if [[ "$PERMISSION" != "admin" && "$PERMISSION" != "write" ]]; then
-          exit 1
-        fi
-
-    - name: Download Codex
-      env:
-        GH_TOKEN: ${{ github.token }}
-      shell: bash
-      run: |
-        set -euo pipefail
-
-        # Determine OS/arch and corresponding Codex artifact name.
-        uname_s=$(uname -s)
-        uname_m=$(uname -m)
-
-        case "$uname_s" in
-          Linux*)   os="linux" ;;
-          Darwin*)  os="apple-darwin" ;;
-          *) echo "Unsupported operating system: $uname_s"; exit 1 ;;
-        esac
-
-        case "$uname_m" in
-          x86_64*) arch="x86_64" ;;
-          arm64*|aarch64*) arch="aarch64" ;;
-          *) echo "Unsupported architecture: $uname_m"; exit 1 ;;
-        esac
-
-        # linux builds differentiate between musl and gnu.
-        if [[ "$os" == "linux" ]]; then
-          if [[ "$arch" == "x86_64" ]]; then
-            triple="${arch}-unknown-linux-musl"
-          else
-            # Only other supported linux build is aarch64 gnu.
-            triple="${arch}-unknown-linux-gnu"
-          fi
-        else
-          # macOS
-          triple="${arch}-apple-darwin"
-        fi
-
-        # Note that if we start baking version numbers into the artifact name,
-        # we will need to update this action.yml file to match.
-        artifact="codex-${triple}.tar.gz"
-
-        TAG_ARG="${{ inputs.codex_release_tag }}"
-        # The usage is `gh release download [<tag>] [flags]`, so if TAG_ARG
-        # is empty, we do not pass it so we can default to the latest release.
-        gh release download ${TAG_ARG:+$TAG_ARG} --repo openai/codex \
-          --pattern "$artifact" --output - \
-        | tar xzO > /usr/local/bin/codex
-        chmod +x /usr/local/bin/codex
-
-        # Display Codex version to confirm binary integrity.
-        codex --version
-
-    - name: Install Bun
-      uses: oven-sh/setup-bun@v2
-      with:
-        bun-version: 1.2.11
-
-    - name: Install dependencies
-      shell: bash
-      run: |
-        cd ${{ github.action_path }}
-        bun install --production
-
-    - name: Run Codex
-      shell: bash
-      run: bun run ${{ github.action_path }}/src/main.ts
-      # Process args plus environment variables often have a max of 128 KiB,
-      # so we should fit within that limit?
-      env:
-        INPUT_CODEX_ARGS: ${{ inputs.codex_args || '' }}
-        INPUT_CODEX_HOME: ${{ inputs.codex_home || ''}}
-        INPUT_TRIGGER_PHRASE: ${{ inputs.trigger_phrase || '' }}
-        OPENAI_API_KEY: ${{ inputs.openai_api_key }}
-        GITHUB_TOKEN: ${{ inputs.github_token }}
-        GITHUB_EVENT_ACTION: ${{ github.event.action || '' }}
-        GITHUB_EVENT_LABEL_NAME: ${{ github.event.label.name || '' }}
-        GITHUB_EVENT_ISSUE_NUMBER: ${{ github.event.issue.number || '' }}
-        GITHUB_EVENT_ISSUE_BODY: ${{ github.event.issue.body || '' }}
-        GITHUB_EVENT_REVIEW_BODY: ${{ github.event.review.body || '' }}
-        GITHUB_EVENT_COMMENT_BODY: ${{ github.event.comment.body || '' }}

.github/actions/codex/bun.lock

@@ -1,91 +0,0 @@
-{
-  "lockfileVersion": 1,
-  "workspaces": {
-    "": {
-      "name": "codex-action",
-      "dependencies": {
-        "@actions/core": "^1.11.1",
-        "@actions/github": "^6.0.1",
-      },
-      "devDependencies": {
-        "@types/bun": "^1.2.20",
-        "@types/node": "^24.3.0",
-        "prettier": "^3.6.2",
-        "typescript": "^5.9.2",
-      },
-    },
-  },
-  "packages": {
-    "@actions/core": ["@actions/core@1.11.1", "", { "dependencies": { "@actions/exec": "^1.1.1", "@actions/http-client": "^2.0.1" } }, "sha512-hXJCSrkwfA46Vd9Z3q4cpEpHB1rL5NG04+/rbqW9d3+CSvtB1tYe8UTpAlixa1vj0m/ULglfEK2UKxMGxCxv5A=="],
-
-    "@actions/exec": ["@actions/exec@1.1.1", "", { "dependencies": { "@actions/io": "^1.0.1" } }, "sha512-+sCcHHbVdk93a0XT19ECtO/gIXoxvdsgQLzb2fE2/5sIZmWQuluYyjPQtrtTHdU1YzTZ7bAPN4sITq2xi1679w=="],
-
-    "@actions/github": ["@actions/github@6.0.1", "", { "dependencies": { "@actions/http-client": "^2.2.0", "@octokit/core": "^5.0.1", "@octokit/plugin-paginate-rest": "^9.2.2", "@octokit/plugin-rest-endpoint-methods": "^10.4.0", "@octokit/request": "^8.4.1", "@octokit/request-error": "^5.1.1", "undici": "^5.28.5" } }, "sha512-xbZVcaqD4XnQAe35qSQqskb3SqIAfRyLBrHMd/8TuL7hJSz2QtbDwnNM8zWx4zO5l2fnGtseNE3MbEvD7BxVMw=="],
-
-    "@actions/http-client": ["@actions/http-client@2.2.3", "", { "dependencies": { "tunnel": "^0.0.6", "undici": "^5.25.4" } }, "sha512-mx8hyJi/hjFvbPokCg4uRd4ZX78t+YyRPtnKWwIl+RzNaVuFpQHfmlGVfsKEJN8LwTCvL+DfVgAM04XaHkm6bA=="],
-
-    "@actions/io": ["@actions/io@1.1.3", "", {}, "sha512-wi9JjgKLYS7U/z8PPbco+PvTb/nRWjeoFlJ1Qer83k/3C5PHQi28hiVdeE2kHXmIL99mQFawx8qt/JPjZilJ8Q=="],
-
-    "@fastify/busboy": ["@fastify/busboy@2.1.1", "", {}, "sha512-vBZP4NlzfOlerQTnba4aqZoMhE/a9HY7HRqoOPaETQcSQuWEIyZMHGfVu6w9wGtGK5fED5qRs2DteVCjOH60sA=="],
-
-    "@octokit/auth-token": ["@octokit/auth-token@4.0.0", "", {}, "sha512-tY/msAuJo6ARbK6SPIxZrPBms3xPbfwBrulZe0Wtr/DIY9lje2HeV1uoebShn6mx7SjCHif6EjMvoREj+gZ+SA=="],
-
-    "@octokit/core": ["@octokit/core@5.2.1", "", { "dependencies": { "@octokit/auth-token": "^4.0.0", "@octokit/graphql": "^7.1.0", "@octokit/request": "^8.4.1", "@octokit/request-error": "^5.1.1", "@octokit/types": "^13.0.0", "before-after-hook": "^2.2.0", "universal-user-agent": "^6.0.0" } }, "sha512-dKYCMuPO1bmrpuogcjQ8z7ICCH3FP6WmxpwC03yjzGfZhj9fTJg6+bS1+UAplekbN2C+M61UNllGOOoAfGCrdQ=="],
-
-    "@octokit/endpoint": ["@octokit/endpoint@9.0.6", "", { "dependencies": { "@octokit/types": "^13.1.0", "universal-user-agent": "^6.0.0" } }, "sha512-H1fNTMA57HbkFESSt3Y9+FBICv+0jFceJFPWDePYlR/iMGrwM5ph+Dd4XRQs+8X+PUFURLQgX9ChPfhJ/1uNQw=="],
-
-    "@octokit/graphql": ["@octokit/graphql@7.1.1", "", { "dependencies": { "@octokit/request": "^8.4.1", "@octokit/types": "^13.0.0", "universal-user-agent": "^6.0.0" } }, "sha512-3mkDltSfcDUoa176nlGoA32RGjeWjl3K7F/BwHwRMJUW/IteSa4bnSV8p2ThNkcIcZU2umkZWxwETSSCJf2Q7g=="],
-
-    "@octokit/openapi-types": ["@octokit/openapi-types@24.2.0", "", {}, "sha512-9sIH3nSUttelJSXUrmGzl7QUBFul0/mB8HRYl3fOlgHbIWG+WnYDXU3v/2zMtAvuzZ/ed00Ei6on975FhBfzrg=="],
-
-    "@octokit/plugin-paginate-rest": ["@octokit/plugin-paginate-rest@9.2.2", "", { "dependencies": { "@octokit/types": "^12.6.0" }, "peerDependencies": { "@octokit/core": "5" } }, "sha512-u3KYkGF7GcZnSD/3UP0S7K5XUFT2FkOQdcfXZGZQPGv3lm4F2Xbf71lvjldr8c1H3nNbF+33cLEkWYbokGWqiQ=="],
-
-    "@octokit/plugin-rest-endpoint-methods": ["@octokit/plugin-rest-endpoint-methods@10.4.1", "", { "dependencies": { "@octokit/types": "^12.6.0" }, "peerDependencies": { "@octokit/core": "5" } }, "sha512-xV1b+ceKV9KytQe3zCVqjg+8GTGfDYwaT1ATU5isiUyVtlVAO3HNdzpS4sr4GBx4hxQ46s7ITtZrAsxG22+rVg=="],
-
-    "@octokit/request": ["@octokit/request@8.4.1", "", { "dependencies": { "@octokit/endpoint": "^9.0.6", "@octokit/request-error": "^5.1.1", "@octokit/types": "^13.1.0", "universal-user-agent": "^6.0.0" } }, "sha512-qnB2+SY3hkCmBxZsR/MPCybNmbJe4KAlfWErXq+rBKkQJlbjdJeS85VI9r8UqeLYLvnAenU8Q1okM/0MBsAGXw=="],
-
-    "@octokit/request-error": ["@octokit/request-error@5.1.1", "", { "dependencies": { "@octokit/types": "^13.1.0", "deprecation": "^2.0.0", "once": "^1.4.0" } }, "sha512-v9iyEQJH6ZntoENr9/yXxjuezh4My67CBSu9r6Ve/05Iu5gNgnisNWOsoJHTP6k0Rr0+HQIpnH+kyammu90q/g=="],
-
-    "@octokit/types": ["@octokit/types@13.10.0", "", { "dependencies": { "@octokit/openapi-types": "^24.2.0" } }, "sha512-ifLaO34EbbPj0Xgro4G5lP5asESjwHracYJvVaPIyXMuiuXLlhic3S47cBdTb+jfODkTE5YtGCLt3Ay3+J97sA=="],
-
-    "@types/bun": ["@types/bun@1.2.20", "", { "dependencies": { "bun-types": "1.2.20" } }, "sha512-dX3RGzQ8+KgmMw7CsW4xT5ITBSCrSbfHc36SNT31EOUg/LA9JWq0VDdEXDRSe1InVWpd2yLUM1FUF/kEOyTzYA=="],
-
-    "@types/node": ["@types/node@24.3.0", "", { "dependencies": { "undici-types": "~7.10.0" } }, "sha512-aPTXCrfwnDLj4VvXrm+UUCQjNEvJgNA8s5F1cvwQU+3KNltTOkBm1j30uNLyqqPNe7gE3KFzImYoZEfLhp4Yow=="],
-
-    "@types/react": ["@types/react@19.1.8", "", { "dependencies": { "csstype": "^3.0.2" } }, "sha512-AwAfQ2Wa5bCx9WP8nZL2uMZWod7J7/JSplxbTmBQ5ms6QpqNYm672H0Vu9ZVKVngQ+ii4R/byguVEUZQyeg44g=="],
-
-    "before-after-hook": ["before-after-hook@2.2.3", "", {}, "sha512-NzUnlZexiaH/46WDhANlyR2bXRopNg4F/zuSA3OpZnllCUgRaOF2znDioDWrmbNVsuZk6l9pMquQB38cfBZwkQ=="],
-
-    "bun-types": ["bun-types@1.2.20", "", { "dependencies": { "@types/node": "*" }, "peerDependencies": { "@types/react": "^19" } }, "sha512-pxTnQYOrKvdOwyiyd/7sMt9yFOenN004Y6O4lCcCUoKVej48FS5cvTw9geRaEcB9TsDZaJKAxPTVvi8tFsVuXA=="],
-
-    "csstype": ["csstype@3.1.3", "", {}, "sha512-M1uQkMl8rQK/szD0LNhtqxIPLpimGm8sOBwU7lLnCpSbTyY3yeU1Vc7l4KT5zT4s/yOxHH5O7tIuuLOCnLADRw=="],
-
-    "deprecation": ["deprecation@2.3.1", "", {}, "sha512-xmHIy4F3scKVwMsQ4WnVaS8bHOx0DmVwRywosKhaILI0ywMDWPtBSku2HNxRvF7jtwDRsoEwYQSfbxj8b7RlJQ=="],
-
-    "once": ["once@1.4.0", "", { "dependencies": { "wrappy": "1" } }, "sha512-lNaJgI+2Q5URQBkccEKHTQOPaXdUxnZZElQTZY0MFUAuaEqe1E+Nyvgdz/aIyNi6Z9MzO5dv1H8n58/GELp3+w=="],
-
-    "prettier": ["prettier@3.6.2", "", { "bin": { "prettier": "bin/prettier.cjs" } }, "sha512-I7AIg5boAr5R0FFtJ6rCfD+LFsWHp81dolrFD8S79U9tb8Az2nGrJncnMSnys+bpQJfRUzqs9hnA81OAA3hCuQ=="],
-
-    "tunnel": ["tunnel@0.0.6", "", {}, "sha512-1h/Lnq9yajKY2PEbBadPXj3VxsDDu844OnaAo52UVmIzIvwwtBPIuNvkjuzBlTWpfJyUbG3ez0KSBibQkj4ojg=="],
-
-    "typescript": ["typescript@5.9.2", "", { "bin": { "tsc": "bin/tsc", "tsserver": "bin/tsserver" } }, "sha512-CWBzXQrc/qOkhidw1OzBTQuYRbfyxDXJMVJ1XNwUHGROVmuaeiEm3OslpZ1RV96d7SKKjZKrSJu3+t/xlw3R9A=="],
-
-    "undici": ["undici@5.29.0", "", { "dependencies": { "@fastify/busboy": "^2.0.0" } }, "sha512-raqeBD6NQK4SkWhQzeYKd1KmIG6dllBOTt55Rmkt4HtI9mwdWtJljnrXjAFUBLTSN67HWrOIZ3EPF4kjUw80Bg=="],
-
-    "undici-types": ["undici-types@7.10.0", "", {}, "sha512-t5Fy/nfn+14LuOc2KNYg75vZqClpAiqscVvMygNnlsHBFpSXdJaYtXMcdNLpl/Qvc3P2cB3s6lOV51nqsFq4ag=="],
-
-    "universal-user-agent": ["universal-user-agent@6.0.1", "", {}, "sha512-yCzhz6FN2wU1NiiQRogkTQszlQSlpWaw8SvVegAc+bDxbzHgh1vX8uIe8OYyMH6DwH+sdTJsgMl36+mSMdRJIQ=="],
-
-    "wrappy": ["wrappy@1.0.2", "", {}, "sha512-l4Sp/DRseor9wL6EvV2+TuQn63dMkPjZ/sp9XkghTEbV9KlPS1xUsZ3u7/IQO4wxtcFB4bgpQPRcR3QCvezPcQ=="],
-
-    "@octokit/plugin-paginate-rest/@octokit/types": ["@octokit/types@12.6.0", "", { "dependencies": { "@octokit/openapi-types": "^20.0.0" } }, "sha512-1rhSOfRa6H9w4YwK0yrf5faDaDTb+yLyBUKOCV4xtCDB5VmIPqd/v9yr9o6SAzOAlRxMiRiCic6JVM1/kunVkw=="],
-
-    "@octokit/plugin-rest-endpoint-methods/@octokit/types": ["@octokit/types@12.6.0", "", { "dependencies": { "@octokit/openapi-types": "^20.0.0" } }, "sha512-1rhSOfRa6H9w4YwK0yrf5faDaDTb+yLyBUKOCV4xtCDB5VmIPqd/v9yr9o6SAzOAlRxMiRiCic6JVM1/kunVkw=="],
-
-    "bun-types/@types/node": ["@types/node@24.2.1", "", { "dependencies": { "undici-types": "~7.10.0" } }, "sha512-DRh5K+ka5eJic8CjH7td8QpYEV6Zo10gfRkjHCO3weqZHWDtAaSTFtl4+VMqOJ4N5jcuhZ9/l+yy8rVgw7BQeQ=="],
-
-    "@octokit/plugin-paginate-rest/@octokit/types/@octokit/openapi-types": ["@octokit/openapi-types@20.0.0", "", {}, "sha512-EtqRBEjp1dL/15V7WiX5LJMIxxkdiGJnabzYx5Apx4FkQIFgAfKumXeYAqqJCj1s+BMX4cPFIFC4OLCR6stlnA=="],
-
-    "@octokit/plugin-rest-endpoint-methods/@octokit/types/@octokit/openapi-types": ["@octokit/openapi-types@20.0.0", "", {}, "sha512-EtqRBEjp1dL/15V7WiX5LJMIxxkdiGJnabzYx5Apx4FkQIFgAfKumXeYAqqJCj1s+BMX4cPFIFC4OLCR6stlnA=="],
-  }
-}

.github/actions/codex/package.json

@@ -1,21 +0,0 @@
-{
-    "name": "codex-action",
-    "version": "0.0.0",
-    "private": true,
-    "scripts": {
-        "format": "prettier --check src",
-        "format:fix": "prettier --write src",
-        "test": "bun test",
-        "typecheck": "tsc"
-    },
-    "dependencies": {
-        "@actions/core": "^1.11.1",
-        "@actions/github": "^6.0.1"
-    },
-    "devDependencies": {
-        "@types/bun": "^1.2.20",
-        "@types/node": "^24.3.0",
-        "prettier": "^3.6.2",
-        "typescript": "^5.9.2"
-    }
-}

.github/actions/codex/src/add-reaction.ts

@@ -1,85 +0,0 @@
-import * as github from "@actions/github";
-import type { EnvContext } from "./env-context";
-
-/**
- * Add an "eyes" reaction to the entity (issue, issue comment, or pull request
- * review comment) that triggered the current Codex invocation.
- *
- * The purpose is to provide immediate feedback to the user – similar to the
- * *-in-progress label flow – indicating that the bot has acknowledged the
- * request and is working on it.
- *
- * We attempt to add the reaction best suited for the current GitHub event:
- *
- *   • issues              → POST /repos/{owner}/{repo}/issues/{issue_number}/reactions
- *   • issue_comment       → POST /repos/{owner}/{repo}/issues/comments/{comment_id}/reactions
- *   • pull_request_review_comment → POST /repos/{owner}/{repo}/pulls/comments/{comment_id}/reactions
- *
- * If the specific target is unavailable (e.g. unexpected payload shape) we
- * silently skip instead of failing the whole action because the reaction is
- * merely cosmetic.
- */
-export async function addEyesReaction(ctx: EnvContext): Promise<void> {
-  const octokit = ctx.getOctokit();
-  const { owner, repo } = github.context.repo;
-  const eventName = github.context.eventName;
-
-  try {
-    switch (eventName) {
-      case "issue_comment": {
-        const commentId = (github.context.payload as any)?.comment?.id;
-        if (commentId) {
-          await octokit.rest.reactions.createForIssueComment({
-            owner,
-            repo,
-            comment_id: commentId,
-            content: "eyes",
-          });
-          return;
-        }
-        break;
-      }
-      case "pull_request_review_comment": {
-        const commentId = (github.context.payload as any)?.comment?.id;
-        if (commentId) {
-          await octokit.rest.reactions.createForPullRequestReviewComment({
-            owner,
-            repo,
-            comment_id: commentId,
-            content: "eyes",
-          });
-          return;
-        }
-        break;
-      }
-      case "issues": {
-        const issueNumber = github.context.issue.number;
-        if (issueNumber) {
-          await octokit.rest.reactions.createForIssue({
-            owner,
-            repo,
-            issue_number: issueNumber,
-            content: "eyes",
-          });
-          return;
-        }
-        break;
-      }
-      default: {
-        // Fallback: try to react to the issue/PR if we have a number.
-        const issueNumber = github.context.issue.number;
-        if (issueNumber) {
-          await octokit.rest.reactions.createForIssue({
-            owner,
-            repo,
-            issue_number: issueNumber,
-            content: "eyes",
-          });
-        }
-      }
-    }
-  } catch (error) {
-    // Do not fail the action if reaction creation fails – log and continue.
-    console.warn(`Failed to add \"eyes\" reaction: ${error}`);
-  }
-}

.github/actions/codex/src/comment.ts

@@ -1,53 +0,0 @@
-import type { EnvContext } from "./env-context";
-import { runCodex } from "./run-codex";
-import { postComment } from "./post-comment";
-import { addEyesReaction } from "./add-reaction";
-
-/**
- * Handle `issue_comment` and `pull_request_review_comment` events once we know
- * the action is supported.
- */
-export async function onComment(ctx: EnvContext): Promise<void> {
-  const triggerPhrase = ctx.tryGet("INPUT_TRIGGER_PHRASE");
-  if (!triggerPhrase) {
-    console.warn("Empty trigger phrase: skipping.");
-    return;
-  }
-
-  // Attempt to get the body of the comment from the environment. Depending on
-  // the event type either `GITHUB_EVENT_COMMENT_BODY` (issue & PR comments) or
-  // `GITHUB_EVENT_REVIEW_BODY` (PR reviews) is set.
-  const commentBody =
-    ctx.tryGetNonEmpty("GITHUB_EVENT_COMMENT_BODY") ??
-    ctx.tryGetNonEmpty("GITHUB_EVENT_REVIEW_BODY") ??
-    ctx.tryGetNonEmpty("GITHUB_EVENT_ISSUE_BODY");
-
-  if (!commentBody) {
-    console.warn("Comment body not found in environment: skipping.");
-    return;
-  }
-
-  // Check if the trigger phrase is present.
-  if (!commentBody.includes(triggerPhrase)) {
-    console.log(
-      `Trigger phrase '${triggerPhrase}' not found: nothing to do for this comment.`,
-    );
-    return;
-  }
-
-  // Derive the prompt by removing the trigger phrase. Remove only the first
-  // occurrence to keep any additional occurrences that might be meaningful.
-  const prompt = commentBody.replace(triggerPhrase, "").trim();
-
-  if (prompt.length === 0) {
-    console.warn("Prompt is empty after removing trigger phrase: skipping");
-    return;
-  }
-
-  // Provide immediate feedback that we are working on the request.
-  await addEyesReaction(ctx);
-
-  // Run Codex and post the response as a new comment.
-  const lastMessage = await runCodex(prompt, ctx);
-  await postComment(lastMessage, ctx);
-}

.github/actions/codex/src/config.ts

@@ -1,11 +0,0 @@
-import { readdirSync, statSync } from "fs";
-import * as path from "path";
-
-export interface Config {
-  labels: Record<string, LabelConfig>;
-}
-
-export interface LabelConfig {
-  /** Returns the prompt template. */
-  getPromptTemplate(): string;
-}

.github/actions/codex/src/default-label-config.ts

@@ -1,44 +0,0 @@
-import type { Config } from "./config";
-
-export function getDefaultConfig(): Config {
-  return {
-    labels: {
-      "codex-investigate-issue": {
-        getPromptTemplate: () =>
-          `
-Troubleshoot whether the reported issue is valid.
-
-Provide a concise and respectful comment summarizing the findings.
-
-### {CODEX_ACTION_ISSUE_TITLE}
-
-{CODEX_ACTION_ISSUE_BODY}
-`.trim(),
-      },
-      "codex-code-review": {
-        getPromptTemplate: () =>
-          `
-Review this PR and respond with a very concise final message, formatted in Markdown.
-
-There should be a summary of the changes (1-2 sentences) and a few bullet points if necessary.
-
-Then provide the **review** (1-2 sentences plus bullet points, friendly tone).
-
-{CODEX_ACTION_GITHUB_EVENT_PATH} contains the JSON that triggered this GitHub workflow. It contains the \`base\` and \`head\` refs that define this PR. Both refs are available locally.
-`.trim(),
-      },
-      "codex-attempt-fix": {
-        getPromptTemplate: () =>
-          `
-Attempt to solve the reported issue.
-
-If a code change is required, create a new branch, commit the fix, and open a pull-request that resolves the problem.
-
-### {CODEX_ACTION_ISSUE_TITLE}
-
-{CODEX_ACTION_ISSUE_BODY}
-`.trim(),
-      },
-    },
-  };
-}

.github/actions/codex/src/env-context.ts

@@ -1,116 +0,0 @@
-/*
- * Centralised access to environment variables used by the Codex GitHub
- * Action.
- *
- * To enable proper unit-testing we avoid reading from `process.env` at module
- * initialisation time.  Instead a `EnvContext` object is created (usually from
- * the real `process.env`) and passed around explicitly or – where that is not
- * yet practical – imported as the shared `defaultContext` singleton. Tests can
- * create their own context backed by a stubbed map of variables without having
- * to mutate global state.
- */
-
-import { fail } from "./fail";
-import * as github from "@actions/github";
-
-export interface EnvContext {
-  /**
-   * Return the value for a given environment variable or terminate the action
-   * via `fail` if it is missing / empty.
-   */
-  get(name: string): string;
-
-  /**
-   * Attempt to read an environment variable. Returns the value when present;
-   * otherwise returns undefined (does not call `fail`).
-   */
-  tryGet(name: string): string | undefined;
-
-  /**
-   * Attempt to read an environment variable. Returns non-empty string value or
-   * null if unset or empty string.
-   */
-  tryGetNonEmpty(name: string): string | null;
-
-  /**
-   * Return a memoised Octokit instance authenticated via the token resolved
-   * from the provided argument (when defined) or the environment variables
-   * `GITHUB_TOKEN`/`GH_TOKEN`.
-   *
-   * Subsequent calls return the same cached instance to avoid spawning
-   * multiple REST clients within a single action run.
-   */
-  getOctokit(token?: string): ReturnType<typeof github.getOctokit>;
-}
-
-/** Internal helper – *not* exported. */
-function _getRequiredEnv(
-  name: string,
-  env: Record<string, string | undefined>,
-): string | undefined {
-  const value = env[name];
-
-  // Avoid leaking secrets into logs while still logging non-secret variables.
-  if (name.endsWith("KEY") || name.endsWith("TOKEN")) {
-    if (value) {
-      console.log(`value for ${name} was found`);
-    }
-  } else {
-    console.log(`${name}=${value}`);
-  }
-
-  return value;
-}
-
-/** Create a context backed by the supplied environment map (defaults to `process.env`). */
-export function createEnvContext(
-  env: Record<string, string | undefined> = process.env,
-): EnvContext {
-  // Lazily instantiated Octokit client – shared across this context.
-  let cachedOctokit: ReturnType<typeof github.getOctokit> | null = null;
-
-  return {
-    get(name: string): string {
-      const value = _getRequiredEnv(name, env);
-      if (value == null) {
-        fail(`Missing required environment variable: ${name}`);
-      }
-      return value;
-    },
-
-    tryGet(name: string): string | undefined {
-      return _getRequiredEnv(name, env);
-    },
-
-    tryGetNonEmpty(name: string): string | null {
-      const value = _getRequiredEnv(name, env);
-      return value == null || value === "" ? null : value;
-    },
-
-    getOctokit(token?: string) {
-      if (cachedOctokit) {
-        return cachedOctokit;
-      }
-
-      // Determine the token to authenticate with.
-      const githubToken = token ?? env["GITHUB_TOKEN"] ?? env["GH_TOKEN"];
-
-      if (!githubToken) {
-        fail(
-          "Unable to locate a GitHub token. `github_token` should have been set on the action.",
-        );
-      }
-
-      cachedOctokit = github.getOctokit(githubToken!);
-      return cachedOctokit;
-    },
-  };
-}
-
-/**
- * Shared context built from the actual `process.env`.  Production code that is
- * not yet refactored to receive a context explicitly may import and use this
- * singleton.  Tests should avoid the singleton and instead pass their own
- * context to the functions they exercise.
- */
-export const defaultContext: EnvContext = createEnvContext();

.github/actions/codex/src/fail.ts

@@ -1,4 +0,0 @@
-export function fail(message: string): never {
-  console.error(message);
-  process.exit(1);
-}

.github/actions/codex/src/git-helpers.ts

@@ -1,149 +0,0 @@
-import { spawnSync } from "child_process";
-import * as github from "@actions/github";
-import { EnvContext } from "./env-context";
-
-function runGit(args: string[], silent = true): string {
-  console.info(`Running git ${args.join(" ")}`);
-  const res = spawnSync("git", args, {
-    encoding: "utf8",
-    stdio: silent ? ["ignore", "pipe", "pipe"] : "inherit",
-  });
-  if (res.error) {
-    throw res.error;
-  }
-  if (res.status !== 0) {
-    // Return stderr so caller may handle; else throw.
-    throw new Error(
-      `git ${args.join(" ")} failed with code ${res.status}: ${res.stderr}`,
-    );
-  }
-  return res.stdout.trim();
-}
-
-function stageAllChanges() {
-  runGit(["add", "-A"]);
-}
-
-function hasStagedChanges(): boolean {
-  const res = spawnSync("git", ["diff", "--cached", "--quiet", "--exit-code"]);
-  return res.status !== 0;
-}
-
-function ensureOnBranch(
-  issueNumber: number,
-  protectedBranches: string[],
-  suggestedSlug?: string,
-): string {
-  let branch = "";
-  try {
-    branch = runGit(["symbolic-ref", "--short", "-q", "HEAD"]);
-  } catch {
-    branch = "";
-  }
-
-  // If detached HEAD or on a protected branch, create a new branch.
-  if (!branch || protectedBranches.includes(branch)) {
-    if (suggestedSlug) {
-      const safeSlug = suggestedSlug
-        .toLowerCase()
-        .replace(/[^\w\s-]/g, "")
-        .trim()
-        .replace(/\s+/g, "-");
-      branch = `codex-fix-${issueNumber}-${safeSlug}`;
-    } else {
-      branch = `codex-fix-${issueNumber}-${Date.now()}`;
-    }
-    runGit(["switch", "-c", branch]);
-  }
-  return branch;
-}
-
-function commitIfNeeded(issueNumber: number) {
-  if (hasStagedChanges()) {
-    runGit([
-      "commit",
-      "-m",
-      `fix: automated fix for #${issueNumber} via Codex`,
-    ]);
-  }
-}
-
-function pushBranch(branch: string, githubToken: string, ctx: EnvContext) {
-  const repoSlug = ctx.get("GITHUB_REPOSITORY"); // owner/repo
-  const remoteUrl = `https://x-access-token:${githubToken}@github.com/${repoSlug}.git`;
-
-  runGit(["push", "--force-with-lease", "-u", remoteUrl, `HEAD:${branch}`]);
-}
-
-/**
- * If this returns a string, it is the URL of the created PR.
- */
-export async function maybePublishPRForIssue(
-  issueNumber: number,
-  lastMessage: string,
-  ctx: EnvContext,
-): Promise<string | undefined> {
-  // Only proceed if GITHUB_TOKEN available.
-  const githubToken =
-    ctx.tryGetNonEmpty("GITHUB_TOKEN") ?? ctx.tryGetNonEmpty("GH_TOKEN");
-  if (!githubToken) {
-    console.warn("No GitHub token - skipping PR creation.");
-    return undefined;
-  }
-
-  // Print `git status` for debugging.
-  runGit(["status"]);
-
-  // Stage any remaining changes so they can be committed and pushed.
-  stageAllChanges();
-
-  const octokit = ctx.getOctokit(githubToken);
-
-  const { owner, repo } = github.context.repo;
-
-  // Determine default branch to treat as protected.
-  let defaultBranch = "main";
-  try {
-    const repoInfo = await octokit.rest.repos.get({ owner, repo });
-    defaultBranch = repoInfo.data.default_branch ?? "main";
-  } catch (e) {
-    console.warn(`Failed to get default branch, assuming 'main': ${e}`);
-  }
-
-  const sanitizedMessage = lastMessage.replace(/\u2022/g, "-");
-  const [summaryLine] = sanitizedMessage.split(/\r?\n/);
-  const branch = ensureOnBranch(issueNumber, [defaultBranch, "master"], summaryLine);
-  commitIfNeeded(issueNumber);
-  pushBranch(branch, githubToken, ctx);
-
-  // Try to find existing PR for this branch
-  const headParam = `${owner}:${branch}`;
-  const existing = await octokit.rest.pulls.list({
-    owner,
-    repo,
-    head: headParam,
-    state: "open",
-  });
-  if (existing.data.length > 0) {
-    return existing.data[0].html_url;
-  }
-
-  // Determine base branch (default to main)
-  let baseBranch = "main";
-  try {
-    const repoInfo = await octokit.rest.repos.get({ owner, repo });
-    baseBranch = repoInfo.data.default_branch ?? "main";
-  } catch (e) {
-    console.warn(`Failed to get default branch, assuming 'main': ${e}`);
-  }
-
-  const pr = await octokit.rest.pulls.create({
-    owner,
-    repo,
-    title: summaryLine,
-    head: branch,
-    base: baseBranch,
-    body: sanitizedMessage,
-  });
-  return pr.data.html_url;
-}

.github/actions/codex/src/git-user.ts

@@ -1,16 +0,0 @@
-export function setGitHubActionsUser(): void {
-  const commands = [
-    ["git", "config", "--global", "user.name", "github-actions[bot]"],
-    [
-      "git",
-      "config",
-      "--global",
-      "user.email",
-      "41898282+github-actions[bot]@users.noreply.github.com",
-    ],
-  ];
-
-  for (const command of commands) {
-    Bun.spawnSync(command);
-  }
-}

.github/actions/codex/src/github-workspace.ts

@@ -1,11 +0,0 @@
-import * as pathMod from "path";
-import { EnvContext } from "./env-context";
-
-export function resolveWorkspacePath(path: string, ctx: EnvContext): string {
-  if (pathMod.isAbsolute(path)) {
-    return path;
-  } else {
-    const workspace = ctx.get("GITHUB_WORKSPACE");
-    return pathMod.join(workspace, path);
-  }
-}

.github/actions/codex/src/load-config.ts

@@ -1,56 +0,0 @@
-import type { Config, LabelConfig } from "./config";
-
-import { getDefaultConfig } from "./default-label-config";
-import { readFileSync, readdirSync, statSync } from "fs";
-import * as path from "path";
-
-/**
- * Build an in-memory configuration object by scanning the repository for
- * Markdown templates located in `.github/codex/labels`.
- *
- * Each `*.md` file in that directory represents a label that can trigger the
- * Codex GitHub Action. The filename **without** the extension is interpreted
- * as the label name, e.g. `codex-review.md` ➜ `codex-review`.
- *
- * For every such label we derive the corresponding `doneLabel` by appending
- * the suffix `-completed`.
- */
-export function loadConfig(workspace: string): Config {
-  const labelsDir = path.join(workspace, ".github", "codex", "labels");
-
-  let entries: string[];
-  try {
-    entries = readdirSync(labelsDir);
-  } catch {
-    // If the directory is missing, return the default configuration.
-    return getDefaultConfig();
-  }
-
-  const labels: Record<string, LabelConfig> = {};
-
-  for (const entry of entries) {
-    if (!entry.endsWith(".md")) {
-      continue;
-    }
-
-    const fullPath = path.join(labelsDir, entry);
-
-    if (!statSync(fullPath).isFile()) {
-      continue;
-    }
-
-    const labelName = entry.slice(0, -3); // trim ".md"
-
-    labels[labelName] = new FileLabelConfig(fullPath);
-  }
-
-  return { labels };
-}
-
-class FileLabelConfig implements LabelConfig {
-  constructor(private readonly promptPath: string) {}
-
-  getPromptTemplate(): string {
-    return readFileSync(this.promptPath, "utf8");
-  }
-}

.github/actions/codex/src/main.ts

@@ -1,80 +0,0 @@
-#!/usr/bin/env bun
-
-import type { Config } from "./config";
-
-import { defaultContext, EnvContext } from "./env-context";
-import { loadConfig } from "./load-config";
-import { setGitHubActionsUser } from "./git-user";
-import { onLabeled } from "./process-label";
-import { ensureBaseAndHeadCommitsForPRAreAvailable } from "./prompt-template";
-import { performAdditionalValidation } from "./verify-inputs";
-import { onComment } from "./comment";
-import { onReview } from "./review";
-
-async function main(): Promise<void> {
-  const ctx: EnvContext = defaultContext;
-
-  // Build the configuration dynamically by scanning `.github/codex/labels`.
-  const GITHUB_WORKSPACE = ctx.get("GITHUB_WORKSPACE");
-  const config: Config = loadConfig(GITHUB_WORKSPACE);
-
-  // Optionally perform additional validation of prompt template files.
-  performAdditionalValidation(config, GITHUB_WORKSPACE);
-
-  const GITHUB_EVENT_NAME = ctx.get("GITHUB_EVENT_NAME");
-  const GITHUB_EVENT_ACTION = ctx.get("GITHUB_EVENT_ACTION");
-
-  // Set user.name and user.email to a bot before Codex runs, just in case it
-  // creates a commit.
-  setGitHubActionsUser();
-
-  switch (GITHUB_EVENT_NAME) {
-    case "issues": {
-      if (GITHUB_EVENT_ACTION === "labeled") {
-        await onLabeled(config, ctx);
-        return;
-      } else if (GITHUB_EVENT_ACTION === "opened") {
-        await onComment(ctx);
-        return;
-      }
-      break;
-    }
-    case "issue_comment": {
-      if (GITHUB_EVENT_ACTION === "created") {
-        await onComment(ctx);
-        return;
-      }
-      break;
-    }
-    case "pull_request": {
-      if (GITHUB_EVENT_ACTION === "labeled") {
-        await ensureBaseAndHeadCommitsForPRAreAvailable(ctx);
-        await onLabeled(config, ctx);
-        return;
-      }
-      break;
-    }
-    case "pull_request_review": {
-      await ensureBaseAndHeadCommitsForPRAreAvailable(ctx);
-      if (GITHUB_EVENT_ACTION === "submitted") {
-        await onReview(ctx);
-        return;
-      }
-      break;
-    }
-    case "pull_request_review_comment": {
-      await ensureBaseAndHeadCommitsForPRAreAvailable(ctx);
-      if (GITHUB_EVENT_ACTION === "created") {
-        await onComment(ctx);
-        return;
-      }
-      break;
-    }
-  }
-
-  console.warn(
-    `Unsupported action '${GITHUB_EVENT_ACTION}' for event '${GITHUB_EVENT_NAME}'.`,
-  );
-}
-
-main();

.github/actions/codex/src/post-comment.ts

@@ -1,62 +0,0 @@
-import { fail } from "./fail";
-import * as github from "@actions/github";
-import { EnvContext } from "./env-context";
-
-/**
- * Post a comment to the issue / pull request currently in scope.
- *
- * Provide the environment context so that token lookup (inside getOctokit) does
- * not rely on global state.
- */
-export async function postComment(
-  commentBody: string,
-  ctx: EnvContext,
-): Promise<void> {
-  // Append a footer with a link back to the workflow run, if available.
-  const footer = buildWorkflowRunFooter(ctx);
-  const bodyWithFooter = footer ? `${commentBody}${footer}` : commentBody;
-
-  const octokit = ctx.getOctokit();
-  console.info("Got Octokit instance for posting comment");
-  const { owner, repo } = github.context.repo;
-  const issueNumber = github.context.issue.number;
-
-  if (!issueNumber) {
-    console.warn(
-      "No issue or pull_request number found in GitHub context; skipping comment creation.",
-    );
-    return;
-  }
-
-  try {
-    console.info("Calling octokit.rest.issues.createComment()");
-    await octokit.rest.issues.createComment({
-      owner,
-      repo,
-      issue_number: issueNumber,
-      body: bodyWithFooter,
-    });
-  } catch (error) {
-    fail(`Failed to create comment via GitHub API: ${error}`);
-  }
-}
-
-/**
- * Helper to build a Markdown fragment linking back to the workflow run that
- * generated the current comment. Returns `undefined` if required environment
- * variables are missing – e.g. when running outside of GitHub Actions – so we
- * can gracefully skip the footer in those cases.
- */
-function buildWorkflowRunFooter(ctx: EnvContext): string | undefined {
-  const serverUrl =
-    ctx.tryGetNonEmpty("GITHUB_SERVER_URL") ?? "https://github.com";
-  const repository = ctx.tryGetNonEmpty("GITHUB_REPOSITORY");
-  const runId = ctx.tryGetNonEmpty("GITHUB_RUN_ID");
-
-  if (!repository || !runId) {
-    return undefined;
-  }
-
-  const url = `${serverUrl}/${repository}/actions/runs/${runId}`;
-  return `\n\n---\n*[_View workflow run_](${url})*`;
-}

.github/actions/codex/src/process-label.ts

@@ -1,226 +0,0 @@
-import { fail } from "./fail";
-import { EnvContext } from "./env-context";
-import { renderPromptTemplate } from "./prompt-template";
-
-import { postComment } from "./post-comment";
-import { runCodex } from "./run-codex";
-
-import * as github from "@actions/github";
-import { Config, LabelConfig } from "./config";
-import { maybePublishPRForIssue } from "./git-helpers";
-
-export async function onLabeled(
-  config: Config,
-  ctx: EnvContext,
-): Promise<void> {
-  const GITHUB_EVENT_LABEL_NAME = ctx.get("GITHUB_EVENT_LABEL_NAME");
-  const labelConfig = config.labels[GITHUB_EVENT_LABEL_NAME] as
-    | LabelConfig
-    | undefined;
-  if (!labelConfig) {
-    fail(
-      `Label \`${GITHUB_EVENT_LABEL_NAME}\` not found in config: ${JSON.stringify(config)}`,
-    );
-  }
-
-  await processLabelConfig(ctx, GITHUB_EVENT_LABEL_NAME, labelConfig);
-}
-
-/**
- * Wrapper that handles `-in-progress` and `-completed` semantics around the core lint/fix/review
- * processing. It will:
- *
- * - Skip execution if the `-in-progress` or `-completed` label is already present.
- * - Mark the PR/issue as `-in-progress`.
- * - After successful execution, mark the PR/issue as `-completed`.
- */
-async function processLabelConfig(
-  ctx: EnvContext,
-  label: string,
-  labelConfig: LabelConfig,
-): Promise<void> {
-  const octokit = ctx.getOctokit();
-  const { owner, repo, issueNumber, labelNames } =
-    await getCurrentLabels(octokit);
-
-  const inProgressLabel = `${label}-in-progress`;
-  const completedLabel = `${label}-completed`;
-  for (const markerLabel of [inProgressLabel, completedLabel]) {
-    if (labelNames.includes(markerLabel)) {
-      console.log(
-        `Label '${markerLabel}' already present on issue/PR #${issueNumber}. Skipping Codex action.`,
-      );
-
-      // Clean up: remove the triggering label to avoid confusion and re-runs.
-      await addAndRemoveLabels(octokit, {
-        owner,
-        repo,
-        issueNumber,
-        remove: markerLabel,
-      });
-
-      return;
-    }
-  }
-
-  // Mark the PR/issue as in progress.
-  await addAndRemoveLabels(octokit, {
-    owner,
-    repo,
-    issueNumber,
-    add: inProgressLabel,
-    remove: label,
-  });
-
-  // Run the core Codex processing.
-  await processLabel(ctx, label, labelConfig);
-
-  // Mark the PR/issue as completed.
-  await addAndRemoveLabels(octokit, {
-    owner,
-    repo,
-    issueNumber,
-    add: completedLabel,
-    remove: inProgressLabel,
-  });
-}
-
-async function processLabel(
-  ctx: EnvContext,
-  label: string,
-  labelConfig: LabelConfig,
-): Promise<void> {
-  const template = labelConfig.getPromptTemplate();
-
-  // If this is a review label, prepend explicit PR-diff scoping guidance to
-  // reduce out-of-scope feedback. Do this before rendering so placeholders in
-  // the guidance (e.g., {CODEX_ACTION_GITHUB_EVENT_PATH}) are substituted.
-  const isReview = label.toLowerCase().includes("review");
-  const reviewScopeGuidance = `
-PR Diff Scope
-- Only review changes between the PR's merge-base and head; do not comment on commits or files outside this range.
-- Derive the base/head SHAs from the event JSON at {CODEX_ACTION_GITHUB_EVENT_PATH}, then compute and use the PR diff for all analysis and comments.
-
-Commands to determine scope
-- Resolve SHAs:
-  - BASE_SHA=$(jq -r '.pull_request.base.sha // .pull_request.base.ref' "{CODEX_ACTION_GITHUB_EVENT_PATH}")
-  - HEAD_SHA=$(jq -r '.pull_request.head.sha // .pull_request.head.ref' "{CODEX_ACTION_GITHUB_EVENT_PATH}")
-  - BASE_SHA=$(git rev-parse "$BASE_SHA")
-  - HEAD_SHA=$(git rev-parse "$HEAD_SHA")
-- Prefer triple-dot (merge-base) semantics for PR diffs:
-  - Changed commits: git log --oneline "$BASE_SHA...$HEAD_SHA"
-  - Changed files: git diff --name-status "$BASE_SHA...$HEAD_SHA"
-  - Review hunks: git diff -U0 "$BASE_SHA...$HEAD_SHA"
-
-Review rules
-- Anchor every comment to a file and hunk present in git diff "$BASE_SHA...$HEAD_SHA".
-- If you mention context outside the diff, label it as "Follow-up (outside this PR scope)" and keep it brief (<=2 bullets).
-- Do not critique commits or files not reachable in the PR range (merge-base(base, head) → head).
-`.trim();
-
-  const effectiveTemplate = isReview
-    ? `${reviewScopeGuidance}\n\n${template}`
-    : template;
-
-  const populatedTemplate = await renderPromptTemplate(effectiveTemplate, ctx);
-
-  // Always run Codex and post the resulting message as a comment.
-  let commentBody = await runCodex(populatedTemplate, ctx);
-
-  // Current heuristic: only try to create a PR if "attempt" or "fix" is in the
-  // label name. (Yes, we plan to evolve this.)
-  if (label.indexOf("fix") !== -1 || label.indexOf("attempt") !== -1) {
-    console.info(`label ${label} indicates we should attempt to create a PR`);
-    const prUrl = await maybeFixIssue(ctx, commentBody);
-    if (prUrl) {
-      commentBody += `\n\n---\nOpened pull request: ${prUrl}`;
-    }
-  } else {
-    console.info(
-      `label ${label} does not indicate we should attempt to create a PR`,
-    );
-  }
-
-  await postComment(commentBody, ctx);
-}
-
-async function maybeFixIssue(
-  ctx: EnvContext,
-  lastMessage: string,
-): Promise<string | undefined> {
-  // Attempt to create a PR out of any changes Codex produced.
-  const issueNumber = github.context.issue.number!; // exists for issues triggering this path
-  try {
-    return await maybePublishPRForIssue(issueNumber, lastMessage, ctx);
-  } catch (e) {
-    console.warn(`Failed to publish PR: ${e}`);
-  }
-}
-
-async function getCurrentLabels(
-  octokit: ReturnType<typeof github.getOctokit>,
-): Promise<{
-  owner: string;
-  repo: string;
-  issueNumber: number;
-  labelNames: Array<string>;
-}> {
-  const { owner, repo } = github.context.repo;
-  const issueNumber = github.context.issue.number;
-
-  if (!issueNumber) {
-    fail("No issue or pull_request number found in GitHub context.");
-  }
-
-  const { data: issueData } = await octokit.rest.issues.get({
-    owner,
-    repo,
-    issue_number: issueNumber,
-  });
-
-  const labelNames =
-    issueData.labels?.map((label: any) =>
-      typeof label === "string" ? label : label.name,
-    ) ?? [];
-
-  return { owner, repo, issueNumber, labelNames };
-}
-
-async function addAndRemoveLabels(
-  octokit: ReturnType<typeof github.getOctokit>,
-  opts: {
-    owner: string;
-    repo: string;
-    issueNumber: number;
-    add?: string;
-    remove?: string;
-  },
-): Promise<void> {
-  const { owner, repo, issueNumber, add, remove } = opts;
-
-  if (add) {
-    try {
-      await octokit.rest.issues.addLabels({
-        owner,
-        repo,
-        issue_number: issueNumber,
-        labels: [add],
-      });
-    } catch (error) {
-      console.warn(`Failed to add label '${add}': ${error}`);
-    }
-  }
-
-  if (remove) {
-    try {
-      await octokit.rest.issues.removeLabel({
-        owner,
-        repo,
-        issue_number: issueNumber,
-        name: remove,
-      });
-    } catch (error) {
-      console.warn(`Failed to remove label '${remove}': ${error}`);
-    }
-  }
-}

.github/actions/codex/src/prompt-template.ts

@@ -1,284 +0,0 @@
-/*
- * Utilities to render Codex prompt templates.
- *
- * A template is a Markdown (or plain-text) file that may contain one or more
- * placeholders of the form `{CODEX_ACTION_<NAME>}`. At runtime these
- * placeholders are substituted with dynamically generated content. Each
- * placeholder is resolved **exactly once** even if it appears multiple times
- * in the same template.
- */
-
-import { readFile } from "fs/promises";
-
-import { EnvContext } from "./env-context";
-
-// ---------------------------------------------------------------------------
-// Helpers
-// ---------------------------------------------------------------------------
-
-/**
- * Lazily caches parsed `$GITHUB_EVENT_PATH` contents keyed by the file path so
- * we only hit the filesystem once per unique event payload.
- */
-const githubEventDataCache: Map<string, Promise<any>> = new Map();
-
-function getGitHubEventData(ctx: EnvContext): Promise<any> {
-  const eventPath = ctx.get("GITHUB_EVENT_PATH");
-  let cached = githubEventDataCache.get(eventPath);
-  if (!cached) {
-    cached = readFile(eventPath, "utf8").then((raw) => JSON.parse(raw));
-    githubEventDataCache.set(eventPath, cached);
-  }
-  return cached;
-}
-
-async function runCommand(args: Array<string>): Promise<string> {
-  const result = Bun.spawnSync(args, {
-    stdout: "pipe",
-    stderr: "pipe",
-  });
-
-  if (result.success) {
-    return result.stdout.toString();
-  }
-
-  console.error(`Error running ${JSON.stringify(args)}: ${result.stderr}`);
-  return "";
-}
-
-// ---------------------------------------------------------------------------
-// Public API
-// ---------------------------------------------------------------------------
-
-// Regex that captures the variable name without the surrounding { } braces.
-const VAR_REGEX = /\{(CODEX_ACTION_[A-Z0-9_]+)\}/g;
-
-// Cache individual placeholder values so each one is resolved at most once per
-// process even if many templates reference it.
-const placeholderCache: Map<string, Promise<string>> = new Map();
-
-/**
- * Parse a template string, resolve all placeholders and return the rendered
- * result.
- */
-export async function renderPromptTemplate(
-  template: string,
-  ctx: EnvContext,
-): Promise<string> {
-  // ---------------------------------------------------------------------
-  // 1) Gather all *unique* placeholders present in the template.
-  // ---------------------------------------------------------------------
-  const variables = new Set<string>();
-  for (const match of template.matchAll(VAR_REGEX)) {
-    variables.add(match[1]);
-  }
-
-  // ---------------------------------------------------------------------
-  // 2) Kick off (or reuse) async resolution for each variable.
-  // ---------------------------------------------------------------------
-  for (const variable of variables) {
-    if (!placeholderCache.has(variable)) {
-      placeholderCache.set(variable, resolveVariable(variable, ctx));
-    }
-  }
-
-  // ---------------------------------------------------------------------
-  // 3) Await completion so we can perform a simple synchronous replace below.
-  // ---------------------------------------------------------------------
-  const resolvedEntries: [string, string][] = [];
-  for (const [key, promise] of placeholderCache.entries()) {
-    resolvedEntries.push([key, await promise]);
-  }
-  const resolvedMap = new Map<string, string>(resolvedEntries);
-
-  // ---------------------------------------------------------------------
-  // 4) Replace each occurrence.  We use replace with a callback to ensure
-  //    correct substitution even if variable names overlap (they shouldn't,
-  //    but better safe than sorry).
-  // ---------------------------------------------------------------------
-  return template.replace(VAR_REGEX, (_, varName: string) => {
-    return resolvedMap.get(varName) ?? "";
-  });
-}
-
-export async function ensureBaseAndHeadCommitsForPRAreAvailable(
-  ctx: EnvContext,
-): Promise<{ baseSha: string; headSha: string } | null> {
-  const prShas = await getPrShas(ctx);
-  if (prShas == null) {
-    console.warn("Unable to resolve PR branches");
-    return null;
-  }
-
-  const event = await getGitHubEventData(ctx);
-  const pr = event.pull_request;
-  if (!pr) {
-    console.warn("event.pull_request is not defined - unexpected");
-    return null;
-  }
-
-  const workspace = ctx.get("GITHUB_WORKSPACE");
-
-  // Refs (branch names)
-  const baseRef: string | undefined = pr.base?.ref;
-  const headRef: string | undefined = pr.head?.ref;
-
-  // Clone URLs
-  const baseRemoteUrl: string | undefined = pr.base?.repo?.clone_url;
-  const headRemoteUrl: string | undefined = pr.head?.repo?.clone_url;
-
-  if (!baseRef || !headRef || !baseRemoteUrl || !headRemoteUrl) {
-    console.warn(
-      "Missing PR ref or remote URL information - cannot fetch commits",
-    );
-    return null;
-  }
-
-  // Ensure we have the base branch.
-  await runCommand([
-    "git",
-    "-C",
-    workspace,
-    "fetch",
-    "--no-tags",
-    "origin",
-    baseRef,
-  ]);
-
-  // Ensure we have the head branch.
-  if (headRemoteUrl === baseRemoteUrl) {
-    // Same repository – the commit is available from `origin`.
-    await runCommand([
-      "git",
-      "-C",
-      workspace,
-      "fetch",
-      "--no-tags",
-      "origin",
-      headRef,
-    ]);
-  } else {
-    // Fork – make sure a `pr` remote exists that points at the fork. Attempting
-    // to add a remote that already exists causes git to error, so we swallow
-    // any non-zero exit codes from that specific command.
-    await runCommand([
-      "git",
-      "-C",
-      workspace,
-      "remote",
-      "add",
-      "pr",
-      headRemoteUrl,
-    ]);
-
-    // Whether adding succeeded or the remote already existed, attempt to fetch
-    // the head ref from the `pr` remote.
-    await runCommand([
-      "git",
-      "-C",
-      workspace,
-      "fetch",
-      "--no-tags",
-      "pr",
-      headRef,
-    ]);
-  }
-
-  return prShas;
-}
-
-// ---------------------------------------------------------------------------
-// Internal helpers – still exported for use by other modules.
-// ---------------------------------------------------------------------------
-
-export async function resolvePrDiff(ctx: EnvContext): Promise<string> {
-  const prShas = await ensureBaseAndHeadCommitsForPRAreAvailable(ctx);
-  if (prShas == null) {
-    console.warn("Unable to resolve PR branches");
-    return "";
-  }
-
-  const workspace = ctx.get("GITHUB_WORKSPACE");
-  const { baseSha, headSha } = prShas;
-  return runCommand([
-    "git",
-    "-C",
-    workspace,
-    "diff",
-    "--color=never",
-    `${baseSha}..${headSha}`,
-  ]);
-}
-
-// ---------------------------------------------------------------------------
-// Placeholder resolution
-// ---------------------------------------------------------------------------
-
-async function resolveVariable(name: string, ctx: EnvContext): Promise<string> {
-  switch (name) {
-    case "CODEX_ACTION_ISSUE_TITLE": {
-      const event = await getGitHubEventData(ctx);
-      const issue = event.issue ?? event.pull_request;
-      return issue?.title ?? "";
-    }
-
-    case "CODEX_ACTION_ISSUE_BODY": {
-      const event = await getGitHubEventData(ctx);
-      const issue = event.issue ?? event.pull_request;
-      return issue?.body ?? "";
-    }
-
-    case "CODEX_ACTION_GITHUB_EVENT_PATH": {
-      return ctx.get("GITHUB_EVENT_PATH");
-    }
-
-    case "CODEX_ACTION_BASE_REF": {
-      const event = await getGitHubEventData(ctx);
-      return event?.pull_request?.base?.ref ?? "";
-    }
-
-    case "CODEX_ACTION_HEAD_REF": {
-      const event = await getGitHubEventData(ctx);
-      return event?.pull_request?.head?.ref ?? "";
-    }
-
-    case "CODEX_ACTION_PR_DIFF": {
-      return resolvePrDiff(ctx);
-    }
-
-    // -------------------------------------------------------------------
-    // Add new template variables here.
-    // -------------------------------------------------------------------
-
-    default: {
-      // Unknown variable – leave it blank to avoid leaking placeholders to the
-      // final prompt.  The alternative would be to `fail()` here, but silently
-      // ignoring unknown placeholders is more forgiving and better matches the
-      // behaviour of typical template engines.
-      console.warn(`Unknown template variable: ${name}`);
-      return "";
-    }
-  }
-}
-
-async function getPrShas(
-  ctx: EnvContext,
-): Promise<{ baseSha: string; headSha: string } | null> {
-  const event = await getGitHubEventData(ctx);
-  const pr = event.pull_request;
-  if (!pr) {
-    console.warn("event.pull_request is not defined");
-    return null;
-  }
-
-  // Prefer explicit SHAs if available to avoid relying on local branch names.
-  const baseSha: string | undefined = pr.base?.sha;
-  const headSha: string | undefined = pr.head?.sha;
-
-  if (!baseSha || !headSha) {
-    console.warn("one of base or head is not defined on event.pull_request");
-    return null;
-  }
-
-  return { baseSha, headSha };
-}

.github/actions/codex/src/review.ts

@@ -1,42 +0,0 @@
-import type { EnvContext } from "./env-context";
-import { runCodex } from "./run-codex";
-import { postComment } from "./post-comment";
-import { addEyesReaction } from "./add-reaction";
-
-/**
- * Handle `pull_request_review` events. We treat the review body the same way
- * as a normal comment.
- */
-export async function onReview(ctx: EnvContext): Promise<void> {
-  const triggerPhrase = ctx.tryGet("INPUT_TRIGGER_PHRASE");
-  if (!triggerPhrase) {
-    console.warn("Empty trigger phrase: skipping.");
-    return;
-  }
-
-  const reviewBody = ctx.tryGet("GITHUB_EVENT_REVIEW_BODY");
-
-  if (!reviewBody) {
-    console.warn("Review body not found in environment: skipping.");
-    return;
-  }
-
-  if (!reviewBody.includes(triggerPhrase)) {
-    console.log(
-      `Trigger phrase '${triggerPhrase}' not found: nothing to do for this review.`,
-    );
-    return;
-  }
-
-  const prompt = reviewBody.replace(triggerPhrase, "").trim();
-
-  if (prompt.length === 0) {
-    console.warn("Prompt is empty after removing trigger phrase: skipping.");
-    return;
-  }
-
-  await addEyesReaction(ctx);
-
-  const lastMessage = await runCodex(prompt, ctx);
-  await postComment(lastMessage, ctx);
-}

.github/actions/codex/src/run-codex.ts

@@ -1,58 +0,0 @@
-import { fail } from "./fail";
-import { EnvContext } from "./env-context";
-import { tmpdir } from "os";
-import { join } from "node:path";
-import { readFile, mkdtemp } from "fs/promises";
-import { resolveWorkspacePath } from "./github-workspace";
-
-/**
- * Runs the Codex CLI with the provided prompt and returns the output written
- * to the "last message" file.
- */
-export async function runCodex(
-  prompt: string,
-  ctx: EnvContext,
-): Promise<string> {
-  const OPENAI_API_KEY = ctx.get("OPENAI_API_KEY");
-
-  const tempDirPath = await mkdtemp(join(tmpdir(), "codex-"));
-  const lastMessageOutput = join(tempDirPath, "codex-prompt.md");
-
-  // Use the unified CLI and its `exec` subcommand instead of the old
-  // standalone `codex-exec` binary.
-  const args = ["/usr/local/bin/codex", "exec"];
-
-  const inputCodexArgs = ctx.tryGet("INPUT_CODEX_ARGS")?.trim();
-  if (inputCodexArgs) {
-    args.push(...inputCodexArgs.split(/\s+/));
-  }
-
-  args.push("--output-last-message", lastMessageOutput, prompt);
-
-  const env: Record<string, string> = { ...process.env, OPENAI_API_KEY };
-  const INPUT_CODEX_HOME = ctx.tryGet("INPUT_CODEX_HOME");
-  if (INPUT_CODEX_HOME) {
-    env.CODEX_HOME = resolveWorkspacePath(INPUT_CODEX_HOME, ctx);
-  }
-
-  console.log(`Running Codex: ${JSON.stringify(args)}`);
-  const result = Bun.spawnSync(args, {
-    stdout: "inherit",
-    stderr: "inherit",
-    env,
-  });
-
-  if (!result.success) {
-    fail(`Codex failed: see above for details.`);
-  }
-
-  // Read the output generated by Codex.
-  let lastMessage: string;
-  try {
-    lastMessage = await readFile(lastMessageOutput, "utf8");
-  } catch (err) {
-    fail(`Failed to read Codex output at '${lastMessageOutput}': ${err}`);
-  }
-
-  return lastMessage;
-}

.github/actions/codex/src/verify-inputs.ts

@@ -1,33 +0,0 @@
-// Validate the inputs passed to the composite action.
-// The script currently ensures that the provided configuration file exists and
-// matches the expected schema.
-
-import type { Config } from "./config";
-
-import { existsSync } from "fs";
-import * as path from "path";
-import { fail } from "./fail";
-
-export function performAdditionalValidation(config: Config, workspace: string) {
-  // Additional validation: ensure referenced prompt files exist and are Markdown.
-  for (const [label, details] of Object.entries(config.labels)) {
-    // Determine which prompt key is present (the schema guarantees exactly one).
-    const promptPathStr =
-      (details as any).prompt ?? (details as any).promptPath;
-
-    if (promptPathStr) {
-      const promptPath = path.isAbsolute(promptPathStr)
-        ? promptPathStr
-        : path.join(workspace, promptPathStr);
-
-      if (!existsSync(promptPath)) {
-        fail(`Prompt file for label '${label}' not found: ${promptPath}`);
-      }
-      if (!promptPath.endsWith(".md")) {
-        fail(
-          `Prompt file for label '${label}' must be a .md file (got ${promptPathStr}).`,
-        );
-      }
-    }
-  }
-}

.github/actions/codex/tsconfig.json

@@ -1,15 +0,0 @@
-{
-  "compilerOptions": {
-    "lib": ["ESNext"],
-    "target": "ESNext",
-    "module": "ESNext",
-    "moduleDetection": "force",
-    "moduleResolution": "bundler",
-
-    "noEmit": true,
-    "strict": true,
-    "skipLibCheck": true
-  },
-
-  "include": ["src"]
-}

.github/workflows/codex.yml

@@ -1,64 +0,0 @@
-name: Codex
-
-on:
-  issues:
-    types: [opened, labeled]
-  pull_request:
-    branches: [main]
-    types: [labeled]
-
-jobs:
-  codex:
-    # This `if` check provides complex filtering logic to avoid running Codex
-    # on every PR. Admittedly, one thing this does not verify is whether the
-    # sender has write access to the repo: that must be done as part of a
-    # runtime step.
-    #
-    # Note the label values should match the ones in the .github/codex/labels
-    # folder.
-    if: |
-      (github.event_name == 'issues' && (
-        (github.event.action == 'labeled' && (github.event.label.name == 'codex-attempt' || github.event.label.name == 'codex-triage'))
-      )) ||
-      (github.event_name == 'pull_request' && github.event.action == 'labeled' && (github.event.label.name == 'codex-review' || github.event.label.name == 'codex-rust-review'))
-    runs-on: ubuntu-latest
-    permissions:
-      contents: write # can push or create branches
-      issues: write # for comments + labels on issues/PRs
-      pull-requests: write # for PR comments/labels
-    steps:
-      # TODO: Consider adding an optional mode (--dry-run?) to actions/codex
-      # that verifies whether Codex should actually be run for this event.
-      # (For example, it may be rejected because the sender does not have
-      # write access to the repo.) The benefit would be two-fold:
-      # 1. As the first step of this job, it gives us a chance to add a reaction
-      #    or comment to the PR/issue ASAP to "ack" the request.
-      # 2. It saves resources by skipping the clone and setup steps below if
-      #    Codex is not going to run.
-
-      - name: Checkout repository
-        uses: actions/checkout@v5
-
-      - uses: dtolnay/rust-toolchain@1.89
-        with:
-          targets: x86_64-unknown-linux-gnu
-          components: clippy
-
-      - uses: actions/cache@v4
-        with:
-          path: |
-            ~/.cargo/bin/
-            ~/.cargo/registry/index/
-            ~/.cargo/registry/cache/
-            ~/.cargo/git/db/
-            ${{ github.workspace }}/codex-rs/target/
-          key: cargo-ubuntu-24.04-x86_64-unknown-linux-gnu-dev-${{ hashFiles('**/Cargo.lock') }}
-
-      # Note it is possible that the `verify` step internal to Run Codex will
-      # fail, in which case the work to setup the repo was worthless :(
-      - name: Run Codex
-        uses: ./.github/actions/codex
-        with:
-          openai_api_key: ${{ secrets.CODEX_OPENAI_API_KEY }}
-          github_token: ${{ secrets.GITHUB_TOKEN }}
-          codex_home: ./.github/codex/home

.github/workflows/rust-ci.yml

@@ -134,7 +134,7 @@ jobs:
 
       - name: cargo clippy
         id: clippy
-        run: cargo clippy --target ${{ matrix.target }} --all-features --tests -- -D warnings
+        run: cargo clippy --target ${{ matrix.target }} --all-features --tests --profile ${{ matrix.profile }} -- -D warnings
 
       # Running `cargo build` from the workspace root builds the workspace using
       # the union of all features from third-party crates. This can mask errors

.github/workflows/rust-release.yml

@@ -79,15 +79,6 @@ jobs:
         with:
           targets: ${{ matrix.target }}
 
-      - uses: actions/cache@v4
-        with:
-          path: |
-            ~/.cargo/bin/
-            ~/.cargo/registry/index/
-            ~/.cargo/registry/cache/
-            ~/.cargo/git/db/
-            ${{ github.workspace }}/codex-rs/target/
-          key: cargo-release-${{ matrix.runner }}-${{ matrix.target }}-release-${{ hashFiles('**/Cargo.lock') }}
 
       - if: ${{ matrix.target == 'x86_64-unknown-linux-musl' || matrix.target == 'aarch64-unknown-linux-musl'}}
         name: Install musl build tools

AGENTS.md

@@ -8,9 +8,10 @@ In the codex-rs folder where the rust code lives:
   - 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.
 
-Before finalizing a change to `codex-rs`, run `just fmt` (in `codex-rs` directory) to format the code and `just fix -p <project>` (in `codex-rs` directory) to fix any linter issues in the code. Additionally, run the tests:
+Before finalizing a change to `codex-rs`, run `just fmt` (in `codex-rs` directory) to format the code and `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 these commands to finalize.
 
 ## TUI style conventions
 

CHANGELOG.md

@@ -1,211 +1 @@
-# Changelog
-
-You can install any of these versions: `npm install -g codex@version`
-
-## `0.1.2505172129`
-
-### 🪲 Bug Fixes
-
-- Add node version check (#1007)
-- Persist token after refresh (#1006)
-
-## `0.1.2505171619`
-
-- `codex --login` + `codex --free` (#998)
-
-## `0.1.2505161800`
-
-- Sign in with chatgpt credits (#974)
-- Add support for OpenAI tool type, local_shell (#961)
-
-## `0.1.2505161243`
-
-- Sign in with chatgpt (#963)
-- Session history viewer (#912)
-- Apply patch issue when using different cwd (#942)
-- Diff command for filenames with special characters (#954)
-
-## `0.1.2505160811`
-
-- `codex-mini-latest` (#951)
-
-## `0.1.2505140839`
-
-### 🪲 Bug Fixes
-
-- Gpt-4.1 apply_patch handling (#930)
-- Add support for fileOpener in config.json (#911)
-- Patch in #366 and #367 for marked-terminal (#916)
-- Remember to set lastIndex = 0 on shared RegExp (#918)
-- Always load version from package.json at runtime (#909)
-- Tweak the label for citations for better rendering (#919)
-- Tighten up some logic around session timestamps and ids (#922)
-- Change EventMsg enum so every variant takes a single struct (#925)
-- Reasoning default to medium, show workdir when supplied (#931)
-- Test_dev_null_write() was not using echo as intended (#923)
-
-## `0.1.2504301751`
-
-### 🚀 Features
-
-- User config api key (#569)
-- `@mention` files in codex (#701)
-- Add `--reasoning` CLI flag (#314)
-- Lower default retry wait time and increase number of tries (#720)
-- Add common package registries domains to allowed-domains list (#414)
-
-### 🪲 Bug Fixes
-
-- Insufficient quota message (#758)
-- Input keyboard shortcut opt+delete (#685)
-- `/diff` should include untracked files (#686)
-- Only allow running without sandbox if explicitly marked in safe container (#699)
-- Tighten up check for /usr/bin/sandbox-exec (#710)
-- Check if sandbox-exec is available (#696)
-- Duplicate messages in quiet mode (#680)
-
-## `0.1.2504251709`
-
-### 🚀 Features
-
-- Add openai model info configuration (#551)
-- Added provider to run quiet mode function (#571)
-- Create parent directories when creating new files (#552)
-- Print bug report URL in terminal instead of opening browser (#510) (#528)
-- Add support for custom provider configuration in the user config (#537)
-- Add support for OpenAI-Organization and OpenAI-Project headers (#626)
-- Add specific instructions for creating API keys in error msg (#581)
-- Enhance toCodePoints to prevent potential unicode 14 errors (#615)
-- More native keyboard navigation in multiline editor (#655)
-- Display error on selection of invalid model (#594)
-
-### 🪲 Bug Fixes
-
-- Model selection (#643)
-- Nits in apply patch (#640)
-- Input keyboard shortcuts (#676)
-- `apply_patch` unicode characters (#625)
-- Don't clear turn input before retries (#611)
-- More loosely match context for apply_patch (#610)
-- Update bug report template - there is no --revision flag (#614)
-- Remove outdated copy of text input and external editor feature (#670)
-- Remove unreachable "disableResponseStorage" logic flow introduced in #543 (#573)
-- Non-openai mode - fix for gemini content: null, fix 429 to throw before stream (#563)
-- Only allow going up in history when not already in history if input is empty (#654)
-- Do not grant "node" user sudo access when using run_in_container.sh (#627)
-- Update scripts/build_container.sh to use pnpm instead of npm (#631)
-- Update lint-staged config to use pnpm --filter (#582)
-- Non-openai mode - don't default temp and top_p (#572)
-- Fix error catching when checking for updates (#597)
-- Close stdin when running an exec tool call (#636)
-
-## `0.1.2504221401`
-
-### 🚀 Features
-
-- Show actionable errors when api keys are missing (#523)
-- Add CLI `--version` flag (#492)
-
-### 🪲 Bug Fixes
-
-- Agent loop for ZDR (`disableResponseStorage`) (#543)
-- Fix relative `workdir` check for `apply_patch` (#556)
-- Minimal mid-stream #429 retry loop using existing back-off (#506)
-- Inconsistent usage of base URL and API key (#507)
-- Remove requirement for api key for ollama (#546)
-- Support `[provider]_BASE_URL` (#542)
-
-## `0.1.2504220136`
-
-### 🚀 Features
-
-- Add support for ZDR orgs (#481)
-- Include fractional portion of chunk that exceeds stdout/stderr limit (#497)
-
-## `0.1.2504211509`
-
-### 🚀 Features
-
-- Support multiple providers via Responses-Completion transformation (#247)
-- Add user-defined safe commands configuration and approval logic #380 (#386)
-- Allow switching approval modes when prompted to approve an edit/command (#400)
-- Add support for `/diff` command autocomplete in TerminalChatInput (#431)
-- Auto-open model selector if user selects deprecated model (#427)
-- Read approvalMode from config file (#298)
-- `/diff` command to view git diff (#426)
-- Tab completions for file paths (#279)
-- Add /command autocomplete (#317)
-- Allow multi-line input (#438)
-
-### 🪲 Bug Fixes
-
-- `full-auto` support in quiet mode (#374)
-- Enable shell option for child process execution (#391)
-- Configure husky and lint-staged for pnpm monorepo (#384)
-- Command pipe execution by improving shell detection (#437)
-- Name of the file not matching the name of the component (#354)
-- Allow proper exit from new Switch approval mode dialog (#453)
-- Ensure /clear resets context and exclude system messages from approximateTokenUsed count (#443)
-- `/clear` now clears terminal screen and resets context left indicator (#425)
-- Correct fish completion function name in CLI script (#485)
-- Auto-open model-selector when model is not found (#448)
-- Remove unnecessary isLoggingEnabled() checks (#420)
-- Improve test reliability for `raw-exec` (#434)
-- Unintended tear down of agent loop (#483)
-- Remove extraneous type casts (#462)
-
-## `0.1.2504181820`
-
-### 🚀 Features
-
-- Add `/bug` report command (#312)
-- Notify when a newer version is available (#333)
-
-### 🪲 Bug Fixes
-
-- Update context left display logic in TerminalChatInput component (#307)
-- Improper spawn of sh on Windows Powershell (#318)
-- `/bug` report command, thinking indicator (#381)
-- Include pnpm lock file (#377)
-
-## `0.1.2504172351`
-
-### 🚀 Features
-
-- Add Nix flake for reproducible development environments (#225)
-
-### 🪲 Bug Fixes
-
-- Handle invalid commands (#304)
-- Raw-exec-process-group.test improve reliability and error handling (#280)
-- Canonicalize the writeable paths used in seatbelt policy (#275)
-
-## `0.1.2504172304`
-
-### 🚀 Features
-
-- Add shell completion subcommand (#138)
-- Add command history persistence (#152)
-- Shell command explanation option (#173)
-- Support bun fallback runtime for codex CLI (#282)
-- Add notifications for MacOS using Applescript (#160)
-- Enhance image path detection in input processing (#189)
-- `--config`/`-c` flag to open global instructions in nvim (#158)
-- Update position of cursor when navigating input history with arrow keys to the end of the text (#255)
-
-### 🪲 Bug Fixes
-
-- Correct word deletion logic for trailing spaces (Ctrl+Backspace) (#131)
-- Improve Windows compatibility for CLI commands and sandbox (#261)
-- Correct typos in thinking texts (transcendent & parroting) (#108)
-- Add empty vite config file to prevent resolving to parent (#273)
-- Update regex to better match the retry error messages (#266)
-- Add missing "as" in prompt prefix in agent loop (#186)
-- Allow continuing after interrupting assistant (#178)
-- Standardize filename to kebab-case 🐍➡️🥙 (#302)
-- Small update to bug report template (#288)
-- Duplicated message on model change (#276)
-- Typos in prompts and comments (#195)
-- Check workdir before spawn (#221)
-
-<!-- generated - do not edit -->
+The changelog can be found on the [releases page](https://github.com/openai/codex/releases)

README.md

@@ -5,73 +5,25 @@
 <p align="center"><strong>Codex CLI</strong> is a coding agent from OpenAI that runs locally on your computer.</br>If you are looking for the <em>cloud-based agent</em> from OpenAI, <strong>Codex Web</strong>, see <a href="https://chatgpt.com/codex">chatgpt.com/codex</a>.</p>
 
 <p align="center">
-  <img src="./.github/codex-cli-splash.png" alt="Codex CLI splash" width="50%" />
+  <img src="./.github/codex-cli-splash.png" alt="Codex CLI splash" width="80%" />
   </p>
 
 ---
 
-<details>
-<summary><strong>Table of contents</strong></summary>
-
-<!-- Begin ToC -->
-
-- [Quickstart](#quickstart)
-  - [Installing and running Codex CLI](#installing-and-running-codex-cli)
-  - [Using Codex with your ChatGPT plan](#using-codex-with-your-chatgpt-plan)
-  - [Connecting on a "Headless" Machine](#connecting-on-a-headless-machine)
-    - [Authenticate locally and copy your credentials to the "headless" machine](#authenticate-locally-and-copy-your-credentials-to-the-headless-machine)
-    - [Connecting through VPS or remote](#connecting-through-vps-or-remote)
-  - [Usage-based billing alternative: Use an OpenAI API key](#usage-based-billing-alternative-use-an-openai-api-key)
-    - [Forcing a specific auth method (advanced)](#forcing-a-specific-auth-method-advanced)
-  - [Choosing Codex's level of autonomy](#choosing-codexs-level-of-autonomy)
-    - [**1. Read/write**](#1-readwrite)
-    - [**2. Read-only**](#2-read-only)
-    - [**3. Advanced configuration**](#3-advanced-configuration)
-    - [Can I run without ANY approvals?](#can-i-run-without-any-approvals)
-    - [Fine-tuning in `config.toml`](#fine-tuning-in-configtoml)
-  - [Example prompts](#example-prompts)
-- [Running with a prompt as input](#running-with-a-prompt-as-input)
-- [Using Open Source Models](#using-open-source-models)
-  - [Platform sandboxing details](#platform-sandboxing-details)
-- [Experimental technology disclaimer](#experimental-technology-disclaimer)
-- [System requirements](#system-requirements)
-- [CLI reference](#cli-reference)
-- [Memory & project docs](#memory--project-docs)
-- [Non-interactive / CI mode](#non-interactive--ci-mode)
-- [Model Context Protocol (MCP)](#model-context-protocol-mcp)
-- [Tracing / verbose logging](#tracing--verbose-logging)
-  - [DotSlash](#dotslash)
-- [Configuration](#configuration)
-- [FAQ](#faq)
-- [Zero data retention (ZDR) usage](#zero-data-retention-zdr-usage)
-- [Codex open source fund](#codex-open-source-fund)
-- [Contributing](#contributing)
-  - [Development workflow](#development-workflow)
-  - [Writing high-impact code changes](#writing-high-impact-code-changes)
-  - [Opening a pull request](#opening-a-pull-request)
-  - [Review process](#review-process)
-  - [Community values](#community-values)
-  - [Getting help](#getting-help)
-  - [Contributor license agreement (CLA)](#contributor-license-agreement-cla)
-    - [Quick fixes](#quick-fixes)
-  - [Releasing `codex`](#releasing-codex)
-- [Security & responsible AI](#security--responsible-ai)
-- [License](#license)
-
-<!-- End ToC -->
-
-</details>
-
----
-
 ## Quickstart
 
 ### Installing and running Codex CLI
 
-Install globally with your preferred package manager:
+Install globally with your preferred package manager. If you use npm:
 
 ```shell
-npm install -g @openai/codex  # Alternatively: `brew install codex`
+npm install -g @openai/codex
+```
+
+Alternatively, if you use Homebrew:
+
+```shell
+brew install codex
 ```
 
 Then simply run `codex` to get started:
@@ -99,607 +51,52 @@ Each archive contains a single entry with the platform baked into the name (e.g.
 ### Using Codex with your ChatGPT plan
 
 <p align="center">
-  <img src="./.github/codex-cli-login.png" alt="Codex CLI login" width="50%" />
+  <img src="./.github/codex-cli-login.png" alt="Codex CLI login" width="80%" />
   </p>
 
-Run `codex` and select **Sign in with ChatGPT**. You'll need a Plus, Pro, or Team ChatGPT account, and will get access to our latest models, including `gpt-5`, at no extra cost to your plan. (Enterprise is coming soon.)
+Run `codex` and select **Sign in with ChatGPT**. We recommend signing into your ChatGPT account to use Codex as part of your Plus, Pro, Team, Edu, or Enterprise plan. [Learn more about what's included in your ChatGPT plan](https://help.openai.com/en/articles/11369540-codex-in-chatgpt).
 
-> Important: If you've used the Codex CLI before, follow these steps to migrate from usage-based billing with your API key:
->
-> 1. Update the CLI and ensure `codex --version` is `0.20.0` or later
-> 2. Delete `~/.codex/auth.json` (this should be `C:\Users\USERNAME\.codex\auth.json` on Windows)
-> 3. Run `codex login` again
+You can also use Codex with an API key, but this requires [additional setup](./docs/authentication.md#usage-based-billing-alternative-use-an-openai-api-key). If you previously used an API key for usage-based billing, see the [migration steps](./docs/authentication.md#migrating-from-usage-based-billing-api-key). If you're having trouble with login, please comment on [this issue](https://github.com/openai/codex/issues/1243).
 
-If you encounter problems with the login flow, please comment on [this issue](https://github.com/openai/codex/issues/1243).
+### Model Context Protocol (MCP)
 
-### Connecting on a "Headless" Machine
+Codex CLI supports [MCP servers](./docs/advanced.md#model-context-protocol-mcp). Enable by adding an `mcp_servers` section to your `~/.codex/config.toml`.
 
-Today, the login process entails running a server on `localhost:1455`. If you are on a "headless" server, such as a Docker container or are `ssh`'d into a remote machine, loading `localhost:1455` in the browser on your local machine will not automatically connect to the webserver running on the _headless_ machine, so you must use one of the following workarounds:
 
-#### Authenticate locally and copy your credentials to the "headless" machine
+### Configuration
 
-The easiest solution is likely to run through the `codex login` process on your local machine such that `localhost:1455` _is_ accessible in your web browser. When you complete the authentication process, an `auth.json` file should be available at `$CODEX_HOME/auth.json` (on Mac/Linux, `$CODEX_HOME` defaults to `~/.codex` whereas on Windows, it defaults to `%USERPROFILE%\.codex`).
-
-Because the `auth.json` file is not tied to a specific host, once you complete the authentication flow locally, you can copy the `$CODEX_HOME/auth.json` file to the headless machine and then `codex` should "just work" on that machine. Note to copy a file to a Docker container, you can do:
-
-```shell
-# substitute MY_CONTAINER with the name or id of your Docker container:
-CONTAINER_HOME=$(docker exec MY_CONTAINER printenv HOME)
-docker exec MY_CONTAINER mkdir -p "$CONTAINER_HOME/.codex"
-docker cp auth.json MY_CONTAINER:"$CONTAINER_HOME/.codex/auth.json"
-```
-
-whereas if you are `ssh`'d into a remote machine, you likely want to use [`scp`](https://en.wikipedia.org/wiki/Secure_copy_protocol):
-
-```shell
-ssh user@remote 'mkdir -p ~/.codex'
-scp ~/.codex/auth.json user@remote:~/.codex/auth.json
-```
-
-or try this one-liner:
-
-```shell
-ssh user@remote 'mkdir -p ~/.codex && cat > ~/.codex/auth.json' < ~/.codex/auth.json
-```
-
-#### Connecting through VPS or remote
-
-If you run Codex on a remote machine (VPS/server) without a local browser, the login helper starts a server on `localhost:1455` on the remote host. To complete login in your local browser, forward that port to your machine before starting the login flow:
-
-```bash
-# From your local machine
-ssh -L 1455:localhost:1455 <user>@<remote-host>
-```
-
-Then, in that SSH session, run `codex` and select "Sign in with ChatGPT". When prompted, open the printed URL (it will be `http://localhost:1455/...`) in your local browser. The traffic will be tunneled to the remote server.
-
-### Usage-based billing alternative: Use an OpenAI API key
-
-If you prefer to pay-as-you-go, you can still authenticate with your OpenAI API key by setting it as an environment variable:
-
-```shell
-export OPENAI_API_KEY="your-api-key-here"
-```
-
-Notes:
-
-- This command only sets the key for your current terminal session, which we recommend. To set it for all future sessions, you can also add the `export` line to your shell's configuration file (e.g., `~/.zshrc`).
-- If you have signed in with ChatGPT, Codex will default to using your ChatGPT credits. If you wish to use your API key, use the `/logout` command to clear your ChatGPT authentication.
-
-#### Forcing a specific auth method (advanced)
-
-You can explicitly choose which authentication Codex should prefer when both are available.
-
-- To always use your API key (even when ChatGPT auth exists), set:
-
-```toml
-# ~/.codex/config.toml
-preferred_auth_method = "apikey"
-```
-
-Or override ad-hoc via CLI:
-
-```bash
-codex --config preferred_auth_method="apikey"
-```
-
-- To prefer ChatGPT auth (default), set:
-
-```toml
-# ~/.codex/config.toml
-preferred_auth_method = "chatgpt"
-```
-
-Notes:
-
-- When `preferred_auth_method = "apikey"` and an API key is available, the login screen is skipped.
-- When `preferred_auth_method = "chatgpt"` (default), Codex prefers ChatGPT auth if present; if only an API key is present, it will use the API key. Certain account types may also require API-key mode.
-
-### Choosing Codex's level of autonomy
-
-We always recommend running Codex in its default sandbox that gives you strong guardrails around what the agent can do. The default sandbox prevents it from editing files outside its workspace, or from accessing the network.
-
-When you launch Codex in a new folder, it detects whether the folder is version controlled and recommends one of two levels of autonomy:
-
-#### **1. Read/write**
-
-- Codex can run commands and write files in the workspace without approval.
-- To write files in other folders, access network, update git or perform other actions protected by the sandbox, Codex will need your permission.
-- By default, the workspace includes the current directory, as well as temporary directories like `/tmp`. You can see what directories are in the workspace with the `/status` command. See the docs for how to customize this behavior.
-- Advanced: You can manually specify this configuration by running `codex --sandbox workspace-write --ask-for-approval on-request`
-- This is the recommended default for version-controlled folders.
-
-#### **2. Read-only**
-
-- Codex can run read-only commands without approval.
-- To edit files, access network, or perform other actions protected by the sandbox, Codex will need your permission.
-- Advanced: You can manually specify this configuration by running `codex --sandbox read-only --ask-for-approval on-request`
-- This is the recommended default non-version-controlled folders.
-
-#### **3. Advanced configuration**
-
-Codex gives you fine-grained control over the sandbox with the `--sandbox` option, and over when it requests approval with the `--ask-for-approval` option. Run `codex help` for more on these options.
-
-#### Can I run without ANY approvals?
-
-Yes, run codex non-interactively with `--ask-for-approval never`. This option works with all `--sandbox` options, so you still have full control over Codex's level of autonomy. It will make its best attempt with whatever contrainsts you provide. For example:
-
-- Use `codex --ask-for-approval never --sandbox read-only` when you are running many agents to answer questions in parallel in the same workspace.
-- Use `codex --ask-for-approval never --sandbox workspace-write` when you want the agent to non-interactively take time to produce the best outcome, with strong guardrails around its behavior.
-- Use `codex --ask-for-approval never --sandbox danger-full-access` to dangerously give the agent full autonomy. Because this disables important safety mechanisms, we recommend against using this unless running Codex in an isolated environment.
-
-#### Fine-tuning in `config.toml`
-
-```toml
-# approval mode
-approval_policy = "untrusted"
-sandbox_mode    = "read-only"
-
-# full-auto mode
-approval_policy = "on-request"
-sandbox_mode    = "workspace-write"
-
-# Optional: allow network in workspace-write mode
-[sandbox_workspace_write]
-network_access = true
-```
-
-You can also save presets as **profiles**:
-
-```toml
-[profiles.full_auto]
-approval_policy = "on-request"
-sandbox_mode    = "workspace-write"
-
-[profiles.readonly_quiet]
-approval_policy = "never"
-sandbox_mode    = "read-only"
-```
-
-### Example prompts
-
-Below are a few bite-size examples you can copy-paste. Replace the text in quotes with your own task. See the [prompting guide](https://github.com/openai/codex/blob/main/codex-cli/examples/prompting_guide.md) for more tips and usage patterns.
-
-| ✨  | What you type                                                                   | What happens                                                               |
-| --- | ------------------------------------------------------------------------------- | -------------------------------------------------------------------------- |
-| 1   | `codex "Refactor the Dashboard component to React Hooks"`                       | Codex rewrites the class component, runs `npm test`, and shows the diff.   |
-| 2   | `codex "Generate SQL migrations for adding a users table"`                      | Infers your ORM, creates migration files, and runs them in a sandboxed DB. |
-| 3   | `codex "Write unit tests for utils/date.ts"`                                    | Generates tests, executes them, and iterates until they pass.              |
-| 4   | `codex "Bulk-rename *.jpeg -> *.jpg with git mv"`                               | Safely renames files and updates imports/usages.                           |
-| 5   | `codex "Explain what this regex does: ^(?=.*[A-Z]).{8,}$"`                      | Outputs a step-by-step human explanation.                                  |
-| 6   | `codex "Carefully review this repo, and propose 3 high impact well-scoped PRs"` | Suggests impactful PRs in the current codebase.                            |
-| 7   | `codex "Look for vulnerabilities and create a security review report"`          | Finds and explains security bugs.                                          |
-
-## Running with a prompt as input
-
-You can also run Codex CLI with a prompt as input:
-
-```shell
-codex "explain this codebase to me"
-```
-
-```shell
-codex --full-auto "create the fanciest todo-list app"
-```
-
-That's it - Codex will scaffold a file, run it inside a sandbox, install any
-missing dependencies, and show you the live result. Approve the changes and
-they'll be committed to your working directory.
-
-## Using Open Source Models
-
-<details>
-<summary><strong>Use <code>--profile</code> to use other models</strong></summary>
-
-Codex also allows you to use other providers that support the OpenAI Chat Completions (or Responses) API.
-
-To do so, you must first define custom [providers](./config.md#model_providers) in `~/.codex/config.toml`. For example, the provider for a standard Ollama setup would be defined as follows:
-
-```toml
-[model_providers.ollama]
-name = "Ollama"
-base_url = "http://localhost:11434/v1"
-```
-
-The `base_url` will have `/chat/completions` appended to it to build the full URL for the request.
-
-For providers that also require an `Authorization` header of the form `Bearer: SECRET`, an `env_key` can be specified, which indicates the environment variable to read to use as the value of `SECRET` when making a request:
-
-```toml
-[model_providers.openrouter]
-name = "OpenRouter"
-base_url = "https://openrouter.ai/api/v1"
-env_key = "OPENROUTER_API_KEY"
-```
-
-Providers that speak the Responses API are also supported by adding `wire_api = "responses"` as part of the definition. Accessing OpenAI models via Azure is an example of such a provider, though it also requires specifying additional `query_params` that need to be appended to the request URL:
-
-```toml
-[model_providers.azure]
-name = "Azure"
-# Make sure you set the appropriate subdomain for this URL.
-base_url = "https://YOUR_PROJECT_NAME.openai.azure.com/openai"
-env_key = "AZURE_OPENAI_API_KEY"  # Or "OPENAI_API_KEY", whichever you use.
-# Newer versions appear to support the responses API, see https://github.com/openai/codex/pull/1321
-query_params = { api-version = "2025-04-01-preview" }
-wire_api = "responses"
-```
-
-Once you have defined a provider you wish to use, you can configure it as your default provider as follows:
-
-```toml
-model_provider = "azure"
-```
-
-> [!TIP]
-> If you find yourself experimenting with a variety of models and providers, then you likely want to invest in defining a _profile_ for each configuration like so:
-
-```toml
-[profiles.o3]
-model_provider = "azure"
-model = "o3"
-
-[profiles.mistral]
-model_provider = "ollama"
-model = "mistral"
-```
-
-This way, you can specify one command-line argument (.e.g., `--profile o3`, `--profile mistral`) to override multiple settings together.
-
-</details>
-
-Codex can run fully locally against an OpenAI-compatible OSS host (like Ollama) using the `--oss` flag:
-
-- Interactive UI:
-  - codex --oss
-- Non-interactive (programmatic) mode:
-  - echo "Refactor utils" | codex exec --oss
-
-Model selection when using `--oss`:
-
-- If you omit `-m/--model`, Codex defaults to -m gpt-oss:20b and will verify it exists locally (downloading if needed).
-- To pick a different size, pass one of:
-  - -m "gpt-oss:20b"
-  - -m "gpt-oss:120b"
-
-Point Codex at your own OSS host:
-
-- By default, `--oss` talks to http://localhost:11434/v1.
-- To use a different host, set one of these environment variables before running Codex:
-  - CODEX_OSS_BASE_URL, for example:
-    - CODEX_OSS_BASE_URL="http://my-ollama.example.com:11434/v1" codex --oss -m gpt-oss:20b
-  - or CODEX_OSS_PORT (when the host is localhost):
-    - CODEX_OSS_PORT=11434 codex --oss
-
-Advanced: you can persist this in your config instead of environment variables by overriding the built-in `oss` provider in `~/.codex/config.toml`:
-
-```toml
-[model_providers.oss]
-name = "Open Source"
-base_url = "http://my-ollama.example.com:11434/v1"
-```
+Codex CLI supports a rich set of configuration options, with preferences stored in `~/.codex/config.toml`. For full configuration options, see [Configuration](./docs/config.md).
 
 ---
 
-### Platform sandboxing details
-
-By default, Codex CLI runs code and shell commands inside a restricted sandbox to protect your system.  
-
-> [!IMPORTANT]
-> Not all tool calls are sandboxed. Specifically, **trusted Model Context Protocol (MCP) tool calls** are executed outside of the sandbox.  
-> This is intentional: MCP tools are explicitly configured and trusted by you, and they often need to connect to **external applications or services** (e.g. issue trackers, databases, messaging systems).  
-> Running them outside the sandbox allows Codex to integrate with these external systems without being blocked by sandbox restrictions.
-
-The mechanism Codex uses to implement the sandbox policy depends on your OS:
-
-- **macOS 12+** uses **Apple Seatbelt** and runs commands using `sandbox-exec` with a profile (`-p`) that corresponds to the `--sandbox` that was specified.
-- **Linux** uses a combination of Landlock/seccomp APIs to enforce the `sandbox` configuration.
-
-Note that when running Linux in a containerized environment such as Docker, sandboxing may not work if the host/container configuration does not support the necessary Landlock/seccomp APIs. In such cases, we recommend configuring your Docker container so that it provides the sandbox guarantees you are looking for and then running `codex` with `--sandbox danger-full-access` (or, more simply, the `--dangerously-bypass-approvals-and-sandbox` flag) within your container.
-
----
-
-## Experimental technology disclaimer
-
-Codex CLI is an experimental project under active development. It is not yet stable, may contain bugs, incomplete features, or undergo breaking changes. We're building it in the open with the community and welcome:
-
-- Bug reports
-- Feature requests
-- Pull requests
-- Good vibes
-
-Help us improve by filing issues or submitting PRs (see the section below for how to contribute)!
-
----
-
-## System requirements
-
-| Requirement                 | Details                                                         |
-| --------------------------- | --------------------------------------------------------------- |
-| Operating systems           | macOS 12+, Ubuntu 20.04+/Debian 10+, or Windows 11 **via WSL2** |
-| Git (optional, recommended) | 2.23+ for built-in PR helpers                                   |
-| RAM                         | 4-GB minimum (8-GB recommended)                                 |
-
----
-
-## CLI reference
-
-| Command            | Purpose                            | Example                         |
-| ------------------ | ---------------------------------- | ------------------------------- |
-| `codex`            | Interactive TUI                    | `codex`                         |
-| `codex "..."`      | Initial prompt for interactive TUI | `codex "fix lint errors"`       |
-| `codex exec "..."` | Non-interactive "automation mode"  | `codex exec "explain utils.ts"` |
-
-Key flags: `--model/-m`, `--ask-for-approval/-a`.
-
----
-
-## Memory & project docs
-
-You can give Codex extra instructions and guidance using `AGENTS.md` files. Codex looks for `AGENTS.md` files in the following places, and merges them top-down:
-
-1. `~/.codex/AGENTS.md` - personal global guidance
-2. `AGENTS.md` at repo root - shared project notes
-3. `AGENTS.md` in the current working directory - sub-folder/feature specifics
-
----
-
-## Non-interactive / CI mode
-
-Run Codex head-less in pipelines. Example GitHub Action step:
-
-```yaml
-- name: Update changelog via Codex
-  run: |
-    npm install -g @openai/codex
-    export OPENAI_API_KEY="${{ secrets.OPENAI_KEY }}"
-    codex exec --full-auto "update CHANGELOG for next release"
-```
-
-## Model Context Protocol (MCP)
-
-The Codex CLI can be configured to leverage MCP servers by defining an [`mcp_servers`](./codex-rs/config.md#mcp_servers) section in `~/.codex/config.toml`. It is intended to mirror how tools such as Claude and Cursor define `mcpServers` in their respective JSON config files, though the Codex format is slightly different since it uses TOML rather than JSON, e.g.:
-
-```toml
-# IMPORTANT: the top-level key is `mcp_servers` rather than `mcpServers`.
-[mcp_servers.server-name]
-command = "npx"
-args = ["-y", "mcp-server"]
-env = { "API_KEY" = "value" }
-```
-
-> [!TIP]
-> It is somewhat experimental, but the Codex CLI can also be run as an MCP _server_ via `codex mcp`. If you launch it with an MCP client such as `npx @modelcontextprotocol/inspector codex mcp` and send it a `tools/list` request, you will see that there is only one tool, `codex`, that accepts a grab-bag of inputs, including a catch-all `config` map for anything you might want to override. Feel free to play around with it and provide feedback via GitHub issues.
-
-## Tracing / verbose logging
-
-Because Codex is written in Rust, it honors the `RUST_LOG` environment variable to configure its logging behavior.
-
-The TUI defaults to `RUST_LOG=codex_core=info,codex_tui=info` and log messages are written to `~/.codex/log/codex-tui.log`, so you can leave the following running in a separate terminal to monitor log messages as they are written:
-
-```
-tail -F ~/.codex/log/codex-tui.log
-```
-
-By comparison, the non-interactive mode (`codex exec`) defaults to `RUST_LOG=error`, but messages are printed inline, so there is no need to monitor a separate file.
-
-See the Rust documentation on [`RUST_LOG`](https://docs.rs/env_logger/latest/env_logger/#enabling-logging) for more information on the configuration options.
-
----
-
-### DotSlash
-
-The GitHub Release also contains a [DotSlash](https://dotslash-cli.com/) file for the Codex CLI named `codex`. Using a DotSlash file makes it possible to make a lightweight commit to source control to ensure all contributors use the same version of an executable, regardless of what platform they use for development.
-
-</details>
-
-<details>
-<summary><strong>Build from source</strong></summary>
-
-```bash
-# Clone the repository and navigate to the root of the Cargo workspace.
-git clone https://github.com/openai/codex.git
-cd codex/codex-rs
-
-# Install the Rust toolchain, if necessary.
-curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
-source "$HOME/.cargo/env"
-rustup component add rustfmt
-rustup component add clippy
-
-# Build Codex.
-cargo build
-
-# Launch the TUI with a sample prompt.
-cargo run --bin codex -- "explain this codebase to me"
-
-# After making changes, ensure the code is clean.
-cargo fmt -- --config imports_granularity=Item
-cargo clippy --tests
-
-# Run the tests.
-cargo test
-```
-
-</details>
-
----
-
-## Configuration
-
-Codex supports a rich set of configuration options documented in [`codex-rs/config.md`](./codex-rs/config.md).
-
-By default, Codex loads its configuration from `~/.codex/config.toml`.
-
-Though `--config` can be used to set/override ad-hoc config values for individual invocations of `codex`.
-
----
-
-## FAQ
-
-<details>
-<summary>OpenAI released a model called Codex in 2021 - is this related?</summary>
-
-In 2021, OpenAI released Codex, an AI system designed to generate code from natural language prompts. That original Codex model was deprecated as of March 2023 and is separate from the CLI tool.
-
-</details>
-
-<details>
-<summary>Which models are supported?</summary>
-
-Any model available with [Responses API](https://platform.openai.com/docs/api-reference/responses). The default is `o4-mini`, but pass `--model gpt-4.1` or set `model: gpt-4.1` in your config file to override.
-
-</details>
-<details>
-<summary>Why does <code>o3</code> or <code>o4-mini</code> not work for me?</summary>
-
-It's possible that your [API account needs to be verified](https://help.openai.com/en/articles/10910291-api-organization-verification) in order to start streaming responses and seeing chain of thought summaries from the API. If you're still running into issues, please let us know!
-
-</details>
-
-<details>
-<summary>How do I stop Codex from editing my files?</summary>
-
-Codex runs model-generated commands in a sandbox. If a proposed command or file change doesn't look right, you can simply type **n** to deny the command or give the model feedback.
-
-</details>
-<details>
-<summary>Does it work on Windows?</summary>
-
-Not directly. It requires [Windows Subsystem for Linux (WSL2)](https://learn.microsoft.com/en-us/windows/wsl/install) - Codex has been tested on macOS and Linux with Node 22.
-
-</details>
-
----
-
-## Zero data retention (ZDR) usage
-
-Codex CLI **does** support OpenAI organizations with [Zero Data Retention (ZDR)](https://platform.openai.com/docs/guides/your-data#zero-data-retention) enabled. If your OpenAI organization has Zero Data Retention enabled and you still encounter errors such as:
-
-```
-OpenAI rejected the request. Error details: Status: 400, Code: unsupported_parameter, Type: invalid_request_error, Message: 400 Previous response cannot be used for this organization due to Zero Data Retention.
-```
-
-Ensure you are running `codex` with `--config disable_response_storage=true` or add this line to `~/.codex/config.toml` to avoid specifying the command line option each time:
-
-```toml
-disable_response_storage = true
-```
-
-See [the configuration documentation on `disable_response_storage`](./codex-rs/config.md#disable_response_storage) for details.
-
----
-
-## Codex open source fund
-
-We're excited to launch a **$1 million initiative** supporting open source projects that use Codex CLI and other OpenAI models.
-
-- Grants are awarded up to **$25,000** API credits.
-- Applications are reviewed **on a rolling basis**.
-
-**Interested? [Apply here](https://openai.com/form/codex-open-source-fund/).**
-
----
-
-## Contributing
-
-This project is under active development and the code will likely change pretty significantly.
-
-**At the moment, we only plan to prioritize reviewing external contributions for bugs or security fixes.**
-
-If you want to add a new feature or change the behavior of an existing one, please open an issue proposing the feature and get approval from an OpenAI team member before spending time building it.
-
-**New contributions that don't go through this process may be closed** if they aren't aligned with our current roadmap or conflict with other priorities/upcoming features.
-
-### Development workflow
-
-- Create a _topic branch_ from `main` - e.g. `feat/interactive-prompt`.
-- Keep your changes focused. Multiple unrelated fixes should be opened as separate PRs.
-- Following the [development setup](#development-workflow) instructions above, ensure your change is free of lint warnings and test failures.
-
-### Writing high-impact code changes
-
-1. **Start with an issue.** Open a new one or comment on an existing discussion so we can agree on the solution before code is written.
-2. **Add or update tests.** Every new feature or bug-fix should come with test coverage that fails before your change and passes afterwards. 100% coverage is not required, but aim for meaningful assertions.
-3. **Document behaviour.** If your change affects user-facing behaviour, update the README, inline help (`codex --help`), or relevant example projects.
-4. **Keep commits atomic.** Each commit should compile and the tests should pass. This makes reviews and potential rollbacks easier.
-
-### Opening a pull request
-
-- Fill in the PR template (or include similar information) - **What? Why? How?**
-- Run **all** checks locally (`cargo test && cargo clippy --tests && cargo fmt -- --config imports_granularity=Item`). CI failures that could have been caught locally slow down the process.
-- Make sure your branch is up-to-date with `main` and that you have resolved merge conflicts.
-- Mark the PR as **Ready for review** only when you believe it is in a merge-able state.
-
-### Review process
-
-1. One maintainer will be assigned as a primary reviewer.
-2. If your PR adds a new feature that was not previously discussed and approved, we may choose to close your PR (see [Contributing](#contributing)).
-3. We may ask for changes - please do not take this personally. We value the work, but we also value consistency and long-term maintainability.
-5. When there is consensus that the PR meets the bar, a maintainer will squash-and-merge.
-
-### Community values
-
-- **Be kind and inclusive.** Treat others with respect; we follow the [Contributor Covenant](https://www.contributor-covenant.org/).
-- **Assume good intent.** Written communication is hard - err on the side of generosity.
-- **Teach & learn.** If you spot something confusing, open an issue or PR with improvements.
-
-### Getting help
-
-If you run into problems setting up the project, would like feedback on an idea, or just want to say _hi_ - please open a Discussion or jump into the relevant issue. We are happy to help.
-
-Together we can make Codex CLI an incredible tool. **Happy hacking!** :rocket:
-
-### Contributor license agreement (CLA)
-
-All contributors **must** accept the CLA. The process is lightweight:
-
-1. Open your pull request.
-2. Paste the following comment (or reply `recheck` if you've signed before):
-
-   ```text
-   I have read the CLA Document and I hereby sign the CLA
-   ```
-
-3. The CLA-Assistant bot records your signature in the repo and marks the status check as passed.
-
-No special Git commands, email attachments, or commit footers required.
-
-#### Quick fixes
-
-| Scenario          | Command                                          |
-| ----------------- | ------------------------------------------------ |
-| Amend last commit | `git commit --amend -s --no-edit && git push -f` |
-
-The **DCO check** blocks merges until every commit in the PR carries the footer (with squash this is just the one).
-
-### Releasing `codex`
-
-_For admins only._
-
-Make sure you are on `main` and have no local changes. Then run:
-
-```shell
-VERSION=0.2.0  # Can also be 0.2.0-alpha.1 or any valid Rust version.
-./codex-rs/scripts/create_github_release.sh "$VERSION"
-```
-
-This will make a local commit on top of `main` with `version` set to `$VERSION` in `codex-rs/Cargo.toml` (note that on `main`, we leave the version as `version = "0.0.0"`).
-
-This will push the commit using the tag `rust-v${VERSION}`, which in turn kicks off [the release workflow](.github/workflows/rust-release.yml). This will create a new GitHub Release named `$VERSION`.
-
-If everything looks good in the generated GitHub Release, uncheck the **pre-release** box so it is the latest release.
-
-Create a PR to update [`Formula/c/codex.rb`](https://github.com/Homebrew/homebrew-core/blob/main/Formula/c/codex.rb) on Homebrew.
-
----
-
-## Security & responsible AI
-
-Have you discovered a vulnerability or have concerns about model output? Please e-mail **security@openai.com** and we will respond promptly.
+### Docs & FAQ
+
+- [**Getting started**](./docs/getting-started.md)
+  - [CLI usage](./docs/getting-started.md#cli-usage)
+  - [Running with a prompt as input](./docs/getting-started.md#running-with-a-prompt-as-input)
+  - [Example prompts](./docs/getting-started.md#example-prompts)
+  - [Memory with AGENTS.md](./docs/getting-started.md#memory--project-docs)
+  - [Configuration](./docs/config.md)
+- [**Sandbox & approvals**](./docs/sandbox.md)
+- [**Authentication**](./docs/authentication.md)
+  - [Auth methods](./docs/authentication.md#forcing-a-specific-auth-method-advanced)
+  - [Login on a "Headless" machine](./docs/authentication.md#connecting-on-a-headless-machine)
+- [**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)
+- [**Contributing**](./docs/contributing.md)
+- [**Install & build**](./docs/install.md)
+  - [System Requirements](./docs/install.md#system-requirements)
+  - [DotSlash](./docs/install.md#dotslash)
+  - [Build from source](./docs/install.md#build-from-source)
+- [**FAQ**](./docs/faq.md)
+- [**Open source fund**](./docs/open-source-fund.md)
 
 ---
 
 ## License
 
 This repository is licensed under the [Apache-2.0 License](LICENSE).
+

codex-rs/Cargo.lock

@@ -635,6 +635,7 @@ name = "codex-apply-patch"
 version = "0.0.0"
 dependencies = [
  "anyhow",
+ "assert_cmd",
  "pretty_assertions",
  "similar",
  "tempfile",
@@ -652,6 +653,7 @@ dependencies = [
  "codex-core",
  "codex-linux-sandbox",
  "dotenvy",
+ "tempfile",
  "tokio",
 ]
 
@@ -731,6 +733,7 @@ dependencies = [
  "mime_guess",
  "openssl-sys",
  "os_info",
+ "portable-pty",
  "predicates",
  "pretty_assertions",
  "rand 0.9.2",
@@ -751,7 +754,7 @@ dependencies = [
  "tokio-test",
  "tokio-util",
  "toml 0.9.5",
- "toml_edit 0.23.3",
+ "toml_edit 0.23.4",
  "tracing",
  "tree-sitter",
  "tree-sitter-bash",
@@ -998,6 +1001,7 @@ dependencies = [
  "tui-markdown",
  "unicode-segmentation",
  "unicode-width 0.1.14",
+ "url",
  "uuid",
  "vt100",
 ]
@@ -1479,6 +1483,12 @@ version = "0.15.7"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "1aaf95b3e5c8f23aa320147307562d361db0ae0d51242340f558153b4eb2439b"
 
+[[package]]
+name = "downcast-rs"
+version = "1.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "75b325c5dbd37f80359721ad39aca5a29fb04c89279657cffdda8736d0c0b9d2"
+
 [[package]]
 name = "dupe"
 version = "0.9.1"
@@ -1724,6 +1734,17 @@ dependencies = [
  "simd-adler32",
 ]
 
+[[package]]
+name = "filedescriptor"
+version = "0.8.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e40758ed24c9b2eeb76c35fb0aebc66c626084edd827e07e1552279814c6682d"
+dependencies = [
+ "libc",
+ "thiserror 1.0.69",
+ "winapi",
+]
+
 [[package]]
 name = "fixedbitset"
 version = "0.4.2"
@@ -2702,6 +2723,7 @@ checksum = "4488594b9328dee448adb906d8b126d9b7deb7cf5c22161ee591610bb1be83c0"
 dependencies = [
  "bitflags 2.9.1",
  "libc",
+ "redox_syscall",
 ]
 
 [[package]]
@@ -3439,6 +3461,27 @@ dependencies = [
  "portable-atomic",
 ]
 
+[[package]]
+name = "portable-pty"
+version = "0.9.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b4a596a2b3d2752d94f51fac2d4a96737b8705dddd311a32b9af47211f08671e"
+dependencies = [
+ "anyhow",
+ "bitflags 1.3.2",
+ "downcast-rs",
+ "filedescriptor",
+ "lazy_static",
+ "libc",
+ "log",
+ "nix",
+ "serial2",
+ "shared_library",
+ "shell-words",
+ "winapi",
+ "winreg",
+]
+
 [[package]]
 name = "potential_utf"
 version = "0.1.2"
@@ -4366,6 +4409,17 @@ dependencies = [
  "syn 2.0.104",
 ]
 
+[[package]]
+name = "serial2"
+version = "0.2.31"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "26e1e5956803a69ddd72ce2de337b577898801528749565def03515f82bad5bb"
+dependencies = [
+ "cfg-if",
+ "libc",
+ "winapi",
+]
+
 [[package]]
 name = "sha1"
 version = "0.10.6"
@@ -4397,6 +4451,22 @@ dependencies = [
  "lazy_static",
 ]
 
+[[package]]
+name = "shared_library"
+version = "0.1.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5a9e7e0f2bfae24d8a5b5a66c5b257a83c7412304311512a0c054cd5e619da11"
+dependencies = [
+ "lazy_static",
+ "libc",
+]
+
+[[package]]
+name = "shell-words"
+version = "1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "24188a676b6ae68c3b2cb3a01be17fbf7240ce009799bb56d5b1409051e78fde"
+
 [[package]]
 name = "shlex"
 version = "1.3.0"
@@ -5126,9 +5196,9 @@ dependencies = [
 
 [[package]]
 name = "toml_edit"
-version = "0.23.3"
+version = "0.23.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "17d3b47e6b7a040216ae5302712c94d1cf88c95b47efa80e2c59ce96c878267e"
+checksum = "7211ff1b8f0d3adae1663b7da9ffe396eabe1ca25f0b0bee42b0da29a9ddce93"
 dependencies = [
  "indexmap 2.10.0",
  "toml_datetime 0.7.0",
@@ -5709,11 +5779,11 @@ dependencies = [
 
 [[package]]
 name = "whoami"
-version = "1.6.0"
+version = "1.6.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6994d13118ab492c3c80c1f81928718159254c53c472bf9ce36f8dae4add02a7"
+checksum = "5d4a4db5077702ca3015d3d02d74974948aba2ad9e12ab7df718ee64ccd7e97d"
 dependencies = [
- "redox_syscall",
+ "libredox",
  "wasite",
  "web-sys",
 ]
@@ -6176,6 +6246,15 @@ dependencies = [
  "memchr",
 ]
 
+[[package]]
+name = "winreg"
+version = "0.10.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "80d0f4e272c85def139476380b12f9ac60926689dd2e01d4923222f40580869d"
+dependencies = [
+ "winapi",
+]
+
 [[package]]
 name = "winsafe"
 version = "0.0.19"

codex-rs/Cargo.toml

@@ -34,6 +34,7 @@ rust = {}
 
 [workspace.lints.clippy]
 expect_used = "deny"
+uninlined_format_args = "deny"
 unwrap_used = "deny"
 
 [profile.release]

codex-rs/README.md

@@ -19,11 +19,11 @@ While we are [working to close the gap between the TypeScript and Rust implement
 
 ### Config
 
-Codex supports a rich set of configuration options. Note that the Rust CLI uses `config.toml` instead of `config.json`. See [`config.md`](./config.md) for details.
+Codex supports a rich set of configuration options. Note that the Rust CLI uses `config.toml` instead of `config.json`. See [`docs/config.md`](../docs/config.md) for details.
 
 ### Model Context Protocol Support
 
-Codex CLI functions as an MCP client that can connect to MCP servers on startup. See the [`mcp_servers`](./config.md#mcp_servers) section in the configuration documentation for details.
+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.
 
 It is still experimental, but you can also launch Codex as an MCP _server_ by running `codex mcp`. Use the [`@modelcontextprotocol/inspector`](https://github.com/modelcontextprotocol/inspector) to try it out:
 
@@ -33,7 +33,7 @@ npx @modelcontextprotocol/inspector codex mcp
 
 ### Notifications
 
-You can enable notifications by configuring a script that is run whenever the agent finishes a turn. The [notify documentation](./config.md#notify) includes a detailed example that explains how to get desktop notifications via [terminal-notifier](https://github.com/julienXX/terminal-notifier) on macOS.
+You can enable notifications by configuring a script that is run whenever the agent finishes a turn. The [notify documentation](../docs/config.md#notify) includes a detailed example that explains how to get desktop notifications via [terminal-notifier](https://github.com/julienXX/terminal-notifier) on macOS.
 
 ### `codex exec` to run Codex programmatially/non-interactively
 
@@ -43,6 +43,12 @@ To run Codex non-interactively, run `codex exec PROMPT` (you can also pass the p
 
 Typing `@` triggers a fuzzy-filename search over the workspace root. Use up/down to select among the results and Tab or Enter to replace the `@` with the selected path. You can use Esc to cancel the search.
 
+### Esc–Esc to edit a previous message
+
+When the chat composer is empty, press Esc to prime “backtrack” mode. Press Esc again to open a transcript preview highlighting the last user message; press Esc repeatedly to step to older user messages. Press Enter to confirm and Codex will fork the conversation from that point, trim the visible transcript accordingly, and pre‑fill the composer with the selected user message so you can edit and resubmit it.
+
+In the transcript preview, the footer shows an `Esc edit prev` hint while editing is active.
+
 ### `--cd`/`-C` flag
 
 Sometimes it is not convenient to `cd` to the directory you want Codex to use as the "working root" before running Codex. Fortunately, `codex` supports a `--cd` option so you can specify whatever folder you want. You can confirm that Codex is honoring `--cd` by double-checking the **workdir** it reports in the TUI at the start of a new session.

codex-rs/apply-patch/Cargo.toml

@@ -7,6 +7,10 @@ version = { workspace = true }
 name = "codex_apply_patch"
 path = "src/lib.rs"
 
+[[bin]]
+name = "apply_patch"
+path = "src/main.rs"
+
 [lints]
 workspace = true
 
@@ -18,5 +22,6 @@ tree-sitter = "0.25.8"
 tree-sitter-bash = "0.25.0"
 
 [dev-dependencies]
+assert_cmd = "2"
 pretty_assertions = "1.4.1"
 tempfile = "3.13.0"

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

@@ -1,5 +1,6 @@
 mod parser;
 mod seek_sequence;
+mod standalone_executable;
 
 use std::collections::HashMap;
 use std::path::Path;
@@ -19,6 +20,8 @@ use tree_sitter::LanguageError;
 use tree_sitter::Parser;
 use tree_sitter_bash::LANGUAGE as BASH;
 
+pub use standalone_executable::main;
+
 /// Detailed instructions for gpt-4.1 on how to use the `apply_patch` tool.
 pub const APPLY_PATCH_TOOL_INSTRUCTIONS: &str = include_str!("../apply_patch_tool_instructions.md");
 

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

@@ -0,0 +1,3 @@
+pub fn main() -> ! {
+    codex_apply_patch::main()
+}

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

@@ -0,0 +1,59 @@
+use std::io::Read;
+use std::io::Write;
+
+pub fn main() -> ! {
+    let exit_code = run_main();
+    std::process::exit(exit_code);
+}
+
+/// We would prefer to return `std::process::ExitCode`, but its `exit_process()`
+/// method is still a nightly API and we want main() to return !.
+pub fn run_main() -> i32 {
+    // Expect either one argument (the full apply_patch payload) or read it from stdin.
+    let mut args = std::env::args_os();
+    let _argv0 = args.next();
+
+    let patch_arg = match args.next() {
+        Some(arg) => match arg.into_string() {
+            Ok(s) => s,
+            Err(_) => {
+                eprintln!("Error: apply_patch requires a UTF-8 PATCH argument.");
+                return 1;
+            }
+        },
+        None => {
+            // No argument provided; attempt to read the patch from stdin.
+            let mut buf = String::new();
+            match std::io::stdin().read_to_string(&mut buf) {
+                Ok(_) => {
+                    if buf.is_empty() {
+                        eprintln!("Usage: apply_patch 'PATCH'\n       echo 'PATCH' | apply-patch");
+                        return 2;
+                    }
+                    buf
+                }
+                Err(err) => {
+                    eprintln!("Error: Failed to read PATCH from stdin.\n{err}");
+                    return 1;
+                }
+            }
+        }
+    };
+
+    // Refuse extra args to avoid ambiguity.
+    if args.next().is_some() {
+        eprintln!("Error: apply_patch accepts exactly one argument.");
+        return 2;
+    }
+
+    let mut stdout = std::io::stdout();
+    let mut stderr = std::io::stderr();
+    match crate::apply_patch(&patch_arg, &mut stdout, &mut stderr) {
+        Ok(()) => {
+            // Flush to ensure output ordering when used in pipelines.
+            let _ = stdout.flush();
+            0
+        }
+        Err(_) => 1,
+    }
+}

codex-rs/apply-patch/tests/all.rs

@@ -0,0 +1,3 @@
+// Single integration test binary that aggregates all test modules.
+// The submodules live in `tests/suite/`.
+mod suite;

codex-rs/apply-patch/tests/suite/cli.rs

@@ -0,0 +1,90 @@
+use assert_cmd::prelude::*;
+use std::fs;
+use std::process::Command;
+use tempfile::tempdir;
+
+#[test]
+fn test_apply_patch_cli_add_and_update() -> anyhow::Result<()> {
+    let tmp = tempdir()?;
+    let file = "cli_test.txt";
+    let absolute_path = tmp.path().join(file);
+
+    // 1) Add a file
+    let add_patch = format!(
+        r#"*** Begin Patch
+*** Add File: {file}
++hello
+*** End Patch"#
+    );
+    Command::cargo_bin("apply_patch")
+        .expect("should find apply_patch binary")
+        .arg(add_patch)
+        .current_dir(tmp.path())
+        .assert()
+        .success()
+        .stdout(format!("Success. Updated the following files:\nA {file}\n"));
+    assert_eq!(fs::read_to_string(&absolute_path)?, "hello\n");
+
+    // 2) Update the file
+    let update_patch = format!(
+        r#"*** Begin Patch
+*** Update File: {file}
+@@
+-hello
++world
+*** End Patch"#
+    );
+    Command::cargo_bin("apply_patch")
+        .expect("should find apply_patch binary")
+        .arg(update_patch)
+        .current_dir(tmp.path())
+        .assert()
+        .success()
+        .stdout(format!("Success. Updated the following files:\nM {file}\n"));
+    assert_eq!(fs::read_to_string(&absolute_path)?, "world\n");
+
+    Ok(())
+}
+
+#[test]
+fn test_apply_patch_cli_stdin_add_and_update() -> anyhow::Result<()> {
+    let tmp = tempdir()?;
+    let file = "cli_test_stdin.txt";
+    let absolute_path = tmp.path().join(file);
+
+    // 1) Add a file via stdin
+    let add_patch = format!(
+        r#"*** Begin Patch
+*** Add File: {file}
++hello
+*** End Patch"#
+    );
+    let mut cmd =
+        assert_cmd::Command::cargo_bin("apply_patch").expect("should find apply_patch binary");
+    cmd.current_dir(tmp.path());
+    cmd.write_stdin(add_patch)
+        .assert()
+        .success()
+        .stdout(format!("Success. Updated the following files:\nA {file}\n"));
+    assert_eq!(fs::read_to_string(&absolute_path)?, "hello\n");
+
+    // 2) Update the file via stdin
+    let update_patch = format!(
+        r#"*** Begin Patch
+*** Update File: {file}
+@@
+-hello
++world
+*** End Patch"#
+    );
+    let mut cmd =
+        assert_cmd::Command::cargo_bin("apply_patch").expect("should find apply_patch binary");
+    cmd.current_dir(tmp.path());
+    cmd.write_stdin(update_patch)
+        .assert()
+        .success()
+        .stdout(format!("Success. Updated the following files:\nM {file}\n"));
+    assert_eq!(fs::read_to_string(&absolute_path)?, "world\n");
+
+    Ok(())
+}

codex-rs/apply-patch/tests/suite/mod.rs

@@ -0,0 +1 @@
+mod cli;

codex-rs/arg0/Cargo.toml

@@ -16,4 +16,5 @@ codex-apply-patch = { path = "../apply-patch" }
 codex-core = { path = "../core" }
 codex-linux-sandbox = { path = "../linux-sandbox" }
 dotenvy = "0.15.7"
+tempfile = "3"
 tokio = { version = "1", features = ["rt-multi-thread"] }

codex-rs/arg0/src/lib.rs

@@ -3,6 +3,13 @@ use std::path::Path;
 use std::path::PathBuf;
 
 use codex_core::CODEX_APPLY_PATCH_ARG1;
+#[cfg(unix)]
+use std::os::unix::fs::symlink;
+use tempfile::TempDir;
+
+const LINUX_SANDBOX_ARG0: &str = "codex-linux-sandbox";
+const APPLY_PATCH_ARG0: &str = "apply_patch";
+const MISSPELLED_APPLY_PATCH_ARG0: &str = "applypatch";
 
 /// While we want to deploy the Codex CLI as a single executable for simplicity,
 /// we also want to expose some of its functionality as distinct CLIs, so we use
@@ -39,9 +46,11 @@ where
         .and_then(|s| s.to_str())
         .unwrap_or("");
 
-    if exe_name == "codex-linux-sandbox" {
+    if exe_name == LINUX_SANDBOX_ARG0 {
         // Safety: [`run_main`] never returns.
         codex_linux_sandbox::run_main();
+    } else if exe_name == APPLY_PATCH_ARG0 || exe_name == MISSPELLED_APPLY_PATCH_ARG0 {
+        codex_apply_patch::main();
     }
 
     let argv1 = args.next().unwrap_or_default();
@@ -68,6 +77,19 @@ where
     // before creating any threads/the Tokio runtime.
     load_dotenv();
 
+    // Retain the TempDir so it exists for the lifetime of the invocation of
+    // this executable. Admittedly, we could invoke `keep()` on it, but it
+    // would be nice to avoid leaving temporary directories behind, if possible.
+    let _path_entry = match prepend_path_entry_for_apply_patch() {
+        Ok(path_entry) => Some(path_entry),
+        Err(err) => {
+            // It is possible that Codex will proceed successfully even if
+            // updating the PATH fails, so warn the user and move on.
+            eprintln!("WARNING: proceeding, even though we could not update PATH: {err}");
+            None
+        }
+    };
+
     // Regular invocation – create a Tokio runtime and execute the provided
     // async entry-point.
     let runtime = tokio::runtime::Runtime::new()?;
@@ -113,3 +135,67 @@ where
         }
     }
 }
+
+/// Creates a temporary directory with either:
+///
+/// - UNIX: `apply_patch` symlink to the current executable
+/// - WINDOWS: `apply_patch.bat` batch script to invoke the current executable
+///   with the "secret" --codex-run-as-apply-patch flag.
+///
+/// This temporary directory is prepended to the PATH environment variable so
+/// that `apply_patch` can be on the PATH without requiring the user to
+/// install a separate `apply_patch` executable, simplifying the deployment of
+/// Codex CLI.
+///
+/// IMPORTANT: This function modifies the PATH environment variable, so it MUST
+/// be called before multiple threads are spawned.
+fn prepend_path_entry_for_apply_patch() -> std::io::Result<TempDir> {
+    let temp_dir = TempDir::new()?;
+    let path = temp_dir.path();
+
+    for filename in &[APPLY_PATCH_ARG0, MISSPELLED_APPLY_PATCH_ARG0] {
+        let exe = std::env::current_exe()?;
+
+        #[cfg(unix)]
+        {
+            let link = path.join(filename);
+            symlink(&exe, &link)?;
+        }
+
+        #[cfg(windows)]
+        {
+            let batch_script = path.join(format!("{filename}.bat"));
+            std::fs::write(
+                &batch_script,
+                format!(
+                    r#"@echo off
+"{}" {CODEX_APPLY_PATCH_ARG1} %*
+"#,
+                    exe.display()
+                ),
+            )?;
+        }
+    }
+
+    #[cfg(unix)]
+    const PATH_SEPARATOR: &str = ":";
+
+    #[cfg(windows)]
+    const PATH_SEPARATOR: &str = ";";
+
+    let path_element = path.display();
+    let updated_path_env_var = match std::env::var("PATH") {
+        Ok(existing_path) => {
+            format!("{path_element}{PATH_SEPARATOR}{existing_path}")
+        }
+        Err(_) => {
+            format!("{path_element}")
+        }
+    };
+
+    unsafe {
+        std::env::set_var("PATH", updated_path_env_var);
+    }
+
+    Ok(temp_dir)
+}

codex-rs/chatgpt/tests/all.rs

@@ -0,0 +1,3 @@
+// Single integration test binary that aggregates all test modules.
+// The submodules live in `tests/suite/`.
+mod suite;

codex-rs/chatgpt/tests/suite/mod.rs

@@ -0,0 +1,2 @@
+// Aggregates all former standalone integration tests as modules.
+mod apply_command_e2e;

codex-rs/config.md

@@ -1,564 +1,6 @@
-# Config
+# Configuration docs moved
 
-Codex supports several mechanisms for setting config values:
+This file has moved. Please see the latest configuration documentation here:
 
-- Config-specific command-line flags, such as `--model o3` (highest precedence).
-- A generic `-c`/`--config` flag that takes a `key=value` pair, such as `--config model="o3"`.
-  - The key can contain dots to set a value deeper than the root, e.g. `--config model_providers.openai.wire_api="chat"`.
-  - Values can contain objects, such as `--config shell_environment_policy.include_only=["PATH", "HOME", "USER"]`.
-  - For consistency with `config.toml`, values are in TOML format rather than JSON format, so use `{a = 1, b = 2}` rather than `{"a": 1, "b": 2}`.
-  - If `value` cannot be parsed as a valid TOML value, it is treated as a string value. This means that both `-c model="o3"` and `-c model=o3` are equivalent.
-- The `$CODEX_HOME/config.toml` configuration file where the `CODEX_HOME` environment value defaults to `~/.codex`. (Note `CODEX_HOME` will also be where logs and other Codex-related information are stored.)
-
-Both the `--config` flag and the `config.toml` file support the following options:
-
-## model
-
-The model that Codex should use.
-
-```toml
-model = "o3"  # overrides the default of "gpt-5"
-```
-
-## model_providers
-
-This option lets you override and amend the default set of model providers bundled with Codex. This value is a map where the key is the value to use with `model_provider` to select the corresponding provider.
-
-For example, if you wanted to add a provider that uses the OpenAI 4o model via the chat completions API, then you could add the following configuration:
-
-```toml
-# Recall that in TOML, root keys must be listed before tables.
-model = "gpt-4o"
-model_provider = "openai-chat-completions"
-
-[model_providers.openai-chat-completions]
-# Name of the provider that will be displayed in the Codex UI.
-name = "OpenAI using Chat Completions"
-# The path `/chat/completions` will be amended to this URL to make the POST
-# request for the chat completions.
-base_url = "https://api.openai.com/v1"
-# If `env_key` is set, identifies an environment variable that must be set when
-# using Codex with this provider. The value of the environment variable must be
-# non-empty and will be used in the `Bearer TOKEN` HTTP header for the POST request.
-env_key = "OPENAI_API_KEY"
-# Valid values for wire_api are "chat" and "responses". Defaults to "chat" if omitted.
-wire_api = "chat"
-# If necessary, extra query params that need to be added to the URL.
-# See the Azure example below.
-query_params = {}
-```
-
-Note this makes it possible to use Codex CLI with non-OpenAI models, so long as they use a wire API that is compatible with the OpenAI chat completions API. For example, you could define the following provider to use Codex CLI with Ollama running locally:
-
-```toml
-[model_providers.ollama]
-name = "Ollama"
-base_url = "http://localhost:11434/v1"
-```
-
-Or a third-party provider (using a distinct environment variable for the API key):
-
-```toml
-[model_providers.mistral]
-name = "Mistral"
-base_url = "https://api.mistral.ai/v1"
-env_key = "MISTRAL_API_KEY"
-```
-
-Note that Azure requires `api-version` to be passed as a query parameter, so be sure to specify it as part of `query_params` when defining the Azure provider:
-
-```toml
-[model_providers.azure]
-name = "Azure"
-# Make sure you set the appropriate subdomain for this URL.
-base_url = "https://YOUR_PROJECT_NAME.openai.azure.com/openai"
-env_key = "AZURE_OPENAI_API_KEY"  # Or "OPENAI_API_KEY", whichever you use.
-query_params = { api-version = "2025-04-01-preview" }
-```
-
-It is also possible to configure a provider to include extra HTTP headers with a request. These can be hardcoded values (`http_headers`) or values read from environment variables (`env_http_headers`):
-
-```toml
-[model_providers.example]
-# name, base_url, ...
-
-# This will add the HTTP header `X-Example-Header` with value `example-value`
-# to each request to the model provider.
-http_headers = { "X-Example-Header" = "example-value" }
-
-# This will add the HTTP header `X-Example-Features` with the value of the
-# `EXAMPLE_FEATURES` environment variable to each request to the model provider
-# _if_ the environment variable is set and its value is non-empty.
-env_http_headers = { "X-Example-Features": "EXAMPLE_FEATURES" }
-```
-
-### Per-provider network tuning
-
-The following optional settings control retry behaviour and streaming idle timeouts **per model provider**. They must be specified inside the corresponding `[model_providers.<id>]` block in `config.toml`. (Older releases accepted top‑level keys; those are now ignored.)
-
-Example:
-
-```toml
-[model_providers.openai]
-name = "OpenAI"
-base_url = "https://api.openai.com/v1"
-env_key = "OPENAI_API_KEY"
-# network tuning overrides (all optional; falls back to built‑in defaults)
-request_max_retries = 4            # retry failed HTTP requests
-stream_max_retries = 10            # retry dropped SSE streams
-stream_idle_timeout_ms = 300000    # 5m idle timeout
-```
-
-#### request_max_retries
-
-How many times Codex will retry a failed HTTP request to the model provider. Defaults to `4`.
-
-#### stream_max_retries
-
-Number of times Codex will attempt to reconnect when a streaming response is interrupted. Defaults to `10`.
-
-#### stream_idle_timeout_ms
-
-How long Codex will wait for activity on a streaming response before treating the connection as lost. Defaults to `300_000` (5 minutes).
-
-## model_provider
-
-Identifies which provider to use from the `model_providers` map. Defaults to `"openai"`. You can override the `base_url` for the built-in `openai` provider via the `OPENAI_BASE_URL` environment variable.
-
-Note that if you override `model_provider`, then you likely want to override
-`model`, as well. For example, if you are running ollama with Mistral locally,
-then you would need to add the following to your config in addition to the new entry in the `model_providers` map:
-
-```toml
-model_provider = "ollama"
-model = "mistral"
-```
-
-## approval_policy
-
-Determines when the user should be prompted to approve whether Codex can execute a command:
-
-```toml
-# Codex has hardcoded logic that defines a set of "trusted" commands.
-# Setting the approval_policy to `untrusted` means that Codex will prompt the
-# user before running a command not in the "trusted" set.
-#
-# See https://github.com/openai/codex/issues/1260 for the plan to enable
-# end-users to define their own trusted commands.
-approval_policy = "untrusted"
-```
-
-If you want to be notified whenever a command fails, use "on-failure":
-
-```toml
-# If the command fails when run in the sandbox, Codex asks for permission to
-# retry the command outside the sandbox.
-approval_policy = "on-failure"
-```
-
-If you want the model to run until it decides that it needs to ask you for escalated permissions, use "on-request":
-
-```toml
-# The model decides when to escalate
-approval_policy = "on-request"
-```
-
-Alternatively, you can have the model run until it is done, and never ask to run a command with escalated permissions:
-
-```toml
-# User is never prompted: if the command fails, Codex will automatically try
-# something out. Note the `exec` subcommand always uses this mode.
-approval_policy = "never"
-```
-
-## profiles
-
-A _profile_ is a collection of configuration values that can be set together. Multiple profiles can be defined in `config.toml` and you can specify the one you
-want to use at runtime via the `--profile` flag.
-
-Here is an example of a `config.toml` that defines multiple profiles:
-
-```toml
-model = "o3"
-approval_policy = "unless-allow-listed"
-disable_response_storage = false
-
-# Setting `profile` is equivalent to specifying `--profile o3` on the command
-# line, though the `--profile` flag can still be used to override this value.
-profile = "o3"
-
-[model_providers.openai-chat-completions]
-name = "OpenAI using Chat Completions"
-base_url = "https://api.openai.com/v1"
-env_key = "OPENAI_API_KEY"
-wire_api = "chat"
-
-[profiles.o3]
-model = "o3"
-model_provider = "openai"
-approval_policy = "never"
-model_reasoning_effort = "high"
-model_reasoning_summary = "detailed"
-
-[profiles.gpt3]
-model = "gpt-3.5-turbo"
-model_provider = "openai-chat-completions"
-
-[profiles.zdr]
-model = "o3"
-model_provider = "openai"
-approval_policy = "on-failure"
-disable_response_storage = true
-```
-
-Users can specify config values at multiple levels. Order of precedence is as follows:
-
-1. custom command-line argument, e.g., `--model o3`
-2. as part of a profile, where the `--profile` is specified via a CLI (or in the config file itself)
-3. as an entry in `config.toml`, e.g., `model = "o3"`
-4. the default value that comes with Codex CLI (i.e., Codex CLI defaults to `gpt-5`)
-
-## model_reasoning_effort
-
-If the selected model is known to support reasoning (for example: `o3`, `o4-mini`, `codex-*`, `gpt-5`), reasoning is enabled by default when using the Responses API. As explained in the [OpenAI Platform documentation](https://platform.openai.com/docs/guides/reasoning?api-mode=responses#get-started-with-reasoning), this can be set to:
-
-- `"minimal"`
-- `"low"`
-- `"medium"` (default)
-- `"high"`
-
-Note: to minimize reasoning, choose `"minimal"`.
-
-## model_reasoning_summary
-
-If the model name starts with `"o"` (as in `"o3"` or `"o4-mini"`) or `"codex"`, reasoning is enabled by default when using the Responses API. As explained in the [OpenAI Platform documentation](https://platform.openai.com/docs/guides/reasoning?api-mode=responses#reasoning-summaries), this can be set to:
-
-- `"auto"` (default)
-- `"concise"`
-- `"detailed"`
-
-To disable reasoning summaries, set `model_reasoning_summary` to `"none"` in your config:
-
-```toml
-model_reasoning_summary = "none"  # disable reasoning summaries
-```
-
-## model_verbosity
-
-Controls output length/detail on GPT‑5 family models when using the Responses API. Supported values:
-
-- `"low"`
-- `"medium"` (default when omitted)
-- `"high"`
-
-When set, Codex includes a `text` object in the request payload with the configured verbosity, for example: `"text": { "verbosity": "low" }`.
-
-Example:
-
-```toml
-model = "gpt-5"
-model_verbosity = "low"
-```
-
-Note: This applies only to providers using the Responses API. Chat Completions providers are unaffected.
-
-## model_supports_reasoning_summaries
-
-By default, `reasoning` is only set on requests to OpenAI models that are known to support them. To force `reasoning` to set on requests to the current model, you can force this behavior by setting the following in `config.toml`:
-
-```toml
-model_supports_reasoning_summaries = true
-```
-
-## sandbox_mode
-
-Codex executes model-generated shell commands inside an OS-level sandbox.
-
-In most cases you can pick the desired behaviour with a single option:
-
-```toml
-# same as `--sandbox read-only`
-sandbox_mode = "read-only"
-```
-
-The default policy is `read-only`, which means commands can read any file on
-disk, but attempts to write a file or access the network will be blocked.
-
-A more relaxed policy is `workspace-write`. When specified, the current working directory for the Codex task will be writable (as well as `$TMPDIR` on macOS). Note that the CLI defaults to using the directory where it was spawned as `cwd`, though this can be overridden using `--cwd/-C`.
-
-On macOS (and soon Linux), all writable roots (including `cwd`) that contain a `.git/` folder _as an immediate child_ will configure the `.git/` folder to be read-only while the rest of the Git repository will be writable. This means that commands like `git commit` will fail, by default (as it entails writing to `.git/`), and will require Codex to ask for permission.
-
-```toml
-# same as `--sandbox workspace-write`
-sandbox_mode = "workspace-write"
-
-# Extra settings that only apply when `sandbox = "workspace-write"`.
-[sandbox_workspace_write]
-# By default, the cwd for the Codex session will be writable as well as $TMPDIR
-# (if set) and /tmp (if it exists). Setting the respective options to `true`
-# will override those defaults.
-exclude_tmpdir_env_var = false
-exclude_slash_tmp = false
-
-# Optional list of _additional_ writable roots beyond $TMPDIR and /tmp.
-writable_roots = ["/Users/YOU/.pyenv/shims"]
-
-# Allow the command being run inside the sandbox to make outbound network
-# requests. Disabled by default.
-network_access = false
-```
-
-To disable sandboxing altogether, specify `danger-full-access` like so:
-
-```toml
-# same as `--sandbox danger-full-access`
-sandbox_mode = "danger-full-access"
-```
-
-This is reasonable to use if Codex is running in an environment that provides its own sandboxing (such as a Docker container) such that further sandboxing is unnecessary.
-
-Though using this option may also be necessary if you try to use Codex in environments where its native sandboxing mechanisms are unsupported, such as older Linux kernels or on Windows.
-
-## Approval presets
-
-Codex provides three main Approval Presets:
-
-- Read Only: Codex can read files and answer questions; edits, running commands, and network access require approval.
-- Auto: Codex can read files, make edits, and run commands in the workspace without approval; asks for approval outside the workspace or for network access.
-- Full Access: Full disk and network access without prompts; extremely risky.
-
-You can further customize how Codex runs at the command line using the `--ask-for-approval` and `--sandbox` options.
-
-## mcp_servers
-
-Defines the list of MCP servers that Codex can consult for tool use. Currently, only servers that are launched by executing a program that communicate over stdio are supported. For servers that use the SSE transport, consider an adapter like [mcp-proxy](https://github.com/sparfenyuk/mcp-proxy).
-
-**Note:** Codex may cache the list of tools and resources from an MCP server so that Codex can include this information in context at startup without spawning all the servers. This is designed to save resources by loading MCP servers lazily.
-
-This config option is comparable to how Claude and Cursor define `mcpServers` in their respective JSON config files, though because Codex uses TOML for its config language, the format is slightly different. For example, the following config in JSON:
-
-```json
-{
-  "mcpServers": {
-    "server-name": {
-      "command": "npx",
-      "args": ["-y", "mcp-server"],
-      "env": {
-        "API_KEY": "value"
-      }
-    }
-  }
-}
-```
-
-Should be represented as follows in `~/.codex/config.toml`:
-
-```toml
-# IMPORTANT: the top-level key is `mcp_servers` rather than `mcpServers`.
-[mcp_servers.server-name]
-command = "npx"
-args = ["-y", "mcp-server"]
-env = { "API_KEY" = "value" }
-```
-
-## disable_response_storage
-
-Currently, customers whose accounts are set to use Zero Data Retention (ZDR) must set `disable_response_storage` to `true` so that Codex uses an alternative to the Responses API that works with ZDR:
-
-```toml
-disable_response_storage = true
-```
-
-## shell_environment_policy
-
-Codex spawns subprocesses (e.g. when executing a `local_shell` tool-call suggested by the assistant). By default it now passes **your full environment** to those subprocesses. You can tune this behavior via the **`shell_environment_policy`** block in `config.toml`:
-
-```toml
-[shell_environment_policy]
-# inherit can be "all" (default), "core", or "none"
-inherit = "core"
-# set to true to *skip* the filter for `"*KEY*"` and `"*TOKEN*"`
-ignore_default_excludes = false
-# exclude patterns (case-insensitive globs)
-exclude = ["AWS_*", "AZURE_*"]
-# force-set / override values
-set = { CI = "1" }
-# if provided, *only* vars matching these patterns are kept
-include_only = ["PATH", "HOME"]
-```
-
-| Field                     | Type                       | Default | Description                                                                                                                                     |
-| ------------------------- | -------------------------- | ------- | ----------------------------------------------------------------------------------------------------------------------------------------------- |
-| `inherit`                 | string                     | `all`   | Starting template for the environment:<br>`all` (clone full parent env), `core` (`HOME`, `PATH`, `USER`, …), or `none` (start empty).           |
-| `ignore_default_excludes` | boolean                    | `false` | When `false`, Codex removes any var whose **name** contains `KEY`, `SECRET`, or `TOKEN` (case-insensitive) before other rules run.              |
-| `exclude`                 | array&lt;string&gt;        | `[]`    | Case-insensitive glob patterns to drop after the default filter.<br>Examples: `"AWS_*"`, `"AZURE_*"`.                                           |
-| `set`                     | table&lt;string,string&gt; | `{}`    | Explicit key/value overrides or additions – always win over inherited values.                                                                   |
-| `include_only`            | array&lt;string&gt;        | `[]`    | If non-empty, a whitelist of patterns; only variables that match _one_ pattern survive the final step. (Generally used with `inherit = "all"`.) |
-
-The patterns are **glob style**, not full regular expressions: `*` matches any
-number of characters, `?` matches exactly one, and character classes like
-`[A-Z]`/`[^0-9]` are supported. Matching is always **case-insensitive**. This
-syntax is documented in code as `EnvironmentVariablePattern` (see
-`core/src/config_types.rs`).
-
-If you just need a clean slate with a few custom entries you can write:
-
-```toml
-[shell_environment_policy]
-inherit = "none"
-set = { PATH = "/usr/bin", MY_FLAG = "1" }
-```
-
-Currently, `CODEX_SANDBOX_NETWORK_DISABLED=1` is also added to the environment, assuming network is disabled. This is not configurable.
-
-## notify
-
-Specify a program that will be executed to get notified about events generated by Codex. Note that the program will receive the notification argument as a string of JSON, e.g.:
-
-```json
-{
-  "type": "agent-turn-complete",
-  "turn-id": "12345",
-  "input-messages": ["Rename `foo` to `bar` and update the callsites."],
-  "last-assistant-message": "Rename complete and verified `cargo build` succeeds."
-}
-```
-
-The `"type"` property will always be set. Currently, `"agent-turn-complete"` is the only notification type that is supported.
-
-As an example, here is a Python script that parses the JSON and decides whether to show a desktop push notification using [terminal-notifier](https://github.com/julienXX/terminal-notifier) on macOS:
-
-```python
-#!/usr/bin/env python3
-
-import json
-import subprocess
-import sys
-
-
-def main() -> int:
-    if len(sys.argv) != 2:
-        print("Usage: notify.py <NOTIFICATION_JSON>")
-        return 1
-
-    try:
-        notification = json.loads(sys.argv[1])
-    except json.JSONDecodeError:
-        return 1
-
-    match notification_type := notification.get("type"):
-        case "agent-turn-complete":
-            assistant_message = notification.get("last-assistant-message")
-            if assistant_message:
-                title = f"Codex: {assistant_message}"
-            else:
-                title = "Codex: Turn Complete!"
-            input_messages = notification.get("input_messages", [])
-            message = " ".join(input_messages)
-            title += message
-        case _:
-            print(f"not sending a push notification for: {notification_type}")
-            return 0
-
-    subprocess.check_output(
-        [
-            "terminal-notifier",
-            "-title",
-            title,
-            "-message",
-            message,
-            "-group",
-            "codex",
-            "-ignoreDnD",
-            "-activate",
-            "com.googlecode.iterm2",
-        ]
-    )
-
-    return 0
-
-
-if __name__ == "__main__":
-    sys.exit(main())
-```
-
-To have Codex use this script for notifications, you would configure it via `notify` in `~/.codex/config.toml` using the appropriate path to `notify.py` on your computer:
-
-```toml
-notify = ["python3", "/Users/mbolin/.codex/notify.py"]
-```
-
-## history
-
-By default, Codex CLI records messages sent to the model in `$CODEX_HOME/history.jsonl`. Note that on UNIX, the file permissions are set to `o600`, so it should only be readable and writable by the owner.
-
-To disable this behavior, configure `[history]` as follows:
-
-```toml
-[history]
-persistence = "none"  # "save-all" is the default value
-```
-
-## file_opener
-
-Identifies the editor/URI scheme to use for hyperlinking citations in model output. If set, citations to files in the model output will be hyperlinked using the specified URI scheme so they can be ctrl/cmd-clicked from the terminal to open them.
-
-For example, if the model output includes a reference such as `【F:/home/user/project/main.py†L42-L50】`, then this would be rewritten to link to the URI `vscode://file/home/user/project/main.py:42`.
-
-Note this is **not** a general editor setting (like `$EDITOR`), as it only accepts a fixed set of values:
-
-- `"vscode"` (default)
-- `"vscode-insiders"`
-- `"windsurf"`
-- `"cursor"`
-- `"none"` to explicitly disable this feature
-
-Currently, `"vscode"` is the default, though Codex does not verify VS Code is installed. As such, `file_opener` may default to `"none"` or something else in the future.
-
-## hide_agent_reasoning
-
-Codex intermittently emits "reasoning" events that show the model's internal "thinking" before it produces a final answer. Some users may find these events distracting, especially in CI logs or minimal terminal output.
-
-Setting `hide_agent_reasoning` to `true` suppresses these events in **both** the TUI as well as the headless `exec` sub-command:
-
-```toml
-hide_agent_reasoning = true   # defaults to false
-```
-
-## show_raw_agent_reasoning
-
-Surfaces the model’s raw chain-of-thought ("raw reasoning content") when available.
-
-Notes:
-
-- Only takes effect if the selected model/provider actually emits raw reasoning content. Many models do not. When unsupported, this option has no visible effect.
-- Raw reasoning may include intermediate thoughts or sensitive context. Enable only if acceptable for your workflow.
-
-Example:
-
-```toml
-show_raw_agent_reasoning = true  # defaults to false
-```
-
-## model_context_window
-
-The size of the context window for the model, in tokens.
-
-In general, Codex knows the context window for the most common OpenAI models, but if you are using a new model with an old version of the Codex CLI, then you can use `model_context_window` to tell Codex what value to use to determine how much context is left during a conversation.
-
-## model_max_output_tokens
-
-This is analogous to `model_context_window`, but for the maximum number of output tokens for the model.
-
-## project_doc_max_bytes
-
-Maximum number of bytes to read from an `AGENTS.md` file to include in the instructions sent with the first turn of a session. Defaults to 32 KiB.
-
-## tui
-
-Options that are specific to the TUI.
-
-```toml
-[tui]
-# More to come here
-```
+- Full config docs: [docs/config.md](../docs/config.md)
+- MCP servers section: [docs/config.md#mcp_servers](../docs/config.md#mcp_servers) 

codex-rs/core/Cargo.toml

@@ -6,6 +6,7 @@ version = { workspace = true }
 [lib]
 name = "codex_core"
 path = "src/lib.rs"
+doctest = false
 
 [lints]
 workspace = true
@@ -28,6 +29,7 @@ libc = "0.2.175"
 mcp-types = { path = "../mcp-types" }
 mime_guess = "2.0"
 os_info = "3.12.0"
+portable-pty = "0.9.0"
 rand = "0.9"
 regex-lite = "0.1.6"
 reqwest = { version = "0.12", features = ["json", "stream"] }
@@ -50,12 +52,12 @@ tokio = { version = "1", features = [
 ] }
 tokio-util = "0.7.16"
 toml = "0.9.5"
-toml_edit = "0.23.3"
+toml_edit = "0.23.4"
 tracing = { version = "0.1.41", features = ["log"] }
 tree-sitter = "0.25.8"
 tree-sitter-bash = "0.25.0"
 uuid = { version = "1", features = ["serde", "v4"] }
-whoami = "1.6.0"
+whoami = "1.6.1"
 wildmatch = "2.4.0"
 
 

codex-rs/core/prompt.md

@@ -134,14 +134,6 @@ If completing the user's task requires writing or modifying files, your code and
 - Do not use one-letter variable names unless explicitly requested.
 - NEVER output inline citations like "【F:README.md†L5-L14】" in your outputs. The CLI is not able to render these so they will just be broken in the UI. Instead, if you output valid filepaths, users will be able to click on them to open the files in their editor.
 
-## Testing your work
-
-If the codebase has tests or the ability to build or run, you should use them to verify that your work is complete. Generally, your testing philosophy should be to start as specific as possible to the code you changed so that you can catch issues efficiently, then make your way to broader tests as you build confidence. If there's no test for the code you changed, and if the adjacent patterns in the codebases show that there's a logical place for you to add a test, you may do so. However, do not add tests to codebases with no tests, or where the patterns don't indicate so.
-
-Once you're confident in correctness, use formatting commands to ensure that your code is well formatted. These commands can take time so you should run them on as precise a target as possible. If there are issues you can iterate up to 3 times to get formatting right, but if you still can't manage it's better to save the user time and present them a correct solution where you call out the formatting in your final message. If the codebase does not have a formatter configured, do not add one.
-
-For all of testing, running, building, and formatting, do not attempt to fix unrelated bugs. It is not your responsibility to fix them. (You may mention them to the user in your final message though.)
-
 ## Sandbox and approvals
 
 The Codex CLI harness supports several different sandboxing, and approval configurations that the user can choose from.
@@ -177,6 +169,22 @@ Note that when sandboxing is set to read-only, you'll need to request approval f
 
 You will be told what filesystem sandboxing, network sandboxing, and approval mode are active in a developer or user message. If you are not told about this, assume that you are running with workspace-write, network sandboxing ON, and approval on-failure.
 
+## Validating your work
+
+If the codebase has tests or the ability to build or run, consider using them to verify that your work is complete. 
+
+When testing, your philosophy should be to start as specific as possible to the code you changed so that you can catch issues efficiently, then make your way to broader tests as you build confidence. If there's no test for the code you changed, and if the adjacent patterns in the codebases show that there's a logical place for you to add a test, you may do so. However, do not add tests to codebases with no tests.
+
+Similarly, once you're confident in correctness, you can suggest or use formatting commands to ensure that your code is well formatted. If there are issues you can iterate up to 3 times to get formatting right, but if you still can't manage it's better to save the user time and present them a correct solution where you call out the formatting in your final message. If the codebase does not have a formatter configured, do not add one.
+
+For all of testing, running, building, and formatting, do not attempt to fix unrelated bugs. It is not your responsibility to fix them. (You may mention them to the user in your final message though.)
+
+Be mindful of whether to run validation commands proactively. In the absence of behavioral guidance:
+
+- When running in non-interactive approval modes like **never** or **on-failure**, proactively run tests, lint and do whatever you need to ensure you've completed the task.
+- When working in interactive approval modes like **untrusted**, or **on-request**, hold off on running tests or lint commands until the user is ready for you to finalize your output, because these commands take time to run and slow down iteration. Instead suggest what you want to do next, and let the user confirm first.
+- When working on test-related tasks, such as adding tests, fixing tests, or reproducing a bug to verify behavior, you may proactively run tests regardless of approval mode. Use your judgement to decide whether this is a test-related task.
+
 ## Ambition vs. precision
 
 For tasks that have no prior context (i.e. the user is starting something brand new), you should feel free to be ambitious and demonstrate creativity with your implementation.

codex-rs/core/src/chat_completions.rs

@@ -129,7 +129,9 @@ pub(crate) async fn stream_chat_completions(
                     "content": output,
                 }));
             }
-            ResponseItem::Reasoning { .. } | ResponseItem::Other => {
+            ResponseItem::Reasoning { .. }
+            | ResponseItem::WebSearchCall { .. }
+            | ResponseItem::Other => {
                 // Omit these items from the conversation history.
                 continue;
             }
@@ -623,6 +625,9 @@ where
                 Poll::Ready(Some(Ok(ResponseEvent::ReasoningSummaryPartAdded))) => {
                     continue;
                 }
+                Poll::Ready(Some(Ok(ResponseEvent::WebSearchCallBegin { call_id }))) => {
+                    return Poll::Ready(Some(Ok(ResponseEvent::WebSearchCallBegin { call_id })));
+                }
             }
         }
     }

codex-rs/core/src/client.rs

@@ -1,6 +1,5 @@
 use std::io::BufRead;
 use std::path::Path;
-use std::sync::OnceLock;
 use std::time::Duration;
 
 use bytes::Bytes;
@@ -8,7 +7,6 @@ use codex_login::AuthManager;
 use codex_login::AuthMode;
 use eventsource_stream::Eventsource;
 use futures::prelude::*;
-use regex_lite::Regex;
 use reqwest::StatusCode;
 use serde::Deserialize;
 use serde::Serialize;
@@ -37,6 +35,7 @@ use crate::flags::CODEX_RS_SSE_FIXTURE;
 use crate::model_family::ModelFamily;
 use crate::model_provider_info::ModelProviderInfo;
 use crate::model_provider_info::WireApi;
+use crate::openai_model_info::get_model_info;
 use crate::openai_tools::create_tools_json_for_responses_api;
 use crate::protocol::TokenUsage;
 use crate::user_agent::get_codex_user_agent;
@@ -54,8 +53,11 @@ struct ErrorResponse {
 #[derive(Debug, Deserialize)]
 struct Error {
     r#type: Option<String>,
-    code: Option<String>,
     message: Option<String>,
+
+    // Optional fields available on "usage_limit_reached" and "usage_not_included" errors
+    plan_type: Option<String>,
+    resets_in_seconds: Option<u64>,
 }
 
 #[derive(Debug, Clone)]
@@ -89,6 +91,12 @@ impl ModelClient {
         }
     }
 
+    pub fn get_model_context_window(&self) -> Option<u64> {
+        self.config
+            .model_context_window
+            .or_else(|| get_model_info(&self.config.model_family).map(|info| info.context_window))
+    }
+
     /// Dispatches to either the Responses or Chat implementation depending on
     /// the provider config.  Public callers always invoke `stream()` – the
     /// specialised helpers are private to avoid accidental misuse.
@@ -142,9 +150,12 @@ impl ModelClient {
         }
 
         let auth_manager = self.auth_manager.clone();
-        let auth = auth_manager.as_ref().and_then(|m| m.auth());
 
-        let auth_mode = auth.as_ref().map(|a| a.mode);
+        let auth_mode = auth_manager
+            .as_ref()
+            .and_then(|m| m.auth())
+            .as_ref()
+            .map(|a| a.mode);
 
         let store = prompt.store && auth_mode != Some(AuthMode::ChatGPT);
 
@@ -197,15 +208,18 @@ impl ModelClient {
         let mut attempt = 0;
         let max_retries = self.provider.request_max_retries();
 
-        trace!(
-            "POST to {}: {}",
-            self.provider.get_full_url(&auth),
-            serde_json::to_string(&payload)?
-        );
-
         loop {
             attempt += 1;
 
+            // Always fetch the latest auth in case a prior attempt refreshed the token.
+            let auth = auth_manager.as_ref().and_then(|m| m.auth());
+
+            trace!(
+                "POST to {}: {}",
+                self.provider.get_full_url(&auth),
+                serde_json::to_string(&payload)?
+            );
+
             let mut req_builder = self
                 .provider
                 .create_request_builder(&self.client, &auth)
@@ -289,19 +303,20 @@ impl ModelClient {
 
                     if status == StatusCode::TOO_MANY_REQUESTS {
                         let body = res.json::<ErrorResponse>().await.ok();
-                        if let Some(ErrorResponse {
-                            error:
-                                Error {
-                                    r#type: Some(error_type),
-                                    ..
-                                },
-                        }) = body
-                        {
-                            if error_type == "usage_limit_reached" {
+                        if let Some(ErrorResponse { error }) = body {
+                            if error.r#type.as_deref() == Some("usage_limit_reached") {
+                                // Prefer the plan_type provided in the error message if present
+                                // because it's more up to date than the one encoded in the auth
+                                // token.
+                                let plan_type = error
+                                    .plan_type
+                                    .or_else(|| auth.and_then(|a| a.get_plan_type()));
+                                let resets_in_seconds = error.resets_in_seconds;
                                 return Err(CodexErr::UsageLimitReached(UsageLimitReachedError {
-                                    plan_type: auth.and_then(|a| a.get_plan_type()),
+                                    plan_type,
+                                    resets_in_seconds,
                                 }));
-                            } else if error_type == "usage_not_included" {
+                            } else if error.r#type.as_deref() == Some("usage_not_included") {
                                 return Err(CodexErr::UsageNotIncluded);
                             }
                         }
@@ -466,7 +481,8 @@ async fn process_sse<S>(
             }
         };
 
-        trace!("SSE event: {}", sse.data);
+        let raw = sse.data.clone();
+        trace!("SSE event: {}", raw);
 
         let event: SseEvent = match serde_json::from_str(&sse.data) {
             Ok(event) => event,
@@ -548,9 +564,8 @@ 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));
+                                response_error = Some(CodexErr::Stream(message, None));
                             }
                             Err(e) => {
                                 debug!("failed to parse ErrorResponse: {e}");
@@ -578,10 +593,24 @@ async fn process_sse<S>(
             | "response.custom_tool_call_input.delta"
             | "response.custom_tool_call_input.done" // also emitted as response.output_item.done
             | "response.in_progress"
-            | "response.output_item.added"
-            | "response.output_text.done" => {
-                // Currently, we ignore this event, but we handle it
-                // separately to skip the logging message in the `other` case.
+            | "response.output_text.done" => {}
+            "response.output_item.added" => {
+                if let Some(item) = event.item.as_ref() {
+                    // Detect web_search_call begin and forward a synthetic event upstream.
+                    if let Some(ty) = item.get("type").and_then(|v| v.as_str())
+                        && ty == "web_search_call"
+                    {
+                        let call_id = item
+                            .get("id")
+                            .and_then(|v| v.as_str())
+                            .unwrap_or("")
+                            .to_string();
+                        let ev = ResponseEvent::WebSearchCallBegin { call_id };
+                        if tx_event.send(Ok(ev)).await.is_err() {
+                            return;
+                        }
+                    }
+                }
             }
             "response.reasoning_summary_part.added" => {
                 // Boundary between reasoning summary sections (e.g., titles).
@@ -591,7 +620,7 @@ async fn process_sse<S>(
                 }
             }
             "response.reasoning_summary_text.done" => {}
-            other => debug!(other, "sse event"),
+            _ => {}
         }
     }
 }
@@ -622,40 +651,6 @@ async fn stream_from_fixture(
     Ok(ResponseStream { rx_event })
 }
 
-fn rate_limit_regex() -> &'static Regex {
-    static RE: OnceLock<Regex> = OnceLock::new();
-
-    #[expect(clippy::unwrap_used)]
-    RE.get_or_init(|| Regex::new(r"Please try again in (\d+(?:\.\d+)?)(s|ms)").unwrap())
-}
-
-fn try_parse_retry_after(err: &Error) -> Option<Duration> {
-    if err.code != Some("rate_limit_exceeded".to_string()) {
-        return None;
-    }
-
-    // parse the Please try again in 1.898s format using regex
-    let re = rate_limit_regex();
-    if let Some(message) = &err.message
-        && let Some(captures) = re.captures(message)
-    {
-        let seconds = captures.get(1);
-        let unit = captures.get(2);
-
-        if let (Some(value), Some(unit)) = (seconds, unit) {
-            let value = value.as_str().parse::<f64>().ok()?;
-            let unit = unit.as_str();
-
-            if unit == "s" {
-                return Some(Duration::from_secs_f64(value));
-            } else if unit == "ms" {
-                return Some(Duration::from_millis(value as u64));
-            }
-        }
-    }
-    None
-}
-
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -876,7 +871,7 @@ mod tests {
                     msg,
                     "Rate limit reached for gpt-5 in organization org-AAA on tokens per min (TPM): Limit 30000, Used 22999, Requested 12528. Please try again in 11.054s. Visit https://platform.openai.com/account/rate-limits to learn more."
                 );
-                assert_eq!(*delay, Some(Duration::from_secs_f64(11.054)));
+                assert_eq!(*delay, None);
             }
             other => panic!("unexpected second event: {other:?}"),
         }
@@ -980,27 +975,4 @@ mod tests {
             );
         }
     }
-
-    #[test]
-    fn test_try_parse_retry_after() {
-        let err = Error {
-            r#type: None,
-            message: Some("Rate limit reached for gpt-5 in organization org- on tokens per min (TPM): Limit 1, Used 1, Requested 19304. Please try again in 28ms. Visit https://platform.openai.com/account/rate-limits to learn more.".to_string()),
-            code: Some("rate_limit_exceeded".to_string()),
-        };
-
-        let delay = try_parse_retry_after(&err);
-        assert_eq!(delay, Some(Duration::from_millis(28)));
-    }
-
-    #[test]
-    fn test_try_parse_retry_after_no_delay() {
-        let err = Error {
-            r#type: None,
-            message: Some("Rate limit reached for gpt-5 in organization <ORG> on tokens per min (TPM): Limit 30000, Used 6899, Requested 24050. Please try again in 1.898s. Visit https://platform.openai.com/account/rate-limits to learn more.".to_string()),
-            code: Some("rate_limit_exceeded".to_string()),
-        };
-        let delay = try_parse_retry_after(&err);
-        assert_eq!(delay, Some(Duration::from_secs_f64(1.898)));
-    }
 }

codex-rs/core/src/client_common.rs

@@ -93,6 +93,9 @@ pub enum ResponseEvent {
     ReasoningSummaryDelta(String),
     ReasoningContentDelta(String),
     ReasoningSummaryPartAdded,
+    WebSearchCallBegin {
+        call_id: String,
+    },
 }
 
 #[derive(Debug, Serialize)]

codex-rs/core/src/codex.rs

@@ -15,6 +15,7 @@ use codex_apply_patch::MaybeApplyPatchVerified;
 use codex_apply_patch::maybe_parse_apply_patch_verified;
 use codex_login::AuthManager;
 use codex_protocol::protocol::ConversationHistoryResponseEvent;
+use codex_protocol::protocol::TaskStartedEvent;
 use codex_protocol::protocol::TurnAbortReason;
 use codex_protocol::protocol::TurnAbortedEvent;
 use futures::prelude::*;
@@ -53,12 +54,19 @@ use crate::exec::SandboxType;
 use crate::exec::StdoutStream;
 use crate::exec::StreamOutput;
 use crate::exec::process_exec_tool_call;
+use crate::exec_command::EXEC_COMMAND_TOOL_NAME;
+use crate::exec_command::ExecCommandParams;
+use crate::exec_command::ExecSessionManager;
+use crate::exec_command::WRITE_STDIN_TOOL_NAME;
+use crate::exec_command::WriteStdinParams;
 use crate::exec_env::create_env;
 use crate::mcp_connection_manager::McpConnectionManager;
 use crate::mcp_tool_call::handle_mcp_tool_call;
 use crate::model_family::find_family_for_model;
+use crate::openai_model_info::get_model_info;
 use crate::openai_tools::ApplyPatchToolArgs;
 use crate::openai_tools::ToolsConfig;
+use crate::openai_tools::ToolsConfigParams;
 use crate::openai_tools::get_openai_tools;
 use crate::parse_command::parse_command;
 use crate::plan_tool::handle_update_plan;
@@ -81,6 +89,7 @@ use crate::protocol::ExecCommandBeginEvent;
 use crate::protocol::ExecCommandEndEvent;
 use crate::protocol::FileChange;
 use crate::protocol::InputItem;
+use crate::protocol::ListCustomPromptsResponseEvent;
 use crate::protocol::Op;
 use crate::protocol::PatchApplyBeginEvent;
 use crate::protocol::PatchApplyEndEvent;
@@ -91,6 +100,8 @@ use crate::protocol::StreamErrorEvent;
 use crate::protocol::Submission;
 use crate::protocol::TaskCompleteEvent;
 use crate::protocol::TurnDiffEvent;
+use crate::protocol::WebSearchBeginEvent;
+use crate::protocol::WebSearchEndEvent;
 use crate::rollout::RolloutRecorder;
 use crate::safety::SafetyCheck;
 use crate::safety::assess_command_safety;
@@ -101,6 +112,7 @@ use crate::user_notification::UserNotification;
 use crate::util::backoff;
 use codex_protocol::config_types::ReasoningEffort as ReasoningEffortConfig;
 use codex_protocol::config_types::ReasoningSummary as ReasoningSummaryConfig;
+use codex_protocol::custom_prompts::CustomPrompt;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::LocalShellAction;
@@ -109,6 +121,7 @@ use codex_protocol::models::ReasoningItemReasoningSummary;
 use codex_protocol::models::ResponseInputItem;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::models::ShellToolCallParams;
+use codex_protocol::models::WebSearchAction;
 
 // A convenience extension trait for acquiring mutex locks where poisoning is
 // unrecoverable and should abort the program. This avoids scattered `.unwrap()`
@@ -142,6 +155,14 @@ pub struct CodexSpawnOk {
 }
 
 pub(crate) const INITIAL_SUBMIT_ID: &str = "";
+pub(crate) const SUBMISSION_CHANNEL_CAPACITY: usize = 64;
+
+// Model-formatting limits: clients get full streams; oonly content sent to the model is truncated.
+pub(crate) const MODEL_FORMAT_MAX_BYTES: usize = 10 * 1024; // 10 KiB
+pub(crate) const MODEL_FORMAT_MAX_LINES: usize = 256; // lines
+pub(crate) const MODEL_FORMAT_HEAD_LINES: usize = MODEL_FORMAT_MAX_LINES / 2;
+pub(crate) const MODEL_FORMAT_TAIL_LINES: usize = MODEL_FORMAT_MAX_LINES - MODEL_FORMAT_HEAD_LINES; // 128
+pub(crate) const MODEL_FORMAT_HEAD_BYTES: usize = MODEL_FORMAT_MAX_BYTES / 2;
 
 impl Codex {
     /// Spawn a new [`Codex`] and initialize the session.
@@ -150,7 +171,7 @@ impl Codex {
         auth_manager: Arc<AuthManager>,
         initial_history: Option<Vec<ResponseItem>>,
     ) -> CodexResult<CodexSpawnOk> {
-        let (tx_sub, rx_sub) = async_channel::bounded(64);
+        let (tx_sub, rx_sub) = async_channel::bounded(SUBMISSION_CHANNEL_CAPACITY);
         let (tx_event, rx_event) = async_channel::unbounded();
 
         let user_instructions = get_user_instructions(&config).await;
@@ -254,6 +275,7 @@ pub(crate) struct Session {
 
     /// Manager for external MCP servers/tools.
     mcp_connection_manager: McpConnectionManager,
+    session_manager: ExecSessionManager,
 
     /// External notifier command (will be passed as args to exec()). When
     /// `None` this feature is disabled.
@@ -492,13 +514,16 @@ impl Session {
         );
         let turn_context = TurnContext {
             client,
-            tools_config: ToolsConfig::new(
-                &config.model_family,
+            tools_config: ToolsConfig::new(&ToolsConfigParams {
+                model_family: &config.model_family,
                 approval_policy,
-                sandbox_policy.clone(),
-                config.include_plan_tool,
-                config.include_apply_patch_tool,
-            ),
+                sandbox_policy: sandbox_policy.clone(),
+                include_plan_tool: config.include_plan_tool,
+                include_apply_patch_tool: config.include_apply_patch_tool,
+                include_web_search_request: config.tools_web_search_request,
+                use_streamable_shell_tool: config.use_experimental_streamable_shell_tool,
+                include_view_image_tool: config.include_view_image_tool,
+            }),
             user_instructions,
             base_instructions,
             approval_policy,
@@ -511,6 +536,7 @@ impl Session {
             session_id,
             tx_event: tx_event.clone(),
             mcp_connection_manager,
+            session_manager: ExecSessionManager::default(),
             notify,
             state: Mutex::new(state),
             rollout: Mutex::new(rollout_recorder),
@@ -631,9 +657,17 @@ impl Session {
     }
 
     pub fn notify_approval(&self, sub_id: &str, decision: ReviewDecision) {
-        let mut state = self.state.lock_unchecked();
-        if let Some(tx_approve) = state.pending_approvals.remove(sub_id) {
-            tx_approve.send(decision).ok();
+        let entry = {
+            let mut state = self.state.lock_unchecked();
+            state.pending_approvals.remove(sub_id)
+        };
+        match entry {
+            Some(tx_approve) => {
+                tx_approve.send(decision).ok();
+            }
+            None => {
+                warn!("No pending approval found for sub_id: {sub_id}");
+            }
         }
     }
 
@@ -722,15 +756,15 @@ impl Session {
         let ExecToolCallOutput {
             stdout,
             stderr,
+            aggregated_output,
             duration,
             exit_code,
         } = output;
-        // Because stdout and stderr could each be up to 100 KiB, we send
-        // truncated versions.
-        const MAX_STREAM_OUTPUT: usize = 5 * 1024; // 5KiB
-        let stdout = stdout.text.chars().take(MAX_STREAM_OUTPUT).collect();
-        let stderr = stderr.text.chars().take(MAX_STREAM_OUTPUT).collect();
+        // Send full stdout/stderr to clients; do not truncate.
+        let stdout = stdout.text.clone();
+        let stderr = stderr.text.clone();
         let formatted_output = format_exec_output_str(output);
+        let aggregated_output: String = aggregated_output.text.clone();
 
         let msg = if is_apply_patch {
             EventMsg::PatchApplyEnd(PatchApplyEndEvent {
@@ -744,9 +778,10 @@ impl Session {
                 call_id: call_id.to_string(),
                 stdout,
                 stderr,
-                formatted_output,
-                duration: *duration,
+                aggregated_output,
                 exit_code: *exit_code,
+                duration: *duration,
+                formatted_output,
             })
         };
 
@@ -804,6 +839,7 @@ impl Session {
                     exit_code: -1,
                     stdout: StreamOutput::new(String::new()),
                     stderr: StreamOutput::new(get_error_message_ui(e)),
+                    aggregated_output: StreamOutput::new(get_error_message_ui(e)),
                     duration: Duration::default(),
                 };
                 &output_stderr
@@ -1058,6 +1094,9 @@ async fn submission_loop(
                 let mut updated_config = (*config).clone();
                 updated_config.model = effective_model.clone();
                 updated_config.model_family = effective_family.clone();
+                if let Some(model_info) = get_model_info(&effective_family) {
+                    updated_config.model_context_window = Some(model_info.context_window);
+                }
 
                 let client = ModelClient::new(
                     Arc::new(updated_config),
@@ -1074,13 +1113,16 @@ async fn submission_loop(
                     .unwrap_or(prev.sandbox_policy.clone());
                 let new_cwd = cwd.clone().unwrap_or_else(|| prev.cwd.clone());
 
-                let tools_config = ToolsConfig::new(
-                    &effective_family,
-                    new_approval_policy,
-                    new_sandbox_policy.clone(),
-                    config.include_plan_tool,
-                    config.include_apply_patch_tool,
-                );
+                let tools_config = ToolsConfig::new(&ToolsConfigParams {
+                    model_family: &effective_family,
+                    approval_policy: new_approval_policy,
+                    sandbox_policy: new_sandbox_policy.clone(),
+                    include_plan_tool: config.include_plan_tool,
+                    include_apply_patch_tool: config.include_apply_patch_tool,
+                    include_web_search_request: config.tools_web_search_request,
+                    use_streamable_shell_tool: config.use_experimental_streamable_shell_tool,
+                    include_view_image_tool: config.include_view_image_tool,
+                });
 
                 let new_turn_context = TurnContext {
                     client,
@@ -1129,6 +1171,7 @@ async fn submission_loop(
                 if let Err(items) = sess.inject_input(items) {
                     // Derive a fresh TurnContext for this turn using the provided overrides.
                     let provider = turn_context.client.get_provider();
+                    let auth_manager = turn_context.client.get_auth_manager();
 
                     // Derive a model family for the requested model; fall back to the session's.
                     let model_family = find_family_for_model(&model)
@@ -1138,12 +1181,15 @@ async fn submission_loop(
                     let mut per_turn_config = (*config).clone();
                     per_turn_config.model = model.clone();
                     per_turn_config.model_family = model_family.clone();
+                    if let Some(model_info) = get_model_info(&model_family) {
+                        per_turn_config.model_context_window = Some(model_info.context_window);
+                    }
 
                     // Build a new client with per‑turn reasoning settings.
                     // Reuse the same provider and session id; auth defaults to env/API key.
                     let client = ModelClient::new(
                         Arc::new(per_turn_config),
-                        None,
+                        auth_manager,
                         provider,
                         effort,
                         summary,
@@ -1152,13 +1198,17 @@ async fn submission_loop(
 
                     let fresh_turn_context = TurnContext {
                         client,
-                        tools_config: ToolsConfig::new(
-                            &model_family,
+                        tools_config: ToolsConfig::new(&ToolsConfigParams {
+                            model_family: &model_family,
                             approval_policy,
-                            sandbox_policy.clone(),
-                            config.include_plan_tool,
-                            config.include_apply_patch_tool,
-                        ),
+                            sandbox_policy: sandbox_policy.clone(),
+                            include_plan_tool: config.include_plan_tool,
+                            include_apply_patch_tool: config.include_apply_patch_tool,
+                            include_web_search_request: config.tools_web_search_request,
+                            use_streamable_shell_tool: config
+                                .use_experimental_streamable_shell_tool,
+                            include_view_image_tool: config.include_view_image_tool,
+                        }),
                         user_instructions: turn_context.user_instructions.clone(),
                         base_instructions: turn_context.base_instructions.clone(),
                         approval_policy,
@@ -1248,6 +1298,27 @@ async fn submission_loop(
                     warn!("failed to send McpListToolsResponse event: {e}");
                 }
             }
+            Op::ListCustomPrompts => {
+                let tx_event = sess.tx_event.clone();
+                let sub_id = sub.id.clone();
+
+                let custom_prompts: Vec<CustomPrompt> =
+                    if let Some(dir) = crate::custom_prompts::default_prompts_dir() {
+                        crate::custom_prompts::discover_prompts_in(&dir).await
+                    } else {
+                        Vec::new()
+                    };
+
+                let event = Event {
+                    id: sub_id,
+                    msg: EventMsg::ListCustomPromptsResponse(ListCustomPromptsResponseEvent {
+                        custom_prompts,
+                    }),
+                };
+                if let Err(e) = tx_event.send(event).await {
+                    warn!("failed to send ListCustomPromptsResponse event: {e}");
+                }
+            }
             Op::Compact => {
                 // Create a summarization request as user input
                 const SUMMARIZATION_PROMPT: &str = include_str!("prompt_for_compact_command.md");
@@ -1343,7 +1414,9 @@ async fn run_task(
     }
     let event = Event {
         id: sub_id.clone(),
-        msg: EventMsg::TaskStarted,
+        msg: EventMsg::TaskStarted(TaskStartedEvent {
+            model_context_window: turn_context.client.get_model_context_window(),
+        }),
     };
     if sess.tx_event.send(event).await.is_err() {
         return;
@@ -1669,6 +1742,7 @@ async fn try_run_turn(
     let mut stream = turn_context.client.clone().stream(&prompt).await?;
 
     let mut output = Vec::new();
+
     loop {
         // Poll the next item from the model stream. We must inspect *both* Ok and Err
         // cases so that transient stream failures (e.g., dropped SSE connection before
@@ -1705,6 +1779,15 @@ async fn try_run_turn(
                 .await?;
                 output.push(ProcessedResponseItem { item, response });
             }
+            ResponseEvent::WebSearchCallBegin { call_id } => {
+                let _ = sess
+                    .tx_event
+                    .send(Event {
+                        id: sub_id.to_string(),
+                        msg: EventMsg::WebSearchBegin(WebSearchBeginEvent { call_id }),
+                    })
+                    .await;
+            }
             ResponseEvent::Completed {
                 response_id: _,
                 token_usage,
@@ -1732,11 +1815,6 @@ async fn try_run_turn(
                 return Ok(output);
             }
             ResponseEvent::OutputTextDelta(delta) => {
-                {
-                    let mut st = sess.state.lock_unchecked();
-                    st.history.append_assistant_text(&delta);
-                }
-
                 let event = Event {
                     id: sub_id.to_string(),
                     msg: EventMsg::AgentMessageDelta(AgentMessageDeltaEvent { delta }),
@@ -1779,9 +1857,12 @@ async fn run_compact_task(
     input: Vec<InputItem>,
     compact_instructions: String,
 ) {
+    let model_context_window = turn_context.client.get_model_context_window();
     let start_event = Event {
         id: sub_id.clone(),
-        msg: EventMsg::TaskStarted,
+        msg: EventMsg::TaskStarted(TaskStartedEvent {
+            model_context_window,
+        }),
     };
     if sess.tx_event.send(start_event).await.is_err() {
         return;
@@ -1835,6 +1916,12 @@ async fn run_compact_task(
     }
 
     sess.remove_task(&sub_id);
+
+    {
+        let mut state = sess.state.lock_unchecked();
+        state.history.keep_last_messages(1);
+    }
+
     let event = Event {
         id: sub_id.clone(),
         msg: EventMsg::AgentMessage(AgentMessageEvent {
@@ -1849,9 +1936,6 @@ async fn run_compact_task(
         }),
     };
     sess.send_event(event).await;
-
-    let mut state = sess.state.lock_unchecked();
-    state.history.keep_last_messages(1);
 }
 
 async fn handle_response_item(
@@ -1999,6 +2083,17 @@ async fn handle_response_item(
             debug!("unexpected CustomToolCallOutput from stream");
             None
         }
+        ResponseItem::WebSearchCall { id, action, .. } => {
+            if let WebSearchAction::Search { query } = action {
+                let call_id = id.unwrap_or_else(|| "".to_string());
+                let event = Event {
+                    id: sub_id.to_string(),
+                    msg: EventMsg::WebSearchEnd(WebSearchEndEvent { call_id, query }),
+                };
+                sess.tx_event.send(event).await.ok();
+            }
+            None
+        }
         ResponseItem::Other => None,
     };
     Ok(output)
@@ -2031,6 +2126,36 @@ async fn handle_function_call(
             )
             .await
         }
+        "view_image" => {
+            #[derive(serde::Deserialize)]
+            struct SeeImageArgs {
+                path: String,
+            }
+            let args = match serde_json::from_str::<SeeImageArgs>(&arguments) {
+                Ok(a) => a,
+                Err(e) => {
+                    return ResponseInputItem::FunctionCallOutput {
+                        call_id,
+                        output: FunctionCallOutputPayload {
+                            content: format!("failed to parse function arguments: {e}"),
+                            success: Some(false),
+                        },
+                    };
+                }
+            };
+            let abs = turn_context.resolve_path(Some(args.path));
+            let output = match sess.inject_input(vec![InputItem::LocalImage { path: abs }]) {
+                Ok(()) => FunctionCallOutputPayload {
+                    content: "attached local image path".to_string(),
+                    success: Some(true),
+                },
+                Err(_) => FunctionCallOutputPayload {
+                    content: "unable to attach image (no active task)".to_string(),
+                    success: Some(false),
+                },
+            };
+            ResponseInputItem::FunctionCallOutput { call_id, output }
+        }
         "apply_patch" => {
             let args = match serde_json::from_str::<ApplyPatchToolArgs>(&arguments) {
                 Ok(a) => a,
@@ -2063,6 +2188,54 @@ async fn handle_function_call(
             .await
         }
         "update_plan" => handle_update_plan(sess, arguments, sub_id, call_id).await,
+        EXEC_COMMAND_TOOL_NAME => {
+            // TODO(mbolin): Sandbox check.
+            let exec_params = match serde_json::from_str::<ExecCommandParams>(&arguments) {
+                Ok(params) => params,
+                Err(e) => {
+                    return ResponseInputItem::FunctionCallOutput {
+                        call_id,
+                        output: FunctionCallOutputPayload {
+                            content: format!("failed to parse function arguments: {e}"),
+                            success: Some(false),
+                        },
+                    };
+                }
+            };
+            let result = sess
+                .session_manager
+                .handle_exec_command_request(exec_params)
+                .await;
+            let function_call_output = crate::exec_command::result_into_payload(result);
+            ResponseInputItem::FunctionCallOutput {
+                call_id,
+                output: function_call_output,
+            }
+        }
+        WRITE_STDIN_TOOL_NAME => {
+            let write_stdin_params = match serde_json::from_str::<WriteStdinParams>(&arguments) {
+                Ok(params) => params,
+                Err(e) => {
+                    return ResponseInputItem::FunctionCallOutput {
+                        call_id,
+                        output: FunctionCallOutputPayload {
+                            content: format!("failed to parse function arguments: {e}"),
+                            success: Some(false),
+                        },
+                    };
+                }
+            };
+            let result = sess
+                .session_manager
+                .handle_write_stdin_request(write_stdin_params)
+                .await;
+            let function_call_output: FunctionCallOutputPayload =
+                crate::exec_command::result_into_payload(result);
+            ResponseInputItem::FunctionCallOutput {
+                call_id,
+                output: function_call_output,
+            }
+        }
         _ => {
             match sess.mcp_connection_manager.parse_tool_name(&name) {
                 Some((server, tool_name)) => {
@@ -2369,11 +2542,15 @@ async fn handle_container_exec_with_params(
                 sandbox_type,
                 sandbox_policy: &turn_context.sandbox_policy,
                 codex_linux_sandbox_exe: &sess.codex_linux_sandbox_exe,
-                stdout_stream: Some(StdoutStream {
-                    sub_id: sub_id.clone(),
-                    call_id: call_id.clone(),
-                    tx_event: sess.tx_event.clone(),
-                }),
+                stdout_stream: if exec_command_context.apply_patch.is_some() {
+                    None
+                } else {
+                    Some(StdoutStream {
+                        sub_id: sub_id.clone(),
+                        call_id: call_id.clone(),
+                        tx_event: sess.tx_event.clone(),
+                    })
+                },
             },
         )
         .await;
@@ -2502,11 +2679,15 @@ async fn handle_sandbox_error(
                         sandbox_type: SandboxType::None,
                         sandbox_policy: &turn_context.sandbox_policy,
                         codex_linux_sandbox_exe: &sess.codex_linux_sandbox_exe,
-                        stdout_stream: Some(StdoutStream {
-                            sub_id: sub_id.clone(),
-                            call_id: call_id.clone(),
-                            tx_event: sess.tx_event.clone(),
-                        }),
+                        stdout_stream: if exec_command_context.apply_patch.is_some() {
+                            None
+                        } else {
+                            Some(StdoutStream {
+                                sub_id: sub_id.clone(),
+                                call_id: call_id.clone(),
+                                tx_event: sess.tx_event.clone(),
+                            })
+                        },
                     },
                 )
                 .await;
@@ -2550,23 +2731,103 @@ async fn handle_sandbox_error(
 
 fn format_exec_output_str(exec_output: &ExecToolCallOutput) -> String {
     let ExecToolCallOutput {
-        exit_code,
-        stdout,
-        stderr,
-        ..
+        aggregated_output, ..
     } = exec_output;
 
-    let is_success = *exit_code == 0;
-    let output = if is_success { stdout } else { stderr };
+    // Head+tail truncation for the model: show the beginning and end with an elision.
+    // Clients still receive full streams; only this formatted summary is capped.
 
-    let mut formatted_output = output.text.clone();
-    if let Some(truncated_after_lines) = output.truncated_after_lines {
-        formatted_output.push_str(&format!(
-            "\n\n[Output truncated after {truncated_after_lines} lines: too many lines or bytes.]",
-        ));
+    let s = aggregated_output.text.as_str();
+    let total_lines = s.lines().count();
+    if s.len() <= MODEL_FORMAT_MAX_BYTES && total_lines <= MODEL_FORMAT_MAX_LINES {
+        return s.to_string();
     }
 
-    formatted_output
+    let lines: Vec<&str> = s.lines().collect();
+    let head_take = MODEL_FORMAT_HEAD_LINES.min(lines.len());
+    let tail_take = MODEL_FORMAT_TAIL_LINES.min(lines.len().saturating_sub(head_take));
+    let omitted = lines.len().saturating_sub(head_take + tail_take);
+
+    // Join head and tail blocks (lines() strips newlines; reinsert them)
+    let head_block = lines
+        .iter()
+        .take(head_take)
+        .cloned()
+        .collect::<Vec<_>>()
+        .join("\n");
+    let tail_block = if tail_take > 0 {
+        lines[lines.len() - tail_take..].join("\n")
+    } else {
+        String::new()
+    };
+    let marker = format!("\n[... omitted {omitted} of {total_lines} lines ...]\n\n");
+
+    // Byte budgets for head/tail around the marker
+    let mut head_budget = MODEL_FORMAT_HEAD_BYTES.min(MODEL_FORMAT_MAX_BYTES);
+    let tail_budget = MODEL_FORMAT_MAX_BYTES.saturating_sub(head_budget + marker.len());
+    if tail_budget == 0 && marker.len() >= MODEL_FORMAT_MAX_BYTES {
+        // Degenerate case: marker alone exceeds budget; return a clipped marker
+        return take_bytes_at_char_boundary(&marker, MODEL_FORMAT_MAX_BYTES).to_string();
+    }
+    if tail_budget == 0 {
+        // Make room for the marker by shrinking head
+        head_budget = MODEL_FORMAT_MAX_BYTES.saturating_sub(marker.len());
+    }
+
+    // Enforce line-count cap by trimming head/tail lines
+    let head_lines_text = head_block;
+    let tail_lines_text = tail_block;
+    // Build final string respecting byte budgets
+    let head_part = take_bytes_at_char_boundary(&head_lines_text, head_budget);
+    let mut result = String::with_capacity(MODEL_FORMAT_MAX_BYTES.min(s.len()));
+    result.push_str(head_part);
+    result.push_str(&marker);
+
+    let remaining = MODEL_FORMAT_MAX_BYTES.saturating_sub(result.len());
+    let tail_budget_final = remaining;
+    let tail_part = take_last_bytes_at_char_boundary(&tail_lines_text, tail_budget_final);
+    result.push_str(tail_part);
+
+    result
+}
+
+// Truncate a &str to a byte budget at a char boundary (prefix)
+#[inline]
+fn take_bytes_at_char_boundary(s: &str, maxb: usize) -> &str {
+    if s.len() <= maxb {
+        return s;
+    }
+    let mut last_ok = 0;
+    for (i, ch) in s.char_indices() {
+        let nb = i + ch.len_utf8();
+        if nb > maxb {
+            break;
+        }
+        last_ok = nb;
+    }
+    &s[..last_ok]
+}
+
+// Take a suffix of a &str within a byte budget at a char boundary
+#[inline]
+fn take_last_bytes_at_char_boundary(s: &str, maxb: usize) -> &str {
+    if s.len() <= maxb {
+        return s;
+    }
+    let mut start = s.len();
+    let mut used = 0usize;
+    for (i, ch) in s.char_indices().rev() {
+        let nb = ch.len_utf8();
+        if used + nb > maxb {
+            break;
+        }
+        start = i;
+        used += nb;
+        if start == 0 {
+            break;
+        }
+    }
+    &s[start..]
 }
 
 /// Exec output is a pre-serialized JSON payload
@@ -2651,15 +2912,9 @@ async fn drain_to_completed(
                 response_id: _,
                 token_usage,
             }) => {
-                let token_usage = match token_usage {
-                    Some(usage) => usage,
-                    None => {
-                        return Err(CodexErr::Stream(
-                            "token_usage was None in ResponseEvent::Completed".into(),
-                            None,
-                        ));
-                    }
-                };
+                // some providers don't return token usage, so we default
+                // TODO: consider approximate token usage
+                let token_usage = token_usage.unwrap_or_default();
                 sess.tx_event
                     .send(Event {
                         id: sub_id.to_string(),
@@ -2667,6 +2922,7 @@ async fn drain_to_completed(
                     })
                     .await
                     .ok();
+
                 return Ok(());
             }
             Ok(_) => continue,
@@ -2717,6 +2973,7 @@ mod tests {
     use mcp_types::TextContent;
     use pretty_assertions::assert_eq;
     use serde_json::json;
+    use std::time::Duration as StdDuration;
 
     fn text_block(s: &str) -> ContentBlock {
         ContentBlock::TextContent(TextContent {
@@ -2751,6 +3008,82 @@ mod tests {
         assert_eq!(expected, got);
     }
 
+    #[test]
+    fn model_truncation_head_tail_by_lines() {
+        // Build 400 short lines so line-count limit, not byte budget, triggers truncation
+        let lines: Vec<String> = (1..=400).map(|i| format!("line{i}")).collect();
+        let full = lines.join("\n");
+
+        let exec = ExecToolCallOutput {
+            exit_code: 0,
+            stdout: StreamOutput::new(String::new()),
+            stderr: StreamOutput::new(String::new()),
+            aggregated_output: StreamOutput::new(full.clone()),
+            duration: StdDuration::from_secs(1),
+        };
+
+        let out = format_exec_output_str(&exec);
+
+        // Expect elision marker with correct counts
+        let omitted = 400 - MODEL_FORMAT_MAX_LINES; // 144
+        let marker = format!("\n[... omitted {omitted} of 400 lines ...]\n\n");
+        assert!(out.contains(&marker), "missing marker: {out}");
+
+        // Validate head and tail
+        let parts: Vec<&str> = out.split(&marker).collect();
+        assert_eq!(parts.len(), 2, "expected one marker split");
+        let head = parts[0];
+        let tail = parts[1];
+
+        let expected_head: String = (1..=MODEL_FORMAT_HEAD_LINES)
+            .map(|i| format!("line{i}"))
+            .collect::<Vec<_>>()
+            .join("\n");
+        assert!(head.starts_with(&expected_head), "head mismatch");
+
+        let expected_tail: String = ((400 - MODEL_FORMAT_TAIL_LINES + 1)..=400)
+            .map(|i| format!("line{i}"))
+            .collect::<Vec<_>>()
+            .join("\n");
+        assert!(tail.ends_with(&expected_tail), "tail mismatch");
+    }
+
+    #[test]
+    fn model_truncation_respects_byte_budget() {
+        // Construct a large output (about 100kB) so byte budget dominates
+        let big_line = "x".repeat(100);
+        let full = std::iter::repeat_n(big_line.clone(), 1000)
+            .collect::<Vec<_>>()
+            .join("\n");
+
+        let exec = ExecToolCallOutput {
+            exit_code: 0,
+            stdout: StreamOutput::new(String::new()),
+            stderr: StreamOutput::new(String::new()),
+            aggregated_output: StreamOutput::new(full.clone()),
+            duration: StdDuration::from_secs(1),
+        };
+
+        let out = format_exec_output_str(&exec);
+        assert!(out.len() <= MODEL_FORMAT_MAX_BYTES, "exceeds byte budget");
+        assert!(out.contains("omitted"), "should contain elision marker");
+
+        // Ensure head and tail are drawn from the original
+        assert!(full.starts_with(out.chars().take(8).collect::<String>().as_str()));
+        assert!(
+            full.ends_with(
+                out.chars()
+                    .rev()
+                    .take(8)
+                    .collect::<String>()
+                    .chars()
+                    .rev()
+                    .collect::<String>()
+                    .as_str()
+            )
+        );
+    }
+
     #[test]
     fn falls_back_to_content_when_structured_is_null() {
         let ctr = CallToolResult {

codex-rs/core/src/config.rs

@@ -169,11 +169,22 @@ pub struct Config {
     /// model family's default preference.
     pub include_apply_patch_tool: bool,
 
+    pub tools_web_search_request: bool,
+
     /// The value for the `originator` header included with Responses API requests.
     pub responses_originator_header: String,
 
     /// If set to `true`, the API key will be signed with the `originator` header.
     pub preferred_auth_method: AuthMode,
+
+    pub use_experimental_streamable_shell_tool: bool,
+
+    /// Include the `view_image` tool that lets the agent attach a local image path to context.
+    pub include_view_image_tool: 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.
+    pub disable_paste_burst: bool,
 }
 
 impl Config {
@@ -469,6 +480,8 @@ pub struct ConfigToml {
     /// Experimental path to a file whose contents replace the built-in BASE_INSTRUCTIONS.
     pub experimental_instructions_file: Option<PathBuf>,
 
+    pub experimental_use_exec_command_tool: Option<bool>,
+
     /// The value for the `originator` header included with Responses API requests.
     pub responses_originator_header_internal_override: Option<String>,
 
@@ -476,6 +489,14 @@ pub struct ConfigToml {
 
     /// If set to `true`, the API key will be signed with the `originator` header.
     pub preferred_auth_method: Option<AuthMode>,
+
+    /// Nested tools section for feature toggles
+    pub tools: Option<ToolsToml>,
+
+    /// 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.
+    pub disable_paste_burst: Option<bool>,
 }
 
 #[derive(Deserialize, Debug, Clone, PartialEq, Eq)]
@@ -483,6 +504,16 @@ pub struct ProjectConfig {
     pub trust_level: Option<String>,
 }
 
+#[derive(Deserialize, Debug, Clone, Default)]
+pub struct ToolsToml {
+    #[serde(default, alias = "web_search_request")]
+    pub web_search: Option<bool>,
+
+    /// Enable the `view_image` tool that lets the agent attach local images.
+    #[serde(default)]
+    pub view_image: Option<bool>,
+}
+
 impl ConfigToml {
     /// Derive the effective sandbox policy from the configuration.
     fn derive_sandbox_policy(&self, sandbox_mode_override: Option<SandboxMode>) -> SandboxPolicy {
@@ -570,8 +601,10 @@ pub struct ConfigOverrides {
     pub base_instructions: Option<String>,
     pub include_plan_tool: Option<bool>,
     pub include_apply_patch_tool: Option<bool>,
+    pub include_view_image_tool: Option<bool>,
     pub disable_response_storage: Option<bool>,
     pub show_raw_agent_reasoning: Option<bool>,
+    pub tools_web_search_request: Option<bool>,
 }
 
 impl Config {
@@ -596,8 +629,10 @@ impl Config {
             base_instructions,
             include_plan_tool,
             include_apply_patch_tool,
+            include_view_image_tool,
             disable_response_storage,
             show_raw_agent_reasoning,
+            tools_web_search_request: override_tools_web_search_request,
         } = overrides;
 
         let config_profile = match config_profile_key.as_ref().or(cfg.profile.as_ref()) {
@@ -659,6 +694,14 @@ impl Config {
 
         let history = cfg.history.unwrap_or_default();
 
+        let tools_web_search_request = override_tools_web_search_request
+            .or(cfg.tools.as_ref().and_then(|t| t.web_search))
+            .unwrap_or(false);
+
+        let include_view_image_tool = include_view_image_tool
+            .or(cfg.tools.as_ref().and_then(|t| t.view_image))
+            .unwrap_or(true);
+
         let model = model
             .or(config_profile.model)
             .or(cfg.model)
@@ -756,8 +799,14 @@ impl Config {
             experimental_resume,
             include_plan_tool: include_plan_tool.unwrap_or(false),
             include_apply_patch_tool: include_apply_patch_tool.unwrap_or(false),
+            tools_web_search_request,
             responses_originator_header,
             preferred_auth_method: cfg.preferred_auth_method.unwrap_or(AuthMode::ChatGPT),
+            use_experimental_streamable_shell_tool: cfg
+                .experimental_use_exec_command_tool
+                .unwrap_or(false),
+            include_view_image_tool,
+            disable_paste_burst: cfg.disable_paste_burst.unwrap_or(false),
         };
         Ok(config)
     }
@@ -1122,8 +1171,12 @@ disable_response_storage = true
                 base_instructions: None,
                 include_plan_tool: false,
                 include_apply_patch_tool: false,
+                tools_web_search_request: false,
                 responses_originator_header: "codex_cli_rs".to_string(),
                 preferred_auth_method: AuthMode::ChatGPT,
+                use_experimental_streamable_shell_tool: false,
+                include_view_image_tool: true,
+                disable_paste_burst: false,
             },
             o3_profile_config
         );
@@ -1176,8 +1229,12 @@ disable_response_storage = true
             base_instructions: None,
             include_plan_tool: false,
             include_apply_patch_tool: false,
+            tools_web_search_request: false,
             responses_originator_header: "codex_cli_rs".to_string(),
             preferred_auth_method: AuthMode::ChatGPT,
+            use_experimental_streamable_shell_tool: false,
+            include_view_image_tool: true,
+            disable_paste_burst: false,
         };
 
         assert_eq!(expected_gpt3_profile_config, gpt3_profile_config);
@@ -1245,8 +1302,12 @@ disable_response_storage = true
             base_instructions: None,
             include_plan_tool: false,
             include_apply_patch_tool: false,
+            tools_web_search_request: false,
             responses_originator_header: "codex_cli_rs".to_string(),
             preferred_auth_method: AuthMode::ChatGPT,
+            use_experimental_streamable_shell_tool: false,
+            include_view_image_tool: true,
+            disable_paste_burst: false,
         };
 
         assert_eq!(expected_zdr_profile_config, zdr_profile_config);
@@ -1268,9 +1329,9 @@ disable_response_storage = true
 
         let raw_path = project_dir.path().to_string_lossy();
         let path_str = if raw_path.contains('\\') {
-            format!("'{}'", raw_path)
+            format!("'{raw_path}'")
         } else {
-            format!("\"{}\"", raw_path)
+            format!("\"{raw_path}\"")
         };
         let expected = format!(
             r#"[projects.{path_str}]
@@ -1291,9 +1352,9 @@ trust_level = "trusted"
         let config_path = codex_home.path().join(CONFIG_TOML_FILE);
         let raw_path = project_dir.path().to_string_lossy();
         let path_str = if raw_path.contains('\\') {
-            format!("'{}'", raw_path)
+            format!("'{raw_path}'")
         } else {
-            format!("\"{}\"", raw_path)
+            format!("\"{raw_path}\"")
         };
         // Use a quoted key so backslashes don't require escaping on Windows
         let initial = format!(

codex-rs/core/src/conversation_history.rs

@@ -28,49 +28,7 @@ impl ConversationHistory {
                 continue;
             }
 
-            // Merge adjacent assistant messages into a single history entry.
-            // This prevents duplicates when a partial assistant message was
-            // streamed into history earlier in the turn and the final full
-            // message is recorded at turn end.
-            match (&*item, self.items.last_mut()) {
-                (
-                    ResponseItem::Message {
-                        role: new_role,
-                        content: new_content,
-                        ..
-                    },
-                    Some(ResponseItem::Message {
-                        role: last_role,
-                        content: last_content,
-                        ..
-                    }),
-                ) if new_role == "assistant" && last_role == "assistant" => {
-                    append_text_content(last_content, new_content);
-                }
-                _ => {
-                    self.items.push(item.clone());
-                }
-            }
-        }
-    }
-
-    /// Append a text `delta` to the latest assistant message, creating a new
-    /// assistant entry if none exists yet (e.g. first delta for this turn).
-    pub(crate) fn append_assistant_text(&mut self, delta: &str) {
-        match self.items.last_mut() {
-            Some(ResponseItem::Message { role, content, .. }) if role == "assistant" => {
-                append_text_delta(content, delta);
-            }
-            _ => {
-                // Start a new assistant message with the delta.
-                self.items.push(ResponseItem::Message {
-                    id: None,
-                    role: "assistant".to_string(),
-                    content: vec![codex_protocol::models::ContentItem::OutputText {
-                        text: delta.to_string(),
-                    }],
-                });
-            }
+            self.items.push(item.clone());
         }
     }
 
@@ -114,35 +72,7 @@ fn is_api_message(message: &ResponseItem) -> bool {
         | ResponseItem::CustomToolCallOutput { .. }
         | ResponseItem::LocalShellCall { .. }
         | ResponseItem::Reasoning { .. } => true,
-        ResponseItem::Other => false,
-    }
-}
-
-/// Helper to append the textual content from `src` into `dst` in place.
-fn append_text_content(
-    dst: &mut Vec<codex_protocol::models::ContentItem>,
-    src: &Vec<codex_protocol::models::ContentItem>,
-) {
-    for c in src {
-        if let codex_protocol::models::ContentItem::OutputText { text } = c {
-            append_text_delta(dst, text);
-        }
-    }
-}
-
-/// Append a single text delta to the last OutputText item in `content`, or
-/// push a new OutputText item if none exists.
-fn append_text_delta(content: &mut Vec<codex_protocol::models::ContentItem>, delta: &str) {
-    if let Some(codex_protocol::models::ContentItem::OutputText { text }) = content
-        .iter_mut()
-        .rev()
-        .find(|c| matches!(c, codex_protocol::models::ContentItem::OutputText { .. }))
-    {
-        text.push_str(delta);
-    } else {
-        content.push(codex_protocol::models::ContentItem::OutputText {
-            text: delta.to_string(),
-        });
+        ResponseItem::WebSearchCall { .. } | ResponseItem::Other => false,
     }
 }
 
@@ -171,49 +101,6 @@ mod tests {
         }
     }
 
-    #[test]
-    fn merges_adjacent_assistant_messages() {
-        let mut h = ConversationHistory::default();
-        let a1 = assistant_msg("Hello");
-        let a2 = assistant_msg(", world!");
-        h.record_items([&a1, &a2]);
-
-        let items = h.contents();
-        assert_eq!(
-            items,
-            vec![ResponseItem::Message {
-                id: None,
-                role: "assistant".to_string(),
-                content: vec![ContentItem::OutputText {
-                    text: "Hello, world!".to_string()
-                }]
-            }]
-        );
-    }
-
-    #[test]
-    fn append_assistant_text_creates_and_appends() {
-        let mut h = ConversationHistory::default();
-        h.append_assistant_text("Hello");
-        h.append_assistant_text(", world");
-
-        // Now record a final full assistant message and verify it merges.
-        let final_msg = assistant_msg("!");
-        h.record_items([&final_msg]);
-
-        let items = h.contents();
-        assert_eq!(
-            items,
-            vec![ResponseItem::Message {
-                id: None,
-                role: "assistant".to_string(),
-                content: vec![ContentItem::OutputText {
-                    text: "Hello, world!".to_string()
-                }]
-            }]
-        );
-    }
-
     #[test]
     fn filters_non_api_messages() {
         let mut h = ConversationHistory::default();

codex-rs/core/src/conversation_manager.rs

@@ -110,6 +110,10 @@ impl ConversationManager {
             .ok_or_else(|| CodexErr::ConversationNotFound(conversation_id))
     }
 
+    pub async fn remove_conversation(&self, conversation_id: Uuid) {
+        self.conversations.write().await.remove(&conversation_id);
+    }
+
     /// Fork an existing conversation by dropping the last `drop_last_messages`
     /// user/assistant messages from its transcript and starting a new
     /// conversation with identical configuration (unless overridden by the

codex-rs/core/src/custom_prompts.rs

@@ -0,0 +1,127 @@
+use codex_protocol::custom_prompts::CustomPrompt;
+use std::collections::HashSet;
+use std::path::Path;
+use std::path::PathBuf;
+use tokio::fs;
+
+/// Return the default prompts directory: `$CODEX_HOME/prompts`.
+/// If `CODEX_HOME` cannot be resolved, returns `None`.
+pub fn default_prompts_dir() -> Option<PathBuf> {
+    crate::config::find_codex_home()
+        .ok()
+        .map(|home| home.join("prompts"))
+}
+
+/// Discover prompt files in the given directory, returning entries sorted by name.
+/// Non-files are ignored. If the directory does not exist or cannot be read, returns empty.
+pub async fn discover_prompts_in(dir: &Path) -> Vec<CustomPrompt> {
+    discover_prompts_in_excluding(dir, &HashSet::new()).await
+}
+
+/// Discover prompt files in the given directory, excluding any with names in `exclude`.
+/// Returns entries sorted by name. Non-files are ignored. Missing/unreadable dir yields empty.
+pub async fn discover_prompts_in_excluding(
+    dir: &Path,
+    exclude: &HashSet<String>,
+) -> Vec<CustomPrompt> {
+    let mut out: Vec<CustomPrompt> = Vec::new();
+    let mut entries = match fs::read_dir(dir).await {
+        Ok(entries) => entries,
+        Err(_) => return out,
+    };
+
+    while let Ok(Some(entry)) = entries.next_entry().await {
+        let path = entry.path();
+        let is_file = entry
+            .file_type()
+            .await
+            .map(|ft| ft.is_file())
+            .unwrap_or(false);
+        if !is_file {
+            continue;
+        }
+        // Only include Markdown files with a .md extension.
+        let is_md = path
+            .extension()
+            .and_then(|s| s.to_str())
+            .map(|ext| ext.eq_ignore_ascii_case("md"))
+            .unwrap_or(false);
+        if !is_md {
+            continue;
+        }
+        let Some(name) = path
+            .file_stem()
+            .and_then(|s| s.to_str())
+            .map(|s| s.to_string())
+        else {
+            continue;
+        };
+        if exclude.contains(&name) {
+            continue;
+        }
+        let content = match fs::read_to_string(&path).await {
+            Ok(s) => s,
+            Err(_) => continue,
+        };
+        out.push(CustomPrompt {
+            name,
+            path,
+            content,
+        });
+    }
+    out.sort_by(|a, b| a.name.cmp(&b.name));
+    out
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::fs;
+    use tempfile::tempdir;
+
+    #[tokio::test]
+    async fn empty_when_dir_missing() {
+        let tmp = tempdir().expect("create TempDir");
+        let missing = tmp.path().join("nope");
+        let found = discover_prompts_in(&missing).await;
+        assert!(found.is_empty());
+    }
+
+    #[tokio::test]
+    async fn discovers_and_sorts_files() {
+        let tmp = tempdir().expect("create TempDir");
+        let dir = tmp.path();
+        fs::write(dir.join("b.md"), b"b").unwrap();
+        fs::write(dir.join("a.md"), b"a").unwrap();
+        fs::create_dir(dir.join("subdir")).unwrap();
+        let found = discover_prompts_in(dir).await;
+        let names: Vec<String> = found.into_iter().map(|e| e.name).collect();
+        assert_eq!(names, vec!["a", "b"]);
+    }
+
+    #[tokio::test]
+    async fn excludes_builtins() {
+        let tmp = tempdir().expect("create TempDir");
+        let dir = tmp.path();
+        fs::write(dir.join("init.md"), b"ignored").unwrap();
+        fs::write(dir.join("foo.md"), b"ok").unwrap();
+        let mut exclude = HashSet::new();
+        exclude.insert("init".to_string());
+        let found = discover_prompts_in_excluding(dir, &exclude).await;
+        let names: Vec<String> = found.into_iter().map(|e| e.name).collect();
+        assert_eq!(names, vec!["foo"]);
+    }
+
+    #[tokio::test]
+    async fn skips_non_utf8_files() {
+        let tmp = tempdir().expect("create TempDir");
+        let dir = tmp.path();
+        // Valid UTF-8 file
+        fs::write(dir.join("good.md"), b"hello").unwrap();
+        // Invalid UTF-8 content in .md file (e.g., lone 0xFF byte)
+        fs::write(dir.join("bad.md"), vec![0xFF, 0xFE, b'\n']).unwrap();
+        let found = discover_prompts_in(dir).await;
+        let names: Vec<String> = found.into_iter().map(|e| e.name).collect();
+        assert_eq!(names, vec!["good"]);
+    }
+}

codex-rs/core/src/environment_context.rs

@@ -85,23 +85,21 @@ impl EnvironmentContext {
         }
         if let Some(approval_policy) = self.approval_policy {
             lines.push(format!(
-                "  <approval_policy>{}</approval_policy>",
-                approval_policy
+                "  <approval_policy>{approval_policy}</approval_policy>"
             ));
         }
         if let Some(sandbox_mode) = self.sandbox_mode {
-            lines.push(format!("  <sandbox_mode>{}</sandbox_mode>", sandbox_mode));
+            lines.push(format!("  <sandbox_mode>{sandbox_mode}</sandbox_mode>"));
         }
         if let Some(network_access) = self.network_access {
             lines.push(format!(
-                "  <network_access>{}</network_access>",
-                network_access
+                "  <network_access>{network_access}</network_access>"
             ));
         }
         if let Some(shell) = self.shell
             && let Some(shell_name) = shell.name()
         {
-            lines.push(format!("  <shell>{}</shell>", shell_name));
+            lines.push(format!("  <shell>{shell_name}</shell>"));
         }
         lines.push(ENVIRONMENT_CONTEXT_END.to_string());
         lines.join("\n")

codex-rs/core/src/error.rs

@@ -128,27 +128,70 @@ pub enum CodexErr {
 #[derive(Debug)]
 pub struct UsageLimitReachedError {
     pub plan_type: Option<String>,
+    pub resets_in_seconds: Option<u64>,
 }
 
 impl std::fmt::Display for UsageLimitReachedError {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        // Base message differs slightly for legacy ChatGPT Plus plan users.
         if let Some(plan_type) = &self.plan_type
             && plan_type == "plus"
         {
             write!(
                 f,
-                "You've hit your usage limit. Upgrade to Pro (https://openai.com/chatgpt/pricing), or wait for limits to reset (every 5h and every week.)."
+                "You've hit your usage limit. Upgrade to Pro (https://openai.com/chatgpt/pricing) or try again"
             )?;
+            if let Some(secs) = self.resets_in_seconds {
+                let reset_duration = format_reset_duration(secs);
+                write!(f, " in {reset_duration}.")?;
+            } else {
+                write!(f, " later.")?;
+            }
         } else {
-            write!(
-                f,
-                "You've hit your usage limit. Limits reset every 5h and every week."
-            )?;
+            write!(f, "You've hit your usage limit.")?;
+
+            if let Some(secs) = self.resets_in_seconds {
+                let reset_duration = format_reset_duration(secs);
+                write!(f, " Try again in {reset_duration}.")?;
+            } else {
+                write!(f, " Try again later.")?;
+            }
         }
+
         Ok(())
     }
 }
 
+fn format_reset_duration(total_secs: u64) -> String {
+    let days = total_secs / 86_400;
+    let hours = (total_secs % 86_400) / 3_600;
+    let minutes = (total_secs % 3_600) / 60;
+
+    let mut parts: Vec<String> = Vec::new();
+    if days > 0 {
+        let unit = if days == 1 { "day" } else { "days" };
+        parts.push(format!("{days} {unit}"));
+    }
+    if hours > 0 {
+        let unit = if hours == 1 { "hour" } else { "hours" };
+        parts.push(format!("{hours} {unit}"));
+    }
+    if minutes > 0 {
+        let unit = if minutes == 1 { "minute" } else { "minutes" };
+        parts.push(format!("{minutes} {unit}"));
+    }
+
+    if parts.is_empty() {
+        return "less than a minute".to_string();
+    }
+
+    match parts.len() {
+        1 => parts[0].clone(),
+        2 => format!("{} {}", parts[0], parts[1]),
+        _ => format!("{} {} {}", parts[0], parts[1], parts[2]),
+    }
+}
+
 #[derive(Debug)]
 pub struct EnvVarError {
     /// Name of the environment variable that is missing.
@@ -181,6 +224,8 @@ impl CodexErr {
 pub fn get_error_message_ui(e: &CodexErr) -> String {
     match e {
         CodexErr::Sandbox(SandboxErr::Denied(_, _, stderr)) => stderr.to_string(),
+        // Timeouts are not sandbox errors from a UX perspective; present them plainly
+        CodexErr::Sandbox(SandboxErr::Timeout) => "error: command timed out".to_string(),
         _ => e.to_string(),
     }
 }
@@ -193,19 +238,23 @@ mod tests {
     fn usage_limit_reached_error_formats_plus_plan() {
         let err = UsageLimitReachedError {
             plan_type: Some("plus".to_string()),
+            resets_in_seconds: None,
         };
         assert_eq!(
             err.to_string(),
-            "You've hit your usage limit. Upgrade to Pro (https://openai.com/chatgpt/pricing), or wait for limits to reset (every 5h and every week.)."
+            "You've hit your usage limit. Upgrade to Pro (https://openai.com/chatgpt/pricing) or try again later."
         );
     }
 
     #[test]
     fn usage_limit_reached_error_formats_default_when_none() {
-        let err = UsageLimitReachedError { plan_type: None };
+        let err = UsageLimitReachedError {
+            plan_type: None,
+            resets_in_seconds: None,
+        };
         assert_eq!(
             err.to_string(),
-            "You've hit your usage limit. Limits reset every 5h and every week."
+            "You've hit your usage limit. Try again later."
         );
     }
 
@@ -213,10 +262,59 @@ mod tests {
     fn usage_limit_reached_error_formats_default_for_other_plans() {
         let err = UsageLimitReachedError {
             plan_type: Some("pro".to_string()),
+            resets_in_seconds: None,
         };
         assert_eq!(
             err.to_string(),
-            "You've hit your usage limit. Limits reset every 5h and every week."
+            "You've hit your usage limit. Try again later."
+        );
+    }
+
+    #[test]
+    fn usage_limit_reached_includes_minutes_when_available() {
+        let err = UsageLimitReachedError {
+            plan_type: None,
+            resets_in_seconds: Some(5 * 60),
+        };
+        assert_eq!(
+            err.to_string(),
+            "You've hit your usage limit. Try again in 5 minutes."
+        );
+    }
+
+    #[test]
+    fn usage_limit_reached_includes_hours_and_minutes() {
+        let err = UsageLimitReachedError {
+            plan_type: Some("plus".to_string()),
+            resets_in_seconds: Some(3 * 3600 + 32 * 60),
+        };
+        assert_eq!(
+            err.to_string(),
+            "You've hit your usage limit. Upgrade to Pro (https://openai.com/chatgpt/pricing) or try again in 3 hours 32 minutes."
+        );
+    }
+
+    #[test]
+    fn usage_limit_reached_includes_days_hours_minutes() {
+        let err = UsageLimitReachedError {
+            plan_type: None,
+            resets_in_seconds: Some(2 * 86_400 + 3 * 3600 + 5 * 60),
+        };
+        assert_eq!(
+            err.to_string(),
+            "You've hit your usage limit. Try again in 2 days 3 hours 5 minutes."
+        );
+    }
+
+    #[test]
+    fn usage_limit_reached_less_than_minute() {
+        let err = UsageLimitReachedError {
+            plan_type: None,
+            resets_in_seconds: Some(30),
+        };
+        assert_eq!(
+            err.to_string(),
+            "You've hit your usage limit. Try again in less than a minute."
         );
     }
 }

codex-rs/core/src/exec.rs

@@ -28,18 +28,21 @@ use crate::spawn::StdioPolicy;
 use crate::spawn::spawn_child_async;
 use serde_bytes::ByteBuf;
 
-// Maximum we send for each stream, which is either:
-// - 10KiB OR
-// - 256 lines
-const MAX_STREAM_OUTPUT: usize = 10 * 1024;
-const MAX_STREAM_OUTPUT_LINES: usize = 256;
-
 const DEFAULT_TIMEOUT_MS: u64 = 10_000;
 
 // Hardcode these since it does not seem worth including the libc crate just
 // for these.
 const SIGKILL_CODE: i32 = 9;
 const TIMEOUT_CODE: i32 = 64;
+const EXIT_CODE_SIGNAL_BASE: i32 = 128; // conventional shell: 128 + signal
+
+// I/O buffer sizing
+const READ_CHUNK_SIZE: usize = 8192; // bytes per read
+const AGGREGATE_BUFFER_INITIAL_CAPACITY: usize = 8 * 1024; // 8 KiB
+
+/// Limit the number of ExecCommandOutputDelta events emitted per exec call.
+/// Aggregation still collects full output; only the live event stream is capped.
+pub(crate) const MAX_EXEC_OUTPUT_DELTAS_PER_CALL: usize = 10_000;
 
 #[derive(Debug, Clone)]
 pub struct ExecParams {
@@ -153,6 +156,7 @@ pub async fn process_exec_tool_call(
                 exit_code,
                 stdout,
                 stderr,
+                aggregated_output: raw_output.aggregated_output.from_utf8_lossy(),
                 duration,
             })
         }
@@ -189,10 +193,11 @@ pub struct StreamOutput<T> {
     pub truncated_after_lines: Option<u32>,
 }
 #[derive(Debug)]
-pub struct RawExecToolCallOutput {
+struct RawExecToolCallOutput {
     pub exit_status: ExitStatus,
     pub stdout: StreamOutput<Vec<u8>>,
     pub stderr: StreamOutput<Vec<u8>>,
+    pub aggregated_output: StreamOutput<Vec<u8>>,
 }
 
 impl StreamOutput<String> {
@@ -213,11 +218,17 @@ impl StreamOutput<Vec<u8>> {
     }
 }
 
+#[inline]
+fn append_all(dst: &mut Vec<u8>, src: &[u8]) {
+    dst.extend_from_slice(src);
+}
+
 #[derive(Debug)]
 pub struct ExecToolCallOutput {
     pub exit_code: i32,
     pub stdout: StreamOutput<String>,
     pub stderr: StreamOutput<String>,
+    pub aggregated_output: StreamOutput<String>,
     pub duration: Duration,
 }
 
@@ -253,7 +264,7 @@ async fn exec(
 
 /// Consumes the output of a child process, truncating it so it is suitable for
 /// use as the output of a `shell` tool call. Also enforces specified timeout.
-pub(crate) async fn consume_truncated_output(
+async fn consume_truncated_output(
     mut child: Child,
     timeout: Duration,
     stdout_stream: Option<StdoutStream>,
@@ -273,19 +284,19 @@ pub(crate) async fn consume_truncated_output(
         ))
     })?;
 
+    let (agg_tx, agg_rx) = async_channel::unbounded::<Vec<u8>>();
+
     let stdout_handle = tokio::spawn(read_capped(
         BufReader::new(stdout_reader),
-        MAX_STREAM_OUTPUT,
-        MAX_STREAM_OUTPUT_LINES,
         stdout_stream.clone(),
         false,
+        Some(agg_tx.clone()),
     ));
     let stderr_handle = tokio::spawn(read_capped(
         BufReader::new(stderr_reader),
-        MAX_STREAM_OUTPUT,
-        MAX_STREAM_OUTPUT_LINES,
         stdout_stream.clone(),
         true,
+        Some(agg_tx.clone()),
     ));
 
     let exit_status = tokio::select! {
@@ -297,38 +308,49 @@ pub(crate) async fn consume_truncated_output(
                     // timeout
                     child.start_kill()?;
                     // Debatable whether `child.wait().await` should be called here.
-                    synthetic_exit_status(128 + TIMEOUT_CODE)
+                    synthetic_exit_status(EXIT_CODE_SIGNAL_BASE + TIMEOUT_CODE)
                 }
             }
         }
         _ = tokio::signal::ctrl_c() => {
             child.start_kill()?;
-            synthetic_exit_status(128 + SIGKILL_CODE)
+            synthetic_exit_status(EXIT_CODE_SIGNAL_BASE + SIGKILL_CODE)
         }
     };
 
     let stdout = stdout_handle.await??;
     let stderr = stderr_handle.await??;
 
+    drop(agg_tx);
+
+    let mut combined_buf = Vec::with_capacity(AGGREGATE_BUFFER_INITIAL_CAPACITY);
+    while let Ok(chunk) = agg_rx.recv().await {
+        append_all(&mut combined_buf, &chunk);
+    }
+    let aggregated_output = StreamOutput {
+        text: combined_buf,
+        truncated_after_lines: None,
+    };
+
     Ok(RawExecToolCallOutput {
         exit_status,
         stdout,
         stderr,
+        aggregated_output,
     })
 }
 
 async fn read_capped<R: AsyncRead + Unpin + Send + 'static>(
     mut reader: R,
-    max_output: usize,
-    max_lines: usize,
     stream: Option<StdoutStream>,
     is_stderr: bool,
+    aggregate_tx: Option<Sender<Vec<u8>>>,
 ) -> io::Result<StreamOutput<Vec<u8>>> {
-    let mut buf = Vec::with_capacity(max_output.min(8 * 1024));
-    let mut tmp = [0u8; 8192];
+    let mut buf = Vec::with_capacity(AGGREGATE_BUFFER_INITIAL_CAPACITY);
+    let mut tmp = [0u8; READ_CHUNK_SIZE];
+    let mut emitted_deltas: usize = 0;
 
-    let mut remaining_bytes = max_output;
-    let mut remaining_lines = max_lines;
+    // No caps: append all bytes
 
     loop {
         let n = reader.read(&mut tmp).await?;
@@ -336,7 +358,9 @@ async fn read_capped<R: AsyncRead + Unpin + Send + 'static>(
             break;
         }
 
-        if let Some(stream) = &stream {
+        if let Some(stream) = &stream
+            && emitted_deltas < MAX_EXEC_OUTPUT_DELTAS_PER_CALL
+        {
             let chunk = tmp[..n].to_vec();
             let msg = EventMsg::ExecCommandOutputDelta(ExecCommandOutputDeltaEvent {
                 call_id: stream.call_id.clone(),
@@ -353,35 +377,20 @@ async fn read_capped<R: AsyncRead + Unpin + Send + 'static>(
             };
             #[allow(clippy::let_unit_value)]
             let _ = stream.tx_event.send(event).await;
+            emitted_deltas += 1;
         }
 
-        // Copy into the buffer only while we still have byte and line budget.
-        if remaining_bytes > 0 && remaining_lines > 0 {
-            let mut copy_len = 0;
-            for &b in &tmp[..n] {
-                if remaining_bytes == 0 || remaining_lines == 0 {
-                    break;
-                }
-                copy_len += 1;
-                remaining_bytes -= 1;
-                if b == b'\n' {
-                    remaining_lines -= 1;
-                }
-            }
-            buf.extend_from_slice(&tmp[..copy_len]);
+        if let Some(tx) = &aggregate_tx {
+            let _ = tx.send(tmp[..n].to_vec()).await;
         }
-        // Continue reading to EOF to avoid back-pressure, but discard once caps are hit.
+
+        append_all(&mut buf, &tmp[..n]);
+        // Continue reading to EOF to avoid back-pressure
     }
 
-    let truncated = remaining_lines == 0 || remaining_bytes == 0;
-
     Ok(StreamOutput {
         text: buf,
-        truncated_after_lines: if truncated {
-            Some((max_lines - remaining_lines) as u32)
-        } else {
-            None
-        },
+        truncated_after_lines: None,
     })
 }
 

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

@@ -0,0 +1,57 @@
+use serde::Deserialize;
+use serde::Serialize;
+
+use crate::exec_command::session_id::SessionId;
+
+#[derive(Debug, Clone, Deserialize)]
+pub struct ExecCommandParams {
+    pub(crate) cmd: String,
+
+    #[serde(default = "default_yield_time")]
+    pub(crate) yield_time_ms: u64,
+
+    #[serde(default = "max_output_tokens")]
+    pub(crate) max_output_tokens: u64,
+
+    #[serde(default = "default_shell")]
+    pub(crate) shell: String,
+
+    #[serde(default = "default_login")]
+    pub(crate) login: bool,
+}
+
+fn default_yield_time() -> u64 {
+    10_000
+}
+
+fn max_output_tokens() -> u64 {
+    10_000
+}
+
+fn default_login() -> bool {
+    true
+}
+
+fn default_shell() -> String {
+    "/bin/bash".to_string()
+}
+
+#[derive(Debug, Deserialize, Serialize)]
+pub struct WriteStdinParams {
+    pub(crate) session_id: SessionId,
+    pub(crate) chars: String,
+
+    #[serde(default = "write_stdin_default_yield_time_ms")]
+    pub(crate) yield_time_ms: u64,
+
+    #[serde(default = "write_stdin_default_max_output_tokens")]
+    pub(crate) max_output_tokens: u64,
+}
+
+fn write_stdin_default_yield_time_ms() -> u64 {
+    250
+}
+
+fn write_stdin_default_max_output_tokens() -> u64 {
+    10_000
+}

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

@@ -0,0 +1,83 @@
+use std::sync::Mutex as StdMutex;
+
+use tokio::sync::broadcast;
+use tokio::sync::mpsc;
+use tokio::task::JoinHandle;
+
+#[derive(Debug)]
+pub(crate) struct ExecCommandSession {
+    /// Queue for writing bytes to the process stdin (PTY master write side).
+    writer_tx: mpsc::Sender<Vec<u8>>,
+    /// Broadcast stream of output chunks read from the PTY. New subscribers
+    /// receive only chunks emitted after they subscribe.
+    output_tx: broadcast::Sender<Vec<u8>>,
+
+    /// Child killer handle for termination on drop (can signal independently
+    /// of a thread blocked in `.wait()`).
+    killer: StdMutex<Option<Box<dyn portable_pty::ChildKiller + Send + Sync>>>,
+
+    /// JoinHandle for the blocking PTY reader task.
+    reader_handle: StdMutex<Option<JoinHandle<()>>>,
+
+    /// JoinHandle for the stdin writer task.
+    writer_handle: StdMutex<Option<JoinHandle<()>>>,
+
+    /// JoinHandle for the child wait task.
+    wait_handle: StdMutex<Option<JoinHandle<()>>>,
+}
+
+impl ExecCommandSession {
+    pub(crate) fn new(
+        writer_tx: mpsc::Sender<Vec<u8>>,
+        output_tx: broadcast::Sender<Vec<u8>>,
+        killer: Box<dyn portable_pty::ChildKiller + Send + Sync>,
+        reader_handle: JoinHandle<()>,
+        writer_handle: JoinHandle<()>,
+        wait_handle: JoinHandle<()>,
+    ) -> Self {
+        Self {
+            writer_tx,
+            output_tx,
+            killer: StdMutex::new(Some(killer)),
+            reader_handle: StdMutex::new(Some(reader_handle)),
+            writer_handle: StdMutex::new(Some(writer_handle)),
+            wait_handle: StdMutex::new(Some(wait_handle)),
+        }
+    }
+
+    pub(crate) fn writer_sender(&self) -> mpsc::Sender<Vec<u8>> {
+        self.writer_tx.clone()
+    }
+
+    pub(crate) fn output_receiver(&self) -> broadcast::Receiver<Vec<u8>> {
+        self.output_tx.subscribe()
+    }
+}
+
+impl Drop for ExecCommandSession {
+    fn drop(&mut self) {
+        // Best-effort: terminate child first so blocking tasks can complete.
+        if let Ok(mut killer_opt) = self.killer.lock()
+            && let Some(mut killer) = killer_opt.take()
+        {
+            let _ = killer.kill();
+        }
+
+        // Abort background tasks; they may already have exited after kill.
+        if let Ok(mut h) = self.reader_handle.lock()
+            && let Some(handle) = h.take()
+        {
+            handle.abort();
+        }
+        if let Ok(mut h) = self.writer_handle.lock()
+            && let Some(handle) = h.take()
+        {
+            handle.abort();
+        }
+        if let Ok(mut h) = self.wait_handle.lock()
+            && let Some(handle) = h.take()
+        {
+            handle.abort();
+        }
+    }
+}

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

@@ -0,0 +1,14 @@
+mod exec_command_params;
+mod exec_command_session;
+mod responses_api;
+mod session_id;
+mod session_manager;
+
+pub use exec_command_params::ExecCommandParams;
+pub use exec_command_params::WriteStdinParams;
+pub use responses_api::EXEC_COMMAND_TOOL_NAME;
+pub use responses_api::WRITE_STDIN_TOOL_NAME;
+pub use responses_api::create_exec_command_tool_for_responses_api;
+pub use responses_api::create_write_stdin_tool_for_responses_api;
+pub use session_manager::SessionManager as ExecSessionManager;
+pub use session_manager::result_into_payload;

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

@@ -0,0 +1,98 @@
+use std::collections::BTreeMap;
+
+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";
+
+pub fn create_exec_command_tool_for_responses_api() -> ResponsesApiTool {
+    let mut properties = BTreeMap::<String, JsonSchema>::new();
+    properties.insert(
+        "cmd".to_string(),
+        JsonSchema::String {
+            description: Some("The shell command to execute.".to_string()),
+        },
+    );
+    properties.insert(
+        "yield_time_ms".to_string(),
+        JsonSchema::Number {
+            description: Some("The maximum time in milliseconds to wait for output.".to_string()),
+        },
+    );
+    properties.insert(
+        "max_output_tokens".to_string(),
+        JsonSchema::Number {
+            description: Some("The maximum number of tokens to output.".to_string()),
+        },
+    );
+    properties.insert(
+        "shell".to_string(),
+        JsonSchema::String {
+            description: Some("The shell to use. Defaults to \"/bin/bash\".".to_string()),
+        },
+    );
+    properties.insert(
+        "login".to_string(),
+        JsonSchema::Boolean {
+            description: Some(
+                "Whether to run the command as a login shell. Defaults to true.".to_string(),
+            ),
+        },
+    );
+
+    ResponsesApiTool {
+        name: EXEC_COMMAND_TOOL_NAME.to_owned(),
+        description: r#"Execute shell commands on the local machine with streaming output."#
+            .to_string(),
+        strict: false,
+        parameters: JsonSchema::Object {
+            properties,
+            required: Some(vec!["cmd".to_string()]),
+            additional_properties: Some(false),
+        },
+    }
+}
+
+pub fn create_write_stdin_tool_for_responses_api() -> ResponsesApiTool {
+    let mut properties = BTreeMap::<String, JsonSchema>::new();
+    properties.insert(
+        "session_id".to_string(),
+        JsonSchema::Number {
+            description: Some("The ID of the exec_command session.".to_string()),
+        },
+    );
+    properties.insert(
+        "chars".to_string(),
+        JsonSchema::String {
+            description: Some("The characters to write to stdin.".to_string()),
+        },
+    );
+    properties.insert(
+        "yield_time_ms".to_string(),
+        JsonSchema::Number {
+            description: Some(
+                "The maximum time in milliseconds to wait for output after writing.".to_string(),
+            ),
+        },
+    );
+    properties.insert(
+        "max_output_tokens".to_string(),
+        JsonSchema::Number {
+            description: Some("The maximum number of tokens to output.".to_string()),
+        },
+    );
+
+    ResponsesApiTool {
+        name: WRITE_STDIN_TOOL_NAME.to_owned(),
+        description: r#"Write characters to an exec session's stdin. Returns all stdout+stderr received within yield_time_ms.
+Can write control characters (\u0003 for Ctrl-C), or an empty string to just poll stdout+stderr."#
+            .to_string(),
+        strict: false,
+        parameters: JsonSchema::Object {
+            properties,
+            required: Some(vec!["session_id".to_string(), "chars".to_string()]),
+            additional_properties: Some(false),
+        },
+    }
+}

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

@@ -0,0 +1,5 @@
+use serde::Deserialize;
+use serde::Serialize;
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, Serialize, Deserialize)]
+pub(crate) struct SessionId(pub u32);

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

@@ -0,0 +1,665 @@
+use std::collections::HashMap;
+use std::io::ErrorKind;
+use std::io::Read;
+use std::sync::Arc;
+use std::sync::Mutex as StdMutex;
+use std::sync::atomic::AtomicU32;
+
+use portable_pty::CommandBuilder;
+use portable_pty::PtySize;
+use portable_pty::native_pty_system;
+use tokio::sync::Mutex;
+use tokio::sync::mpsc;
+use tokio::sync::oneshot;
+use tokio::time::Duration;
+use tokio::time::Instant;
+use tokio::time::timeout;
+
+use crate::exec_command::exec_command_params::ExecCommandParams;
+use crate::exec_command::exec_command_params::WriteStdinParams;
+use crate::exec_command::exec_command_session::ExecCommandSession;
+use crate::exec_command::session_id::SessionId;
+use codex_protocol::models::FunctionCallOutputPayload;
+
+#[derive(Debug, Default)]
+pub struct SessionManager {
+    next_session_id: AtomicU32,
+    sessions: Mutex<HashMap<SessionId, ExecCommandSession>>,
+}
+
+#[derive(Debug)]
+pub struct ExecCommandOutput {
+    wall_time: Duration,
+    exit_status: ExitStatus,
+    original_token_count: Option<u64>,
+    output: String,
+}
+
+impl ExecCommandOutput {
+    fn to_text_output(&self) -> String {
+        let wall_time_secs = self.wall_time.as_secs_f32();
+        let termination_status = match self.exit_status {
+            ExitStatus::Exited(code) => format!("Process exited with code {code}"),
+            ExitStatus::Ongoing(session_id) => {
+                format!("Process running with session ID {}", session_id.0)
+            }
+        };
+        let truncation_status = match self.original_token_count {
+            Some(tokens) => {
+                format!("\nWarning: truncated output (original token count: {tokens})")
+            }
+            None => "".to_string(),
+        };
+        format!(
+            r#"Wall time: {wall_time_secs:.3} seconds
+{termination_status}{truncation_status}
+Output:
+{output}"#,
+            output = self.output
+        )
+    }
+}
+
+#[derive(Debug)]
+pub enum ExitStatus {
+    Exited(i32),
+    Ongoing(SessionId),
+}
+
+pub fn result_into_payload(result: Result<ExecCommandOutput, String>) -> FunctionCallOutputPayload {
+    match result {
+        Ok(output) => FunctionCallOutputPayload {
+            content: output.to_text_output(),
+            success: Some(true),
+        },
+        Err(err) => FunctionCallOutputPayload {
+            content: err,
+            success: Some(false),
+        },
+    }
+}
+
+impl SessionManager {
+    /// Processes the request and is required to send a response via `outgoing`.
+    pub async fn handle_exec_command_request(
+        &self,
+        params: ExecCommandParams,
+    ) -> Result<ExecCommandOutput, String> {
+        // Allocate a session id.
+        let session_id = SessionId(
+            self.next_session_id
+                .fetch_add(1, std::sync::atomic::Ordering::SeqCst),
+        );
+
+        let (session, mut exit_rx) =
+            create_exec_command_session(params.clone())
+                .await
+                .map_err(|err| {
+                    format!(
+                        "failed to create exec command session for session id {}: {err}",
+                        session_id.0
+                    )
+                })?;
+
+        // Insert into session map.
+        let mut output_rx = session.output_receiver();
+        self.sessions.lock().await.insert(session_id, session);
+
+        // Collect output until either timeout expires or process exits.
+        // Do not cap during collection; truncate at the end if needed.
+        // Use a modest initial capacity to avoid large preallocation.
+        let cap_bytes_u64 = params.max_output_tokens.saturating_mul(4);
+        let cap_bytes: usize = cap_bytes_u64.min(usize::MAX as u64) as usize;
+        let mut collected: Vec<u8> = Vec::with_capacity(4096);
+
+        let start_time = Instant::now();
+        let deadline = start_time + Duration::from_millis(params.yield_time_ms);
+        let mut exit_code: Option<i32> = None;
+
+        loop {
+            if Instant::now() >= deadline {
+                break;
+            }
+            let remaining = deadline.saturating_duration_since(Instant::now());
+            tokio::select! {
+                biased;
+                exit = &mut exit_rx => {
+                    exit_code = exit.ok();
+                    // Small grace period to pull remaining buffered output
+                    let grace_deadline = Instant::now() + Duration::from_millis(25);
+                    while Instant::now() < grace_deadline {
+                        match timeout(Duration::from_millis(1), output_rx.recv()).await {
+                            Ok(Ok(chunk)) => {
+                                collected.extend_from_slice(&chunk);
+                            }
+                            Ok(Err(tokio::sync::broadcast::error::RecvError::Lagged(_))) => {
+                                // Skip missed messages; keep trying within grace period.
+                                continue;
+                            }
+                            Ok(Err(tokio::sync::broadcast::error::RecvError::Closed)) => break,
+                            Err(_) => break,
+                        }
+                    }
+                    break;
+                }
+                chunk = timeout(remaining, output_rx.recv()) => {
+                    match chunk {
+                        Ok(Ok(chunk)) => {
+                            collected.extend_from_slice(&chunk);
+                        }
+                        Ok(Err(tokio::sync::broadcast::error::RecvError::Lagged(_))) => {
+                            // Skip missed messages; continue collecting fresh output.
+                        }
+                        Ok(Err(tokio::sync::broadcast::error::RecvError::Closed)) => { break; }
+                        Err(_) => { break; }
+                    }
+                }
+            }
+        }
+
+        let output = String::from_utf8_lossy(&collected).to_string();
+
+        let exit_status = if let Some(code) = exit_code {
+            ExitStatus::Exited(code)
+        } else {
+            ExitStatus::Ongoing(session_id)
+        };
+
+        // If output exceeds cap, truncate the middle and record original token estimate.
+        let (output, original_token_count) = truncate_middle(&output, cap_bytes);
+        Ok(ExecCommandOutput {
+            wall_time: Instant::now().duration_since(start_time),
+            exit_status,
+            original_token_count,
+            output,
+        })
+    }
+
+    /// Write characters to a session's stdin and collect combined output for up to `yield_time_ms`.
+    pub async fn handle_write_stdin_request(
+        &self,
+        params: WriteStdinParams,
+    ) -> Result<ExecCommandOutput, String> {
+        let WriteStdinParams {
+            session_id,
+            chars,
+            yield_time_ms,
+            max_output_tokens,
+        } = params;
+
+        // Grab handles without holding the sessions lock across await points.
+        let (writer_tx, mut output_rx) = {
+            let sessions = self.sessions.lock().await;
+            match sessions.get(&session_id) {
+                Some(session) => (session.writer_sender(), session.output_receiver()),
+                None => {
+                    return Err(format!("unknown session id {}", session_id.0));
+                }
+            }
+        };
+
+        // Write stdin if provided.
+        if !chars.is_empty() && writer_tx.send(chars.into_bytes()).await.is_err() {
+            return Err("failed to write to stdin".to_string());
+        }
+
+        // Collect output up to yield_time_ms, truncating to max_output_tokens bytes.
+        let mut collected: Vec<u8> = Vec::with_capacity(4096);
+        let start_time = Instant::now();
+        let deadline = start_time + Duration::from_millis(yield_time_ms);
+        loop {
+            let now = Instant::now();
+            if now >= deadline {
+                break;
+            }
+            let remaining = deadline - now;
+            match timeout(remaining, output_rx.recv()).await {
+                Ok(Ok(chunk)) => {
+                    // Collect all output within the time budget; truncate at the end.
+                    collected.extend_from_slice(&chunk);
+                }
+                Ok(Err(tokio::sync::broadcast::error::RecvError::Lagged(_))) => {
+                    // Skip missed messages; continue collecting fresh output.
+                }
+                Ok(Err(tokio::sync::broadcast::error::RecvError::Closed)) => break,
+                Err(_) => break, // timeout
+            }
+        }
+
+        // Return structured output, truncating middle if over cap.
+        let output = String::from_utf8_lossy(&collected).to_string();
+        let cap_bytes_u64 = max_output_tokens.saturating_mul(4);
+        let cap_bytes: usize = cap_bytes_u64.min(usize::MAX as u64) as usize;
+        let (output, original_token_count) = truncate_middle(&output, cap_bytes);
+        Ok(ExecCommandOutput {
+            wall_time: Instant::now().duration_since(start_time),
+            exit_status: ExitStatus::Ongoing(session_id),
+            original_token_count,
+            output,
+        })
+    }
+}
+
+/// Spawn PTY and child process per spawn_exec_command_session logic.
+async fn create_exec_command_session(
+    params: ExecCommandParams,
+) -> anyhow::Result<(ExecCommandSession, oneshot::Receiver<i32>)> {
+    let ExecCommandParams {
+        cmd,
+        yield_time_ms: _,
+        max_output_tokens: _,
+        shell,
+        login,
+    } = params;
+
+    // Use the native pty implementation for the system
+    let pty_system = native_pty_system();
+
+    // Create a new pty
+    let pair = pty_system.openpty(PtySize {
+        rows: 24,
+        cols: 80,
+        pixel_width: 0,
+        pixel_height: 0,
+    })?;
+
+    // Spawn a shell into the pty
+    let mut command_builder = CommandBuilder::new(shell);
+    let shell_mode_opt = if login { "-lc" } else { "-c" };
+    command_builder.arg(shell_mode_opt);
+    command_builder.arg(cmd);
+
+    let mut child = pair.slave.spawn_command(command_builder)?;
+    // Obtain a killer that can signal the process independently of `.wait()`.
+    let killer = child.clone_killer();
+
+    // Channel to forward write requests to the PTY writer.
+    let (writer_tx, mut writer_rx) = mpsc::channel::<Vec<u8>>(128);
+    // Broadcast for streaming PTY output to readers: subscribers receive from subscription time.
+    let (output_tx, _) = tokio::sync::broadcast::channel::<Vec<u8>>(256);
+
+    // Reader task: drain PTY and forward chunks to output channel.
+    let mut reader = pair.master.try_clone_reader()?;
+    let output_tx_clone = output_tx.clone();
+    let reader_handle = tokio::task::spawn_blocking(move || {
+        let mut buf = [0u8; 8192];
+        loop {
+            match reader.read(&mut buf) {
+                Ok(0) => break, // EOF
+                Ok(n) => {
+                    // Forward to broadcast; best-effort if there are subscribers.
+                    let _ = output_tx_clone.send(buf[..n].to_vec());
+                }
+                Err(ref e) if e.kind() == ErrorKind::Interrupted => {
+                    // Retry on EINTR
+                    continue;
+                }
+                Err(ref e) if e.kind() == ErrorKind::WouldBlock => {
+                    // We're in a blocking thread; back off briefly and retry.
+                    std::thread::sleep(Duration::from_millis(5));
+                    continue;
+                }
+                Err(_) => break,
+            }
+        }
+    });
+
+    // Writer task: apply stdin writes to the PTY writer.
+    let writer = pair.master.take_writer()?;
+    let writer = Arc::new(StdMutex::new(writer));
+    let writer_handle = tokio::spawn({
+        let writer = writer.clone();
+        async move {
+            while let Some(bytes) = writer_rx.recv().await {
+                let writer = writer.clone();
+                // Perform blocking write on a blocking thread.
+                let _ = tokio::task::spawn_blocking(move || {
+                    if let Ok(mut guard) = writer.lock() {
+                        use std::io::Write;
+                        let _ = guard.write_all(&bytes);
+                        let _ = guard.flush();
+                    }
+                })
+                .await;
+            }
+        }
+    });
+
+    // Keep the child alive until it exits, then signal exit code.
+    let (exit_tx, exit_rx) = oneshot::channel::<i32>();
+    let wait_handle = tokio::task::spawn_blocking(move || {
+        let code = match child.wait() {
+            Ok(status) => status.exit_code() as i32,
+            Err(_) => -1,
+        };
+        let _ = exit_tx.send(code);
+    });
+
+    // Create and store the session with channels.
+    let session = ExecCommandSession::new(
+        writer_tx,
+        output_tx,
+        killer,
+        reader_handle,
+        writer_handle,
+        wait_handle,
+    );
+    Ok((session, exit_rx))
+}
+
+/// Truncate the middle of a UTF-8 string to at most `max_bytes` bytes,
+/// preserving the beginning and the end. Returns the possibly truncated
+/// string and `Some(original_token_count)` (estimated at 4 bytes/token)
+/// if truncation occurred; otherwise returns the original string and `None`.
+fn truncate_middle(s: &str, max_bytes: usize) -> (String, Option<u64>) {
+    // No truncation needed
+    if s.len() <= max_bytes {
+        return (s.to_string(), None);
+    }
+    let est_tokens = (s.len() as u64).div_ceil(4);
+    if max_bytes == 0 {
+        // Cannot keep any content; still return a full marker (never truncated).
+        return (format!("…{est_tokens} tokens truncated…"), Some(est_tokens));
+    }
+
+    // Helper to truncate a string to a given byte length on a char boundary.
+    fn truncate_on_boundary(input: &str, max_len: usize) -> &str {
+        if input.len() <= max_len {
+            return input;
+        }
+        let mut end = max_len;
+        while end > 0 && !input.is_char_boundary(end) {
+            end -= 1;
+        }
+        &input[..end]
+    }
+
+    // Given a left/right budget, prefer newline boundaries; otherwise fall back
+    // to UTF-8 char boundaries.
+    fn pick_prefix_end(s: &str, left_budget: usize) -> usize {
+        if let Some(head) = s.get(..left_budget)
+            && let Some(i) = head.rfind('\n')
+        {
+            return i + 1; // keep the newline so suffix starts on a fresh line
+        }
+        truncate_on_boundary(s, left_budget).len()
+    }
+
+    fn pick_suffix_start(s: &str, right_budget: usize) -> usize {
+        let start_tail = s.len().saturating_sub(right_budget);
+        if let Some(tail) = s.get(start_tail..)
+            && let Some(i) = tail.find('\n')
+        {
+            return start_tail + i + 1; // start after newline
+        }
+        // Fall back to a char boundary at or after start_tail.
+        let mut idx = start_tail.min(s.len());
+        while idx < s.len() && !s.is_char_boundary(idx) {
+            idx += 1;
+        }
+        idx
+    }
+
+    // Refine marker length and budgets until stable. Marker is never truncated.
+    let mut guess_tokens = est_tokens; // worst-case: everything truncated
+    for _ in 0..4 {
+        let marker = format!("…{guess_tokens} tokens truncated…");
+        let marker_len = marker.len();
+        let keep_budget = max_bytes.saturating_sub(marker_len);
+        if keep_budget == 0 {
+            // No room for any content within the cap; return a full, untruncated marker
+            // that reflects the entire truncated content.
+            return (format!("…{est_tokens} tokens truncated…"), Some(est_tokens));
+        }
+
+        let left_budget = keep_budget / 2;
+        let right_budget = keep_budget - left_budget;
+        let prefix_end = pick_prefix_end(s, left_budget);
+        let mut suffix_start = pick_suffix_start(s, right_budget);
+        if suffix_start < prefix_end {
+            suffix_start = prefix_end;
+        }
+        let kept_content_bytes = prefix_end + (s.len() - suffix_start);
+        let truncated_content_bytes = s.len().saturating_sub(kept_content_bytes);
+        let new_tokens = (truncated_content_bytes as u64).div_ceil(4);
+        if new_tokens == guess_tokens {
+            let mut out = String::with_capacity(marker_len + kept_content_bytes + 1);
+            out.push_str(&s[..prefix_end]);
+            out.push_str(&marker);
+            // Place marker on its own line for symmetry when we keep line boundaries.
+            out.push('\n');
+            out.push_str(&s[suffix_start..]);
+            return (out, Some(est_tokens));
+        }
+        guess_tokens = new_tokens;
+    }
+
+    // Fallback: use last guess to build output.
+    let marker = format!("…{guess_tokens} tokens truncated…");
+    let marker_len = marker.len();
+    let keep_budget = max_bytes.saturating_sub(marker_len);
+    if keep_budget == 0 {
+        return (format!("…{est_tokens} tokens truncated…"), Some(est_tokens));
+    }
+    let left_budget = keep_budget / 2;
+    let right_budget = keep_budget - left_budget;
+    let prefix_end = pick_prefix_end(s, left_budget);
+    let suffix_start = pick_suffix_start(s, right_budget);
+    let mut out = String::with_capacity(marker_len + prefix_end + (s.len() - suffix_start) + 1);
+    out.push_str(&s[..prefix_end]);
+    out.push_str(&marker);
+    out.push('\n');
+    out.push_str(&s[suffix_start..]);
+    (out, Some(est_tokens))
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::exec_command::session_id::SessionId;
+
+    /// Test that verifies that [`SessionManager::handle_exec_command_request()`]
+    /// and [`SessionManager::handle_write_stdin_request()`] work as expected
+    /// in the presence of a process that never terminates (but produces
+    /// output continuously).
+    #[cfg(unix)]
+    #[allow(clippy::print_stderr)]
+    #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
+    async fn session_manager_streams_and_truncates_from_now() {
+        use crate::exec_command::exec_command_params::ExecCommandParams;
+        use crate::exec_command::exec_command_params::WriteStdinParams;
+        use tokio::time::sleep;
+
+        let session_manager = SessionManager::default();
+        // Long-running loop that prints an increasing counter every ~100ms.
+        // Use Python for a portable, reliable sleep across shells/PTYs.
+        let cmd = r#"python3 - <<'PY'
+import sys, time
+count = 0
+while True:
+    print(count)
+    sys.stdout.flush()
+    count += 100
+    time.sleep(0.1)
+PY"#
+        .to_string();
+
+        // Start the session and collect ~3s of output.
+        let params = ExecCommandParams {
+            cmd,
+            yield_time_ms: 3_000,
+            max_output_tokens: 1_000, // large enough to avoid truncation here
+            shell: "/bin/bash".to_string(),
+            login: false,
+        };
+        let initial_output = match session_manager
+            .handle_exec_command_request(params.clone())
+            .await
+        {
+            Ok(v) => v,
+            Err(e) => {
+                // PTY may be restricted in some sandboxes; skip in that case.
+                if e.contains("openpty") || e.contains("Operation not permitted") {
+                    eprintln!("skipping test due to restricted PTY: {e}");
+                    return;
+                }
+                panic!("exec request failed unexpectedly: {e}");
+            }
+        };
+        eprintln!("initial output: {initial_output:?}");
+
+        // Should be ongoing (we launched a never-ending loop).
+        let session_id = match initial_output.exit_status {
+            ExitStatus::Ongoing(id) => id,
+            _ => panic!("expected ongoing session"),
+        };
+
+        // Parse the numeric lines and get the max observed value in the first window.
+        let first_nums = extract_monotonic_numbers(&initial_output.output);
+        assert!(
+            !first_nums.is_empty(),
+            "expected some output from first window"
+        );
+        let first_max = *first_nums.iter().max().unwrap();
+
+        // Wait ~4s so counters progress while we're not reading.
+        sleep(Duration::from_millis(4_000)).await;
+
+        // Now read ~3s of output "from now" only.
+        // Use a small token cap so truncation occurs and we test middle truncation.
+        let write_params = WriteStdinParams {
+            session_id,
+            chars: String::new(),
+            yield_time_ms: 3_000,
+            max_output_tokens: 16, // 16 tokens ~= 64 bytes -> likely truncation
+        };
+        let second = session_manager
+            .handle_write_stdin_request(write_params)
+            .await
+            .expect("write stdin should succeed");
+
+        // Verify truncation metadata and size bound (cap is tokens*4 bytes).
+        assert!(second.original_token_count.is_some());
+        let cap_bytes = (16u64 * 4) as usize;
+        assert!(second.output.len() <= cap_bytes);
+        // New middle marker should be present.
+        assert!(
+            second.output.contains("tokens truncated") && second.output.contains('…'),
+            "expected truncation marker in output, got: {}",
+            second.output
+        );
+
+        // Minimal freshness check: the earliest number we see in the second window
+        // should be significantly larger than the last from the first window.
+        let second_nums = extract_monotonic_numbers(&second.output);
+        assert!(
+            !second_nums.is_empty(),
+            "expected some numeric output from second window"
+        );
+        let second_min = *second_nums.iter().min().unwrap();
+
+        // We slept 4 seconds (~40 ticks at 100ms/tick, each +100), so expect
+        // an increase of roughly 4000 or more. Allow a generous margin.
+        assert!(
+            second_min >= first_max + 2000,
+            "second_min={second_min} first_max={first_max}",
+        );
+    }
+
+    #[cfg(unix)]
+    fn extract_monotonic_numbers(s: &str) -> Vec<i64> {
+        s.lines()
+            .filter_map(|line| {
+                if !line.is_empty()
+                    && line.chars().all(|c| c.is_ascii_digit())
+                    && let Ok(n) = line.parse::<i64>()
+                {
+                    // Our generator increments by 100; ignore spurious fragments.
+                    if n % 100 == 0 {
+                        return Some(n);
+                    }
+                }
+                None
+            })
+            .collect()
+    }
+
+    #[test]
+    fn to_text_output_exited_no_truncation() {
+        let out = ExecCommandOutput {
+            wall_time: Duration::from_millis(1234),
+            exit_status: ExitStatus::Exited(0),
+            original_token_count: None,
+            output: "hello".to_string(),
+        };
+        let text = out.to_text_output();
+        let expected = r#"Wall time: 1.234 seconds
+Process exited with code 0
+Output:
+hello"#;
+        assert_eq!(expected, text);
+    }
+
+    #[test]
+    fn to_text_output_ongoing_with_truncation() {
+        let out = ExecCommandOutput {
+            wall_time: Duration::from_millis(500),
+            exit_status: ExitStatus::Ongoing(SessionId(42)),
+            original_token_count: Some(1000),
+            output: "abc".to_string(),
+        };
+        let text = out.to_text_output();
+        let expected = r#"Wall time: 0.500 seconds
+Process running with session ID 42
+Warning: truncated output (original token count: 1000)
+Output:
+abc"#;
+        assert_eq!(expected, text);
+    }
+
+    #[test]
+    fn truncate_middle_no_newlines_fallback() {
+        // A long string with no newlines that exceeds the cap.
+        let s = "abcdefghijklmnopqrstuvwxyz0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ";
+        let max_bytes = 16; // force truncation
+        let (out, original) = truncate_middle(s, max_bytes);
+        // For very small caps, we return the full, untruncated marker,
+        // even if it exceeds the cap.
+        assert_eq!(out, "…16 tokens truncated…");
+        // Original string length is 62 bytes => ceil(62/4) = 16 tokens.
+        assert_eq!(original, Some(16));
+    }
+
+    #[test]
+    fn truncate_middle_prefers_newline_boundaries() {
+        // Build a multi-line string of 20 numbered lines (each "NNN\n").
+        let mut s = String::new();
+        for i in 1..=20 {
+            s.push_str(&format!("{i:03}\n"));
+        }
+        // Total length: 20 lines * 4 bytes per line = 80 bytes.
+        assert_eq!(s.len(), 80);
+
+        // Choose a cap that forces truncation while leaving room for
+        // a few lines on each side after accounting for the marker.
+        let max_bytes = 64;
+        // Expect exact output: first 4 lines, marker, last 4 lines, and correct token estimate (80/4 = 20).
+        assert_eq!(
+            truncate_middle(&s, max_bytes),
+            (
+                r#"001
+002
+003
+004
+…12 tokens truncated…
+017
+018
+019
+020
+"#
+                .to_string(),
+                Some(20)
+            )
+        );
+    }
+}

codex-rs/core/src/git_info.rs

@@ -385,7 +385,9 @@ async fn find_closest_sha(cwd: &Path, branches: &[String], remotes: &[String]) -
 }
 
 async fn diff_against_sha(cwd: &Path, sha: &GitSha) -> Option<String> {
-    let output = run_git_command_with_timeout(&["diff", &sha.0], cwd).await?;
+    let output =
+        run_git_command_with_timeout(&["diff", "--no-textconv", "--no-ext-diff", &sha.0], cwd)
+            .await?;
     // 0 is success and no diff.
     // 1 is success but there is a diff.
     let exit_ok = output.status.code().is_some_and(|c| c == 0 || c == 1);
@@ -406,10 +408,21 @@ async fn diff_against_sha(cwd: &Path, sha: &GitSha) -> Option<String> {
             .collect();
 
         if !untracked.is_empty() {
+            // Use platform-appropriate null device and guard paths with `--`.
+            let null_device: &str = if cfg!(windows) { "NUL" } else { "/dev/null" };
             let futures_iter = untracked.into_iter().map(|file| async move {
                 let file_owned = file;
-                let args_vec: Vec<&str> =
-                    vec!["diff", "--binary", "--no-index", "/dev/null", &file_owned];
+                let args_vec: Vec<&str> = vec![
+                    "diff",
+                    "--no-textconv",
+                    "--no-ext-diff",
+                    "--binary",
+                    "--no-index",
+                    // -- ensures that filenames that start with - are not treated as options.
+                    "--",
+                    null_device,
+                    &file_owned,
+                ];
                 run_git_command_with_timeout(&args_vec, cwd).await
             });
             let results = join_all(futures_iter).await;

codex-rs/core/src/lib.rs

@@ -17,9 +17,11 @@ pub mod config;
 pub mod config_profile;
 pub mod config_types;
 mod conversation_history;
+pub mod custom_prompts;
 mod environment_context;
 pub mod error;
 pub mod exec;
+mod exec_command;
 pub mod exec_env;
 mod flags;
 pub mod git_info;

codex-rs/core/src/model_family.rs

@@ -90,7 +90,6 @@ pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
         model_family!(
             slug, slug,
             supports_reasoning_summaries: true,
-            apply_patch_tool_type: Some(ApplyPatchToolType::Freeform),
         )
     } else if slug.starts_with("gpt-4.1") {
         model_family!(
@@ -107,7 +106,6 @@ pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
         model_family!(
             slug, "gpt-5",
             supports_reasoning_summaries: true,
-            apply_patch_tool_type: Some(ApplyPatchToolType::Freeform),
         )
     } else {
         None

codex-rs/core/src/model_provider_info.rs

@@ -17,6 +17,10 @@ use crate::error::EnvVarError;
 const DEFAULT_STREAM_IDLE_TIMEOUT_MS: u64 = 300_000;
 const DEFAULT_STREAM_MAX_RETRIES: u64 = 5;
 const DEFAULT_REQUEST_MAX_RETRIES: u64 = 4;
+/// Hard cap for user-configured `stream_max_retries`.
+const MAX_STREAM_MAX_RETRIES: u64 = 100;
+/// Hard cap for user-configured `request_max_retries`.
+const MAX_REQUEST_MAX_RETRIES: u64 = 100;
 
 /// Wire protocol that the provider speaks. Most third-party services only
 /// implement the classic OpenAI Chat Completions JSON schema, whereas OpenAI
@@ -207,12 +211,14 @@ impl ModelProviderInfo {
     pub fn request_max_retries(&self) -> u64 {
         self.request_max_retries
             .unwrap_or(DEFAULT_REQUEST_MAX_RETRIES)
+            .min(MAX_REQUEST_MAX_RETRIES)
     }
 
     /// Effective maximum number of stream reconnection attempts for this provider.
     pub fn stream_max_retries(&self) -> u64 {
         self.stream_max_retries
             .unwrap_or(DEFAULT_STREAM_MAX_RETRIES)
+            .min(MAX_STREAM_MAX_RETRIES)
     }
 
     /// Effective idle timeout for streaming responses.

codex-rs/core/src/openai_model_info.rs

@@ -79,13 +79,13 @@ pub(crate) fn get_model_info(model_family: &ModelFamily) -> Option<ModelInfo> {
         }),
 
         "gpt-5" => Some(ModelInfo {
-            context_window: 200_000,
-            max_output_tokens: 100_000,
+            context_window: 400_000,
+            max_output_tokens: 128_000,
         }),
 
         _ if slug.starts_with("codex-") => Some(ModelInfo {
-            context_window: 200_000,
-            max_output_tokens: 100_000,
+            context_window: 400_000,
+            max_output_tokens: 128_000,
         }),
 
         _ => None,

codex-rs/core/src/openai_tools.rs

@@ -47,6 +47,10 @@ pub(crate) enum OpenAiTool {
     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_preview")]
+    WebSearch {},
     #[serde(rename = "custom")]
     Freeform(FreeformTool),
 }
@@ -56,29 +60,49 @@ pub enum ConfigShellToolType {
     DefaultShell,
     ShellWithRequest { sandbox_policy: SandboxPolicy },
     LocalShell,
+    StreamableShell,
 }
 
 #[derive(Debug, Clone)]
-pub struct ToolsConfig {
+pub(crate) struct ToolsConfig {
     pub shell_type: ConfigShellToolType,
     pub plan_tool: bool,
     pub apply_patch_tool_type: Option<ApplyPatchToolType>,
+    pub web_search_request: bool,
+    pub include_view_image_tool: bool,
+}
+
+pub(crate) struct ToolsConfigParams<'a> {
+    pub(crate) model_family: &'a ModelFamily,
+    pub(crate) approval_policy: AskForApproval,
+    pub(crate) sandbox_policy: SandboxPolicy,
+    pub(crate) include_plan_tool: bool,
+    pub(crate) include_apply_patch_tool: bool,
+    pub(crate) include_web_search_request: bool,
+    pub(crate) use_streamable_shell_tool: bool,
+    pub(crate) include_view_image_tool: bool,
 }
 
 impl ToolsConfig {
-    pub fn new(
-        model_family: &ModelFamily,
-        approval_policy: AskForApproval,
-        sandbox_policy: SandboxPolicy,
-        include_plan_tool: bool,
-        include_apply_patch_tool: bool,
-    ) -> Self {
-        let mut shell_type = if model_family.uses_local_shell_tool {
+    pub fn new(params: &ToolsConfigParams) -> Self {
+        let ToolsConfigParams {
+            model_family,
+            approval_policy,
+            sandbox_policy,
+            include_plan_tool,
+            include_apply_patch_tool,
+            include_web_search_request,
+            use_streamable_shell_tool,
+            include_view_image_tool,
+        } = params;
+        let mut shell_type = if *use_streamable_shell_tool {
+            ConfigShellToolType::StreamableShell
+        } else if model_family.uses_local_shell_tool {
             ConfigShellToolType::LocalShell
         } else {
             ConfigShellToolType::DefaultShell
         };
-        if matches!(approval_policy, AskForApproval::OnRequest) {
+        if matches!(approval_policy, AskForApproval::OnRequest) && !use_streamable_shell_tool {
             shell_type = ConfigShellToolType::ShellWithRequest {
                 sandbox_policy: sandbox_policy.clone(),
             }
@@ -88,7 +112,7 @@ impl ToolsConfig {
             Some(ApplyPatchToolType::Freeform) => Some(ApplyPatchToolType::Freeform),
             Some(ApplyPatchToolType::Function) => Some(ApplyPatchToolType::Function),
             None => {
-                if include_apply_patch_tool {
+                if *include_apply_patch_tool {
                     Some(ApplyPatchToolType::Freeform)
                 } else {
                     None
@@ -98,8 +122,10 @@ impl ToolsConfig {
 
         Self {
             shell_type,
-            plan_tool: include_plan_tool,
+            plan_tool: *include_plan_tool,
             apply_patch_tool_type,
+            web_search_request: *include_web_search_request,
+            include_view_image_tool: *include_view_image_tool,
         }
     }
 }
@@ -272,6 +298,30 @@ The shell tool is used to execute shell commands.
         },
     })
 }
+
+fn create_view_image_tool() -> OpenAiTool {
+    // Support only local filesystem path.
+    let mut properties = BTreeMap::new();
+    properties.insert(
+        "path".to_string(),
+        JsonSchema::String {
+            description: Some("Local filesystem path to an image file".to_string()),
+        },
+    );
+
+    OpenAiTool::Function(ResponsesApiTool {
+        name: "view_image".to_string(),
+        description:
+            "Attach a local image (by filesystem path) to the conversation context for this turn."
+                .to_string(),
+        strict: false,
+        parameters: JsonSchema::Object {
+            properties,
+            required: Some(vec!["path".to_string()]),
+            additional_properties: Some(false),
+        },
+    })
+}
 /// TODO(dylan): deprecate once we get rid of json tool
 #[derive(Serialize, Deserialize)]
 pub(crate) struct ApplyPatchToolArgs {
@@ -287,12 +337,12 @@ pub fn create_tools_json_for_responses_api(
     let mut tools_json = Vec::new();
 
     for tool in tools {
-        tools_json.push(serde_json::to_value(tool)?);
+        let json = serde_json::to_value(tool)?;
+        tools_json.push(json);
     }
 
     Ok(tools_json)
 }
-
 /// Returns JSON values that are compatible with Function Calling in the
 /// Chat Completions API:
 /// https://platform.openai.com/docs/guides/function-calling?api-mode=chat
@@ -492,6 +542,14 @@ pub(crate) fn get_openai_tools(
         ConfigShellToolType::LocalShell => {
             tools.push(OpenAiTool::LocalShell {});
         }
+        ConfigShellToolType::StreamableShell => {
+            tools.push(OpenAiTool::Function(
+                crate::exec_command::create_exec_command_tool_for_responses_api(),
+            ));
+            tools.push(OpenAiTool::Function(
+                crate::exec_command::create_write_stdin_tool_for_responses_api(),
+            ));
+        }
     }
 
     if config.plan_tool {
@@ -509,8 +567,22 @@ pub(crate) fn get_openai_tools(
         }
     }
 
+    if config.web_search_request {
+        tools.push(OpenAiTool::WebSearch {});
+    }
+
+    // Include the view_image tool so the agent can attach images to context.
+    if config.include_view_image_tool {
+        tools.push(create_view_image_tool());
+    }
+
     if let Some(mcp_tools) = mcp_tools {
-        for (name, tool) in mcp_tools {
+        // Ensure deterministic ordering to maximize prompt cache hits.
+        // HashMap iteration order is non-deterministic, so sort by fully-qualified tool name.
+        let mut entries: Vec<(String, mcp_types::Tool)> = mcp_tools.into_iter().collect();
+        entries.sort_by(|a, b| a.0.cmp(&b.0));
+
+        for (name, tool) in entries.into_iter() {
             match mcp_tool_to_openai_tool(name.clone(), tool.clone()) {
                 Ok(converted_tool) => tools.push(OpenAiTool::Function(converted_tool)),
                 Err(e) => {
@@ -537,6 +609,7 @@ mod tests {
             .map(|tool| match tool {
                 OpenAiTool::Function(ResponsesApiTool { name, .. }) => name,
                 OpenAiTool::LocalShell {} => "local_shell",
+                OpenAiTool::WebSearch {} => "web_search",
                 OpenAiTool::Freeform(FreeformTool { name, .. }) => name,
             })
             .collect::<Vec<_>>();
@@ -558,43 +631,58 @@ mod tests {
     fn test_get_openai_tools() {
         let model_family = find_family_for_model("codex-mini-latest")
             .expect("codex-mini-latest should be a valid model family");
-        let config = ToolsConfig::new(
-            &model_family,
-            AskForApproval::Never,
-            SandboxPolicy::ReadOnly,
-            true,
-            false,
-        );
+        let config = ToolsConfig::new(&ToolsConfigParams {
+            model_family: &model_family,
+            approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::ReadOnly,
+            include_plan_tool: true,
+            include_apply_patch_tool: false,
+            include_web_search_request: true,
+            use_streamable_shell_tool: false,
+            include_view_image_tool: true,
+        });
         let tools = get_openai_tools(&config, Some(HashMap::new()));
 
-        assert_eq_tool_names(&tools, &["local_shell", "update_plan"]);
+        assert_eq_tool_names(
+            &tools,
+            &["local_shell", "update_plan", "web_search", "view_image"],
+        );
     }
 
     #[test]
     fn test_get_openai_tools_default_shell() {
         let model_family = find_family_for_model("o3").expect("o3 should be a valid model family");
-        let config = ToolsConfig::new(
-            &model_family,
-            AskForApproval::Never,
-            SandboxPolicy::ReadOnly,
-            true,
-            false,
-        );
+        let config = ToolsConfig::new(&ToolsConfigParams {
+            model_family: &model_family,
+            approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::ReadOnly,
+            include_plan_tool: true,
+            include_apply_patch_tool: false,
+            include_web_search_request: true,
+            use_streamable_shell_tool: false,
+            include_view_image_tool: true,
+        });
         let tools = get_openai_tools(&config, Some(HashMap::new()));
 
-        assert_eq_tool_names(&tools, &["shell", "update_plan"]);
+        assert_eq_tool_names(
+            &tools,
+            &["shell", "update_plan", "web_search", "view_image"],
+        );
     }
 
     #[test]
     fn test_get_openai_tools_mcp_tools() {
         let model_family = find_family_for_model("o3").expect("o3 should be a valid model family");
-        let config = ToolsConfig::new(
-            &model_family,
-            AskForApproval::Never,
-            SandboxPolicy::ReadOnly,
-            false,
-            false,
-        );
+        let config = ToolsConfig::new(&ToolsConfigParams {
+            model_family: &model_family,
+            approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::ReadOnly,
+            include_plan_tool: false,
+            include_apply_patch_tool: false,
+            include_web_search_request: true,
+            use_streamable_shell_tool: false,
+            include_view_image_tool: true,
+        });
         let tools = get_openai_tools(
             &config,
             Some(HashMap::from([(
@@ -617,7 +705,7 @@ mod tests {
                                 },
                                 "required": [
                                     "string_property",
-                                    "number_property"
+                                    "number_property",
                                 ],
                                 "additionalProperties": Some(false),
                             },
@@ -633,10 +721,18 @@ mod tests {
             )])),
         );
 
-        assert_eq_tool_names(&tools, &["shell", "test_server/do_something_cool"]);
+        assert_eq_tool_names(
+            &tools,
+            &[
+                "shell",
+                "web_search",
+                "view_image",
+                "test_server/do_something_cool",
+            ],
+        );
 
         assert_eq!(
-            tools[1],
+            tools[3],
             OpenAiTool::Function(ResponsesApiTool {
                 name: "test_server/do_something_cool".to_string(),
                 parameters: JsonSchema::Object {
@@ -679,16 +775,96 @@ mod tests {
         );
     }
 
+    #[test]
+    fn test_get_openai_tools_mcp_tools_sorted_by_name() {
+        let model_family = find_family_for_model("o3").expect("o3 should be a valid model family");
+        let config = ToolsConfig::new(&ToolsConfigParams {
+            model_family: &model_family,
+            approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::ReadOnly,
+            include_plan_tool: false,
+            include_apply_patch_tool: false,
+            include_web_search_request: false,
+            use_streamable_shell_tool: false,
+            include_view_image_tool: true,
+        });
+
+        // Intentionally construct a map with keys that would sort alphabetically.
+        let tools_map: HashMap<String, mcp_types::Tool> = HashMap::from([
+            (
+                "test_server/do".to_string(),
+                mcp_types::Tool {
+                    name: "a".to_string(),
+                    input_schema: ToolInputSchema {
+                        properties: Some(serde_json::json!({})),
+                        required: None,
+                        r#type: "object".to_string(),
+                    },
+                    output_schema: None,
+                    title: None,
+                    annotations: None,
+                    description: Some("a".to_string()),
+                },
+            ),
+            (
+                "test_server/something".to_string(),
+                mcp_types::Tool {
+                    name: "b".to_string(),
+                    input_schema: ToolInputSchema {
+                        properties: Some(serde_json::json!({})),
+                        required: None,
+                        r#type: "object".to_string(),
+                    },
+                    output_schema: None,
+                    title: None,
+                    annotations: None,
+                    description: Some("b".to_string()),
+                },
+            ),
+            (
+                "test_server/cool".to_string(),
+                mcp_types::Tool {
+                    name: "c".to_string(),
+                    input_schema: ToolInputSchema {
+                        properties: Some(serde_json::json!({})),
+                        required: None,
+                        r#type: "object".to_string(),
+                    },
+                    output_schema: None,
+                    title: None,
+                    annotations: None,
+                    description: Some("c".to_string()),
+                },
+            ),
+        ]);
+
+        let tools = get_openai_tools(&config, Some(tools_map));
+        // Expect shell first, followed by MCP tools sorted by fully-qualified name.
+        assert_eq_tool_names(
+            &tools,
+            &[
+                "shell",
+                "view_image",
+                "test_server/cool",
+                "test_server/do",
+                "test_server/something",
+            ],
+        );
+    }
+
     #[test]
     fn test_mcp_tool_property_missing_type_defaults_to_string() {
         let model_family = find_family_for_model("o3").expect("o3 should be a valid model family");
-        let config = ToolsConfig::new(
-            &model_family,
-            AskForApproval::Never,
-            SandboxPolicy::ReadOnly,
-            false,
-            false,
-        );
+        let config = ToolsConfig::new(&ToolsConfigParams {
+            model_family: &model_family,
+            approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::ReadOnly,
+            include_plan_tool: false,
+            include_apply_patch_tool: false,
+            include_web_search_request: true,
+            use_streamable_shell_tool: false,
+            include_view_image_tool: true,
+        });
 
         let tools = get_openai_tools(
             &config,
@@ -713,10 +889,13 @@ mod tests {
             )])),
         );
 
-        assert_eq_tool_names(&tools, &["shell", "dash/search"]);
+        assert_eq_tool_names(
+            &tools,
+            &["shell", "web_search", "view_image", "dash/search"],
+        );
 
         assert_eq!(
-            tools[1],
+            tools[3],
             OpenAiTool::Function(ResponsesApiTool {
                 name: "dash/search".to_string(),
                 parameters: JsonSchema::Object {
@@ -738,13 +917,16 @@ mod tests {
     #[test]
     fn test_mcp_tool_integer_normalized_to_number() {
         let model_family = find_family_for_model("o3").expect("o3 should be a valid model family");
-        let config = ToolsConfig::new(
-            &model_family,
-            AskForApproval::Never,
-            SandboxPolicy::ReadOnly,
-            false,
-            false,
-        );
+        let config = ToolsConfig::new(&ToolsConfigParams {
+            model_family: &model_family,
+            approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::ReadOnly,
+            include_plan_tool: false,
+            include_apply_patch_tool: false,
+            include_web_search_request: true,
+            use_streamable_shell_tool: false,
+            include_view_image_tool: true,
+        });
 
         let tools = get_openai_tools(
             &config,
@@ -767,9 +949,12 @@ mod tests {
             )])),
         );
 
-        assert_eq_tool_names(&tools, &["shell", "dash/paginate"]);
+        assert_eq_tool_names(
+            &tools,
+            &["shell", "web_search", "view_image", "dash/paginate"],
+        );
         assert_eq!(
-            tools[1],
+            tools[3],
             OpenAiTool::Function(ResponsesApiTool {
                 name: "dash/paginate".to_string(),
                 parameters: JsonSchema::Object {
@@ -789,13 +974,16 @@ mod tests {
     #[test]
     fn test_mcp_tool_array_without_items_gets_default_string_items() {
         let model_family = find_family_for_model("o3").expect("o3 should be a valid model family");
-        let config = ToolsConfig::new(
-            &model_family,
-            AskForApproval::Never,
-            SandboxPolicy::ReadOnly,
-            false,
-            false,
-        );
+        let config = ToolsConfig::new(&ToolsConfigParams {
+            model_family: &model_family,
+            approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::ReadOnly,
+            include_plan_tool: false,
+            include_apply_patch_tool: false,
+            include_web_search_request: true,
+            use_streamable_shell_tool: false,
+            include_view_image_tool: true,
+        });
 
         let tools = get_openai_tools(
             &config,
@@ -818,9 +1006,9 @@ mod tests {
             )])),
         );
 
-        assert_eq_tool_names(&tools, &["shell", "dash/tags"]);
+        assert_eq_tool_names(&tools, &["shell", "web_search", "view_image", "dash/tags"]);
         assert_eq!(
-            tools[1],
+            tools[3],
             OpenAiTool::Function(ResponsesApiTool {
                 name: "dash/tags".to_string(),
                 parameters: JsonSchema::Object {
@@ -843,13 +1031,16 @@ mod tests {
     #[test]
     fn test_mcp_tool_anyof_defaults_to_string() {
         let model_family = find_family_for_model("o3").expect("o3 should be a valid model family");
-        let config = ToolsConfig::new(
-            &model_family,
-            AskForApproval::Never,
-            SandboxPolicy::ReadOnly,
-            false,
-            false,
-        );
+        let config = ToolsConfig::new(&ToolsConfigParams {
+            model_family: &model_family,
+            approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::ReadOnly,
+            include_plan_tool: false,
+            include_apply_patch_tool: false,
+            include_web_search_request: true,
+            use_streamable_shell_tool: false,
+            include_view_image_tool: true,
+        });
 
         let tools = get_openai_tools(
             &config,
@@ -872,9 +1063,9 @@ mod tests {
             )])),
         );
 
-        assert_eq_tool_names(&tools, &["shell", "dash/value"]);
+        assert_eq_tool_names(&tools, &["shell", "web_search", "view_image", "dash/value"]);
         assert_eq!(
-            tools[1],
+            tools[3],
             OpenAiTool::Function(ResponsesApiTool {
                 name: "dash/value".to_string(),
                 parameters: JsonSchema::Object {

codex-rs/core/src/rollout.rs

@@ -135,7 +135,7 @@ impl RolloutRecorder {
                 | ResponseItem::CustomToolCall { .. }
                 | ResponseItem::CustomToolCallOutput { .. }
                 | ResponseItem::Reasoning { .. } => filtered.push(item.clone()),
-                ResponseItem::Other => {
+                ResponseItem::WebSearchCall { .. } | ResponseItem::Other => {
                     // These should never be serialized.
                     continue;
                 }
@@ -199,7 +199,7 @@ impl RolloutRecorder {
                     | ResponseItem::CustomToolCall { .. }
                     | ResponseItem::CustomToolCallOutput { .. }
                     | ResponseItem::Reasoning { .. } => items.push(item),
-                    ResponseItem::Other => {}
+                    ResponseItem::WebSearchCall { .. } | ResponseItem::Other => {}
                 },
                 Err(e) => {
                     warn!("failed to parse item: {v:?}, error: {e}");
@@ -326,7 +326,7 @@ async fn rollout_writer(
                         | ResponseItem::Reasoning { .. } => {
                             writer.write_line(&item).await?;
                         }
-                        ResponseItem::Other => {}
+                        ResponseItem::WebSearchCall { .. } | ResponseItem::Other => {}
                     }
                 }
             }

codex-rs/core/tests/all.rs

@@ -0,0 +1,3 @@
+// Single integration test binary that aggregates all test modules.
+// The submodules live in `tests/all/`.
+mod suite;

codex-rs/core/tests/suite/client.rs

@@ -12,6 +12,7 @@ use codex_login::CodexAuth;
 use core_test_support::load_default_config_for_test;
 use core_test_support::load_sse_fixture_with_id;
 use core_test_support::wait_for_event;
+use serde_json::json;
 use tempfile::TempDir;
 use wiremock::Mock;
 use wiremock::MockServer;
@@ -66,7 +67,6 @@ fn write_auth_json(
     account_id: Option<&str>,
 ) -> String {
     use base64::Engine as _;
-    use serde_json::json;
 
     let header = json!({ "alg": "none", "typ": "JWT" });
     let payload = json!({
@@ -418,7 +418,7 @@ async fn prefers_chatgpt_token_when_config_prefers_chatgpt() {
         match CodexAuth::from_codex_home(codex_home.path(), config.preferred_auth_method) {
             Ok(Some(auth)) => codex_login::AuthManager::from_auth_for_testing(auth),
             Ok(None) => panic!("No CodexAuth found in codex_home"),
-            Err(e) => panic!("Failed to load CodexAuth: {}", e),
+            Err(e) => panic!("Failed to load CodexAuth: {e}"),
         };
     let conversation_manager = ConversationManager::new(auth_manager);
     let NewConversation {
@@ -499,7 +499,7 @@ async fn prefers_apikey_when_config_prefers_apikey_even_with_chatgpt_tokens() {
         match CodexAuth::from_codex_home(codex_home.path(), config.preferred_auth_method) {
             Ok(Some(auth)) => codex_login::AuthManager::from_auth_for_testing(auth),
             Ok(None) => panic!("No CodexAuth found in codex_home"),
-            Err(e) => panic!("Failed to load CodexAuth: {}", e),
+            Err(e) => panic!("Failed to load CodexAuth: {e}"),
         };
     let conversation_manager = ConversationManager::new(auth_manager);
     let NewConversation {
@@ -746,3 +746,151 @@ async fn env_var_overrides_loaded_auth() {
 fn create_dummy_codex_auth() -> CodexAuth {
     CodexAuth::create_dummy_chatgpt_auth_for_testing()
 }
+
+/// Scenario:
+/// - Turn 1: user sends U1; model streams deltas then a final assistant message A.
+/// - Turn 2: user sends U2; model streams a delta then the same final assistant message A.
+/// - Turn 3: user sends U3; model responds (same SSE again, not important).
+///
+/// We assert that the `input` sent on each turn contains the expected conversation history
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn history_dedupes_streamed_and_final_messages_across_turns() {
+    // Skip under Codex sandbox network restrictions (mirrors other tests).
+    if std::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;
+    }
+
+    // Mock server that will receive three sequential requests and return the same SSE stream
+    // each time: a few deltas, then a final assistant message, then completed.
+    let server = MockServer::start().await;
+
+    // Build a small SSE stream with deltas and a final assistant message.
+    // We emit the same body for all 3 turns; ids vary but are unused by assertions.
+    let sse_raw = r##"[
+        {"type":"response.output_text.delta", "delta":"Hey "},
+        {"type":"response.output_text.delta", "delta":"there"},
+        {"type":"response.output_text.delta", "delta":"!\n"},
+        {"type":"response.output_item.done", "item":{
+            "type":"message", "role":"assistant",
+            "content":[{"type":"output_text","text":"Hey there!\n"}]
+        }},
+        {"type":"response.completed", "response": {"id": "__ID__"}}
+    ]"##;
+    let sse1 = core_test_support::load_sse_fixture_with_id_from_str(sse_raw, "resp1");
+
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .respond_with(
+            ResponseTemplate::new(200)
+                .insert_header("content-type", "text/event-stream")
+                .set_body_raw(sse1.clone(), "text/event-stream"),
+        )
+        .expect(3) // respond identically to the three sequential turns
+        .mount(&server)
+        .await;
+
+    // Configure provider to point to mock server (Responses API) and use API key auth.
+    let model_provider = ModelProviderInfo {
+        base_url: Some(format!("{}/v1", server.uri())),
+        ..built_in_model_providers()["openai"].clone()
+    };
+
+    // Init session with isolated codex home.
+    let codex_home = TempDir::new().unwrap();
+    let mut config = load_default_config_for_test(&codex_home);
+    config.model_provider = model_provider;
+
+    let conversation_manager =
+        ConversationManager::with_auth(CodexAuth::from_api_key("Test API Key"));
+    let NewConversation {
+        conversation: codex,
+        ..
+    } = conversation_manager
+        .new_conversation(config)
+        .await
+        .expect("create new conversation");
+
+    // Turn 1: user sends U1; wait for completion.
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text { text: "U1".into() }],
+        })
+        .await
+        .unwrap();
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    // Turn 2: user sends U2; wait for completion.
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text { text: "U2".into() }],
+        })
+        .await
+        .unwrap();
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    // Turn 3: user sends U3; wait for completion.
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text { text: "U3".into() }],
+        })
+        .await
+        .unwrap();
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    // Inspect the three captured requests.
+    let requests = server.received_requests().await.unwrap();
+    assert_eq!(requests.len(), 3, "expected 3 requests (one per turn)");
+
+    // Replace full-array compare with tail-only raw JSON compare using a single hard-coded value.
+    let r3_tail_expected = serde_json::json!([
+        {
+            "type": "message",
+            "id": null,
+            "role": "user",
+            "content": [{"type":"input_text","text":"U1"}]
+        },
+        {
+            "type": "message",
+            "id": null,
+            "role": "assistant",
+            "content": [{"type":"output_text","text":"Hey there!\n"}]
+        },
+        {
+            "type": "message",
+            "id": null,
+            "role": "user",
+            "content": [{"type":"input_text","text":"U2"}]
+        },
+        {
+            "type": "message",
+            "id": null,
+            "role": "assistant",
+            "content": [{"type":"output_text","text":"Hey there!\n"}]
+        },
+        {
+            "type": "message",
+            "id": null,
+            "role": "user",
+            "content": [{"type":"input_text","text":"U3"}]
+        }
+    ]);
+
+    let r3_input_array = requests[2]
+        .body_json::<serde_json::Value>()
+        .unwrap()
+        .get("input")
+        .and_then(|v| v.as_array())
+        .cloned()
+        .expect("r3 missing input array");
+    // skipping earlier context and developer messages
+    let tail_len = r3_tail_expected.as_array().unwrap().len();
+    let actual_tail = &r3_input_array[r3_input_array.len() - tail_len..];
+    assert_eq!(
+        serde_json::Value::Array(actual_tail.to_vec()),
+        r3_tail_expected,
+        "request 3 tail mismatch",
+    );
+}

codex-rs/core/tests/suite/exec.rs

@@ -70,12 +70,12 @@ async fn truncates_output_lines() {
 
     let output = run_test_cmd(tmp, cmd).await.unwrap();
 
-    let expected_output = (1..=256)
+    let expected_output = (1..=300)
         .map(|i| format!("{i}\n"))
         .collect::<Vec<_>>()
         .join("");
     assert_eq!(output.stdout.text, expected_output);
-    assert_eq!(output.stdout.truncated_after_lines, Some(256));
+    assert_eq!(output.stdout.truncated_after_lines, None);
 }
 
 /// Command succeeds with exit code 0 normally
@@ -91,8 +91,8 @@ async fn truncates_output_bytes() {
 
     let output = run_test_cmd(tmp, cmd).await.unwrap();
 
-    assert_eq!(output.stdout.text.len(), 10240);
-    assert_eq!(output.stdout.truncated_after_lines, Some(10));
+    assert!(output.stdout.text.len() >= 15000);
+    assert_eq!(output.stdout.truncated_after_lines, None);
 }
 
 /// Command not found returns exit code 127, this is not considered a sandbox error

codex-rs/core/tests/suite/exec_stream_events.rs

@@ -139,3 +139,34 @@ async fn test_exec_stderr_stream_events_echo() {
     }
     assert_eq!(String::from_utf8_lossy(&err), "oops\n");
 }
+
+#[tokio::test]
+async fn test_aggregated_output_interleaves_in_order() {
+    // Spawn a shell that alternates stdout and stderr with sleeps to enforce order.
+    let cmd = vec![
+        "/bin/sh".to_string(),
+        "-c".to_string(),
+        "printf 'O1\\n'; sleep 0.01; printf 'E1\\n' 1>&2; sleep 0.01; printf 'O2\\n'; sleep 0.01; printf 'E2\\n' 1>&2".to_string(),
+    ];
+
+    let params = ExecParams {
+        command: cmd,
+        cwd: std::env::current_dir().unwrap_or_else(|_| PathBuf::from(".")),
+        timeout_ms: Some(5_000),
+        env: HashMap::new(),
+        with_escalated_permissions: None,
+        justification: None,
+    };
+
+    let policy = SandboxPolicy::new_read_only_policy();
+
+    let result = process_exec_tool_call(params, SandboxType::None, &policy, &None, None)
+        .await
+        .expect("process_exec_tool_call");
+
+    assert_eq!(result.exit_code, 0);
+    assert_eq!(result.stdout.text, "O1\nO2\n");
+    assert_eq!(result.stderr.text, "E1\nE2\n");
+    assert_eq!(result.aggregated_output.text, "O1\nE1\nO2\nE2\n");
+    assert_eq!(result.aggregated_output.truncated_after_lines, None);
+}

codex-rs/core/tests/suite/mod.rs

@@ -0,0 +1,12 @@
+// Aggregates all former standalone integration tests as modules.
+
+mod cli_stream;
+mod client;
+mod compact;
+mod exec;
+mod exec_stream_events;
+mod live_cli;
+mod prompt_caching;
+mod seatbelt;
+mod stream_error_allows_next_turn;
+mod stream_no_completed;

codex-rs/core/tests/suite/prompt_caching.rs

@@ -107,8 +107,8 @@ async fn codex_mini_latest_tools() {
     assert_eq!(requests.len(), 2, "expected two POST requests");
 
     let expected_instructions = [
-        include_str!("../prompt.md"),
-        include_str!("../../apply-patch/apply_patch_tool_instructions.md"),
+        include_str!("../../prompt.md"),
+        include_str!("../../../apply-patch/apply_patch_tool_instructions.md"),
     ]
     .join("\n");
 
@@ -188,10 +188,10 @@ async fn prompt_tools_are_consistent_across_requests() {
     let requests = server.received_requests().await.unwrap();
     assert_eq!(requests.len(), 2, "expected two POST requests");
 
-    let expected_instructions: &str = include_str!("../prompt.md");
+    let expected_instructions: &str = include_str!("../../prompt.md");
     // our internal implementation is responsible for keeping tools in sync
     // with the OpenAI schema, so we just verify the tool presence here
-    let expected_tools_names: &[&str] = &["shell", "update_plan", "apply_patch"];
+    let expected_tools_names: &[&str] = &["shell", "update_plan", "apply_patch", "view_image"];
     let body0 = requests[0].body_json::<serde_json::Value>().unwrap();
     assert_eq!(
         body0["instructions"],
@@ -280,7 +280,7 @@ async fn prefixes_context_and_instructions_once_and_consistently_across_requests
 {}</environment_context>"#,
         cwd.path().to_string_lossy(),
         match shell.name() {
-            Some(name) => format!("  <shell>{}</shell>\n", name),
+            Some(name) => format!("  <shell>{name}</shell>\n"),
             None => String::new(),
         }
     );

codex-rs/exec/src/event_processor_with_human_output.rs

@@ -24,6 +24,8 @@ use codex_core::protocol::StreamErrorEvent;
 use codex_core::protocol::TaskCompleteEvent;
 use codex_core::protocol::TurnAbortReason;
 use codex_core::protocol::TurnDiffEvent;
+use codex_core::protocol::WebSearchBeginEvent;
+use codex_core::protocol::WebSearchEndEvent;
 use owo_colors::OwoColorize;
 use owo_colors::Style;
 use shlex::try_join;
@@ -178,7 +180,7 @@ impl EventProcessor for EventProcessorWithHumanOutput {
             EventMsg::StreamError(StreamErrorEvent { message }) => {
                 ts_println!(self, "{}", message.style(self.dimmed));
             }
-            EventMsg::TaskStarted => {
+            EventMsg::TaskStarted(_) => {
                 // Ignore.
             }
             EventMsg::TaskComplete(TaskCompleteEvent { last_agent_message }) => {
@@ -287,8 +289,7 @@ impl EventProcessor for EventProcessorWithHumanOutput {
             EventMsg::ExecCommandOutputDelta(_) => {}
             EventMsg::ExecCommandEnd(ExecCommandEndEvent {
                 call_id,
-                stdout,
-                stderr,
+                aggregated_output,
                 duration,
                 exit_code,
                 ..
@@ -304,8 +305,7 @@ impl EventProcessor for EventProcessorWithHumanOutput {
                     ("".to_string(), format!("exec('{call_id}')"))
                 };
 
-                let output = if exit_code == 0 { stdout } else { stderr };
-                let truncated_output = output
+                let truncated_output = aggregated_output
                     .lines()
                     .take(MAX_OUTPUT_LINES_FOR_EXEC_TOOL_CALL)
                     .collect::<Vec<_>>()
@@ -363,6 +363,10 @@ impl EventProcessor for EventProcessorWithHumanOutput {
                     }
                 }
             }
+            EventMsg::WebSearchBegin(WebSearchBeginEvent { call_id: _ }) => {}
+            EventMsg::WebSearchEnd(WebSearchEndEvent { call_id: _, query }) => {
+                ts_println!(self, "🌐 Searched: {query}");
+            }
             EventMsg::PatchApplyBegin(PatchApplyBeginEvent {
                 call_id,
                 auto_approved,
@@ -531,6 +535,9 @@ impl EventProcessor for EventProcessorWithHumanOutput {
             EventMsg::McpListToolsResponse(_) => {
                 // Currently ignored in exec output.
             }
+            EventMsg::ListCustomPromptsResponse(_) => {
+                // Currently ignored in exec output.
+            }
             EventMsg::TurnAborted(abort_reason) => match abort_reason.reason {
                 TurnAbortReason::Interrupted => {
                     ts_println!(self, "task interrupted");

codex-rs/exec/src/lib.rs

@@ -148,8 +148,10 @@ pub async fn run_main(cli: Cli, codex_linux_sandbox_exe: Option<PathBuf>) -> any
         base_instructions: None,
         include_plan_tool: None,
         include_apply_patch_tool: None,
+        include_view_image_tool: None,
         disable_response_storage: oss.then_some(true),
         show_raw_agent_reasoning: oss.then_some(true),
+        tools_web_search_request: None,
     };
     // Parse `-c` overrides.
     let cli_kv_overrides = match config_overrides.parse_overrides() {

codex-rs/exec/tests/all.rs

@@ -0,0 +1,3 @@
+// Single integration test binary that aggregates all test modules.
+// The submodules live in `tests/suite/`.
+mod suite;

codex-rs/exec/tests/apply_patch.rs

@@ -1,339 +0,0 @@
-#![allow(clippy::expect_used, clippy::unwrap_used)]
-
-use anyhow::Context;
-use assert_cmd::prelude::*;
-use codex_core::CODEX_APPLY_PATCH_ARG1;
-use std::fs;
-use std::process::Command;
-use tempfile::tempdir;
-
-/// While we may add an `apply-patch` subcommand to the `codex` CLI multitool
-/// at some point, we must ensure that the smaller `codex-exec` CLI can still
-/// emulate the `apply_patch` CLI.
-#[test]
-fn test_standalone_exec_cli_can_use_apply_patch() -> anyhow::Result<()> {
-    let tmp = tempdir()?;
-    let relative_path = "source.txt";
-    let absolute_path = tmp.path().join(relative_path);
-    fs::write(&absolute_path, "original content\n")?;
-
-    Command::cargo_bin("codex-exec")
-        .context("should find binary for codex-exec")?
-        .arg(CODEX_APPLY_PATCH_ARG1)
-        .arg(
-            r#"*** Begin Patch
-*** Update File: source.txt
-@@
--original content
-+modified by apply_patch
-*** End Patch"#,
-        )
-        .current_dir(tmp.path())
-        .assert()
-        .success()
-        .stdout("Success. Updated the following files:\nM source.txt\n")
-        .stderr(predicates::str::is_empty());
-    assert_eq!(
-        fs::read_to_string(absolute_path)?,
-        "modified by apply_patch\n"
-    );
-    Ok(())
-}
-
-#[cfg(not(target_os = "windows"))]
-#[tokio::test]
-async fn test_apply_patch_tool() -> anyhow::Result<()> {
-    use core_test_support::load_sse_fixture_with_id_from_str;
-    use tempfile::TempDir;
-    use wiremock::Mock;
-    use wiremock::MockServer;
-    use wiremock::ResponseTemplate;
-    use wiremock::matchers::method;
-    use wiremock::matchers::path;
-
-    const SSE_TOOL_CALL_ADD: &str = r#"[
-  {
-    "type": "response.output_item.done",
-    "item": {
-      "type": "function_call",
-      "name": "apply_patch",
-      "arguments": "{\n  \"input\": \"*** Begin Patch\\n*** Add File: test.md\\n+Hello world\\n*** End Patch\"\n}",
-      "call_id": "__ID__"
-    }
-  },
-  {
-    "type": "response.completed",
-    "response": {
-      "id": "__ID__",
-      "usage": {
-        "input_tokens": 0,
-        "input_tokens_details": null,
-        "output_tokens": 0,
-        "output_tokens_details": null,
-        "total_tokens": 0
-      },
-      "output": []
-    }
-  }
-]"#;
-
-    const SSE_TOOL_CALL_UPDATE: &str = r#"[
-  {
-    "type": "response.output_item.done",
-    "item": {
-      "type": "function_call",
-      "name": "apply_patch",
-      "arguments": "{\n  \"input\": \"*** Begin Patch\\n*** Update File: test.md\\n@@\\n-Hello world\\n+Final text\\n*** End Patch\"\n}",
-      "call_id": "__ID__"
-    }
-  },
-  {
-    "type": "response.completed",
-    "response": {
-      "id": "__ID__",
-      "usage": {
-        "input_tokens": 0,
-        "input_tokens_details": null,
-        "output_tokens": 0,
-        "output_tokens_details": null,
-        "total_tokens": 0
-      },
-      "output": []
-    }
-  }
-]"#;
-
-    const SSE_TOOL_CALL_COMPLETED: &str = r#"[
-  {
-    "type": "response.completed",
-    "response": {
-      "id": "__ID__",
-      "usage": {
-        "input_tokens": 0,
-        "input_tokens_details": null,
-        "output_tokens": 0,
-        "output_tokens_details": null,
-        "total_tokens": 0
-      },
-      "output": []
-    }
-  }
-]"#;
-
-    // Start a mock model server
-    let server = MockServer::start().await;
-
-    // First response: model calls apply_patch to create test.md
-    let first = ResponseTemplate::new(200)
-        .insert_header("content-type", "text/event-stream")
-        .set_body_raw(
-            load_sse_fixture_with_id_from_str(SSE_TOOL_CALL_ADD, "call1"),
-            "text/event-stream",
-        );
-
-    Mock::given(method("POST"))
-        // .and(path("/v1/responses"))
-        .respond_with(first)
-        .up_to_n_times(1)
-        .mount(&server)
-        .await;
-
-    // Second response: model calls apply_patch to update test.md
-    let second = ResponseTemplate::new(200)
-        .insert_header("content-type", "text/event-stream")
-        .set_body_raw(
-            load_sse_fixture_with_id_from_str(SSE_TOOL_CALL_UPDATE, "call2"),
-            "text/event-stream",
-        );
-
-    Mock::given(method("POST"))
-        .and(path("/v1/responses"))
-        .respond_with(second)
-        .up_to_n_times(1)
-        .mount(&server)
-        .await;
-
-    let final_completed = ResponseTemplate::new(200)
-        .insert_header("content-type", "text/event-stream")
-        .set_body_raw(
-            load_sse_fixture_with_id_from_str(SSE_TOOL_CALL_COMPLETED, "resp3"),
-            "text/event-stream",
-        );
-
-    Mock::given(method("POST"))
-        .and(path("/v1/responses"))
-        .respond_with(final_completed)
-        .expect(1)
-        .mount(&server)
-        .await;
-
-    let tmp_cwd = TempDir::new().unwrap();
-    Command::cargo_bin("codex-exec")
-        .context("should find binary for codex-exec")?
-        .current_dir(tmp_cwd.path())
-        .env("CODEX_HOME", tmp_cwd.path())
-        .env("OPENAI_API_KEY", "dummy")
-        .env("OPENAI_BASE_URL", format!("{}/v1", server.uri()))
-        .arg("--skip-git-repo-check")
-        .arg("-s")
-        .arg("workspace-write")
-        .arg("foo")
-        .assert()
-        .success();
-
-    // Verify final file contents
-    let final_path = tmp_cwd.path().join("test.md");
-    let contents = std::fs::read_to_string(&final_path)
-        .unwrap_or_else(|e| panic!("failed reading {}: {e}", final_path.display()));
-    assert_eq!(contents, "Final text\n");
-    Ok(())
-}
-
-#[cfg(not(target_os = "windows"))]
-#[tokio::test]
-async fn test_apply_patch_freeform_tool() -> anyhow::Result<()> {
-    use core_test_support::load_sse_fixture_with_id_from_str;
-    use tempfile::TempDir;
-    use wiremock::Mock;
-    use wiremock::MockServer;
-    use wiremock::ResponseTemplate;
-    use wiremock::matchers::method;
-    use wiremock::matchers::path;
-
-    const SSE_TOOL_CALL_ADD: &str = r#"[
-  {
-    "type": "response.output_item.done",
-    "item": {
-      "type": "custom_tool_call",
-      "name": "apply_patch",
-      "input": "*** Begin Patch\n*** Add File: test.md\n+Hello world\n*** End Patch",
-      "call_id": "__ID__"
-    }
-  },
-  {
-    "type": "response.completed",
-    "response": {
-      "id": "__ID__",
-      "usage": {
-        "input_tokens": 0,
-        "input_tokens_details": null,
-        "output_tokens": 0,
-        "output_tokens_details": null,
-        "total_tokens": 0
-      },
-      "output": []
-    }
-  }
-]"#;
-
-    const SSE_TOOL_CALL_UPDATE: &str = r#"[
-  {
-    "type": "response.output_item.done",
-    "item": {
-      "type": "custom_tool_call",
-      "name": "apply_patch",
-      "input": "*** Begin Patch\n*** Update File: test.md\n@@\n-Hello world\n+Final text\n*** End Patch",
-      "call_id": "__ID__"
-    }
-  },
-  {
-    "type": "response.completed",
-    "response": {
-      "id": "__ID__",
-      "usage": {
-        "input_tokens": 0,
-        "input_tokens_details": null,
-        "output_tokens": 0,
-        "output_tokens_details": null,
-        "total_tokens": 0
-      },
-      "output": []
-    }
-  }
-]"#;
-
-    const SSE_TOOL_CALL_COMPLETED: &str = r#"[
-  {
-    "type": "response.completed",
-    "response": {
-      "id": "__ID__",
-      "usage": {
-        "input_tokens": 0,
-        "input_tokens_details": null,
-        "output_tokens": 0,
-        "output_tokens_details": null,
-        "total_tokens": 0
-      },
-      "output": []
-    }
-  }
-]"#;
-
-    // Start a mock model server
-    let server = MockServer::start().await;
-
-    // First response: model calls apply_patch to create test.md
-    let first = ResponseTemplate::new(200)
-        .insert_header("content-type", "text/event-stream")
-        .set_body_raw(
-            load_sse_fixture_with_id_from_str(SSE_TOOL_CALL_ADD, "call1"),
-            "text/event-stream",
-        );
-
-    Mock::given(method("POST"))
-        // .and(path("/v1/responses"))
-        .respond_with(first)
-        .up_to_n_times(1)
-        .mount(&server)
-        .await;
-
-    // Second response: model calls apply_patch to update test.md
-    let second = ResponseTemplate::new(200)
-        .insert_header("content-type", "text/event-stream")
-        .set_body_raw(
-            load_sse_fixture_with_id_from_str(SSE_TOOL_CALL_UPDATE, "call2"),
-            "text/event-stream",
-        );
-
-    Mock::given(method("POST"))
-        .and(path("/v1/responses"))
-        .respond_with(second)
-        .up_to_n_times(1)
-        .mount(&server)
-        .await;
-
-    let final_completed = ResponseTemplate::new(200)
-        .insert_header("content-type", "text/event-stream")
-        .set_body_raw(
-            load_sse_fixture_with_id_from_str(SSE_TOOL_CALL_COMPLETED, "resp3"),
-            "text/event-stream",
-        );
-
-    Mock::given(method("POST"))
-        // .and(path("/v1/responses"))
-        .respond_with(final_completed)
-        .expect(1)
-        .mount(&server)
-        .await;
-
-    let tmp_cwd = TempDir::new().unwrap();
-    Command::cargo_bin("codex-exec")
-        .context("should find binary for codex-exec")?
-        .current_dir(tmp_cwd.path())
-        .env("CODEX_HOME", tmp_cwd.path())
-        .env("OPENAI_API_KEY", "dummy")
-        .env("OPENAI_BASE_URL", format!("{}/v1", server.uri()))
-        .arg("--skip-git-repo-check")
-        .arg("-s")
-        .arg("workspace-write")
-        .arg("foo")
-        .assert()
-        .success();
-
-    // Verify final file contents
-    let final_path = tmp_cwd.path().join("test.md");
-    let contents = std::fs::read_to_string(&final_path)
-        .unwrap_or_else(|e| panic!("failed reading {}: {e}", final_path.display()));
-    assert_eq!(contents, "Final text\n");
-    Ok(())
-}

codex-rs/exec/tests/fixtures/apply_patch_freeform_final.txt

@@ -0,0 +1,4 @@
+class BaseClass:
+  def method():
+
+    return True

codex-rs/exec/tests/fixtures/sse_apply_patch_add.json

@@ -0,0 +1,25 @@
+[
+  {
+    "type": "response.output_item.done",
+    "item": {
+      "type": "custom_tool_call",
+      "name": "apply_patch",
+      "input": "*** Begin Patch\n*** Add File: test.md\n+Hello world\n*** End Patch",
+      "call_id": "__ID__"
+    }
+  },
+  {
+    "type": "response.completed",
+    "response": {
+      "id": "__ID__",
+      "usage": {
+        "input_tokens": 0,
+        "input_tokens_details": null,
+        "output_tokens": 0,
+        "output_tokens_details": null,
+        "total_tokens": 0
+      },
+      "output": []
+    }
+  }
+]

codex-rs/exec/tests/fixtures/sse_apply_patch_freeform_add.json

@@ -0,0 +1,25 @@
+[
+  {
+    "type": "response.output_item.done",
+    "item": {
+      "type": "custom_tool_call",
+      "name": "apply_patch",
+      "input": "*** Begin Patch\n*** Add File: app.py\n+class BaseClass:\n+  def method():\n+    return False\n*** End Patch",
+      "call_id": "__ID__"
+    }
+  },
+  {
+    "type": "response.completed",
+    "response": {
+      "id": "__ID__",
+      "usage": {
+        "input_tokens": 0,
+        "input_tokens_details": null,
+        "output_tokens": 0,
+        "output_tokens_details": null,
+        "total_tokens": 0
+      },
+      "output": []
+    }
+  }
+]

codex-rs/exec/tests/fixtures/sse_apply_patch_freeform_update.json

@@ -0,0 +1,25 @@
+[
+  {
+    "type": "response.output_item.done",
+    "item": {
+      "type": "custom_tool_call",
+      "name": "apply_patch",
+      "input": "*** Begin Patch\n*** Update File: app.py\n@@  def method():\n-    return False\n+\n+    return True\n*** End Patch",
+      "call_id": "__ID__"
+    }
+  },
+  {
+    "type": "response.completed",
+    "response": {
+      "id": "__ID__",
+      "usage": {
+        "input_tokens": 0,
+        "input_tokens_details": null,
+        "output_tokens": 0,
+        "output_tokens_details": null,
+        "total_tokens": 0
+      },
+      "output": []
+    }
+  }
+]

codex-rs/exec/tests/fixtures/sse_apply_patch_update.json

@@ -0,0 +1,25 @@
+[
+  {
+    "type": "response.output_item.done",
+    "item": {
+      "type": "function_call",
+      "name": "apply_patch",
+      "arguments": "{\n  \"input\": \"*** Begin Patch\\n*** Update File: test.md\\n@@\\n-Hello world\\n+Final text\\n*** End Patch\"\n}",
+      "call_id": "__ID__"
+    }
+  },
+  {
+    "type": "response.completed",
+    "response": {
+      "id": "__ID__",
+      "usage": {
+        "input_tokens": 0,
+        "input_tokens_details": null,
+        "output_tokens": 0,
+        "output_tokens_details": null,
+        "total_tokens": 0
+      },
+      "output": []
+    }
+  }
+]

codex-rs/exec/tests/fixtures/sse_response_completed.json

@@ -0,0 +1,16 @@
+[
+  {
+    "type": "response.completed",
+    "response": {
+      "id": "__ID__",
+      "usage": {
+        "input_tokens": 0,
+        "input_tokens_details": null,
+        "output_tokens": 0,
+        "output_tokens_details": null,
+        "total_tokens": 0
+      },
+      "output": []
+    }
+  }
+]

codex-rs/exec/tests/suite/apply_patch.rs

@@ -0,0 +1,108 @@
+#![allow(clippy::expect_used, clippy::unwrap_used)]
+
+use anyhow::Context;
+use assert_cmd::prelude::*;
+use codex_core::CODEX_APPLY_PATCH_ARG1;
+use std::fs;
+use std::process::Command;
+use tempfile::tempdir;
+
+/// While we may add an `apply-patch` subcommand to the `codex` CLI multitool
+/// at some point, we must ensure that the smaller `codex-exec` CLI can still
+/// emulate the `apply_patch` CLI.
+#[test]
+fn test_standalone_exec_cli_can_use_apply_patch() -> anyhow::Result<()> {
+    let tmp = tempdir()?;
+    let relative_path = "source.txt";
+    let absolute_path = tmp.path().join(relative_path);
+    fs::write(&absolute_path, "original content\n")?;
+
+    Command::cargo_bin("codex-exec")
+        .context("should find binary for codex-exec")?
+        .arg(CODEX_APPLY_PATCH_ARG1)
+        .arg(
+            r#"*** Begin Patch
+*** Update File: source.txt
+@@
+-original content
++modified by apply_patch
+*** End Patch"#,
+        )
+        .current_dir(tmp.path())
+        .assert()
+        .success()
+        .stdout("Success. Updated the following files:\nM source.txt\n")
+        .stderr(predicates::str::is_empty());
+    assert_eq!(
+        fs::read_to_string(absolute_path)?,
+        "modified by apply_patch\n"
+    );
+    Ok(())
+}
+
+#[cfg(not(target_os = "windows"))]
+#[tokio::test(flavor = "multi_thread", worker_threads = 4)]
+async fn test_apply_patch_tool() -> anyhow::Result<()> {
+    use crate::suite::common::run_e2e_exec_test;
+    use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
+
+    if std::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 Ok(());
+    }
+
+    let tmp_cwd = tempdir().expect("failed to create temp dir");
+    let tmp_path = tmp_cwd.path().to_path_buf();
+    run_e2e_exec_test(
+        tmp_cwd.path(),
+        vec![
+            include_str!("../fixtures/sse_apply_patch_add.json").to_string(),
+            include_str!("../fixtures/sse_apply_patch_update.json").to_string(),
+            include_str!("../fixtures/sse_response_completed.json").to_string(),
+        ],
+    )
+    .await;
+
+    let final_path = tmp_path.join("test.md");
+    let contents = std::fs::read_to_string(&final_path)
+        .unwrap_or_else(|e| panic!("failed reading {}: {e}", final_path.display()));
+    assert_eq!(contents, "Final text\n");
+    Ok(())
+}
+
+#[cfg(not(target_os = "windows"))]
+#[tokio::test(flavor = "multi_thread", worker_threads = 4)]
+async fn test_apply_patch_freeform_tool() -> anyhow::Result<()> {
+    use crate::suite::common::run_e2e_exec_test;
+    use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
+
+    if std::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 Ok(());
+    }
+
+    let tmp_cwd = tempdir().expect("failed to create temp dir");
+    run_e2e_exec_test(
+        tmp_cwd.path(),
+        vec![
+            include_str!("../fixtures/sse_apply_patch_freeform_add.json").to_string(),
+            include_str!("../fixtures/sse_apply_patch_freeform_update.json").to_string(),
+            include_str!("../fixtures/sse_response_completed.json").to_string(),
+        ],
+    )
+    .await;
+
+    // Verify final file contents
+    let final_path = tmp_cwd.path().join("app.py");
+    let contents = std::fs::read_to_string(&final_path)
+        .unwrap_or_else(|e| panic!("failed reading {}: {e}", final_path.display()));
+    assert_eq!(
+        contents,
+        include_str!("../fixtures/apply_patch_freeform_final.txt")
+    );
+    Ok(())
+}

codex-rs/exec/tests/suite/common.rs

@@ -0,0 +1,73 @@
+// this file is only used for e2e tests which are currently disabled on windows
+#![cfg(not(target_os = "windows"))]
+#![allow(clippy::expect_used)]
+
+use anyhow::Context;
+use assert_cmd::prelude::*;
+use core_test_support::load_sse_fixture_with_id_from_str;
+use std::path::Path;
+use std::process::Command;
+use std::sync::atomic::AtomicUsize;
+use std::sync::atomic::Ordering;
+use wiremock::Mock;
+use wiremock::MockServer;
+use wiremock::matchers::method;
+use wiremock::matchers::path;
+
+use wiremock::Respond;
+
+struct SeqResponder {
+    num_calls: AtomicUsize,
+    responses: Vec<String>,
+}
+
+impl Respond for SeqResponder {
+    fn respond(&self, _: &wiremock::Request) -> wiremock::ResponseTemplate {
+        let call_num = self.num_calls.fetch_add(1, Ordering::SeqCst);
+        match self.responses.get(call_num) {
+            Some(body) => wiremock::ResponseTemplate::new(200)
+                .insert_header("content-type", "text/event-stream")
+                .set_body_raw(
+                    load_sse_fixture_with_id_from_str(body, &format!("request_{call_num}")),
+                    "text/event-stream",
+                ),
+            None => panic!("no response for {call_num}"),
+        }
+    }
+}
+
+/// Helper function to run an E2E test of a codex-exec call. Starts a wiremock
+/// server, and returns the response_streams in order for each api call. Runs
+/// the codex-exec command with the wiremock server as the model server.
+pub(crate) async fn run_e2e_exec_test(cwd: &Path, response_streams: Vec<String>) {
+    let server = MockServer::start().await;
+
+    let num_calls = response_streams.len();
+    let seq_responder = SeqResponder {
+        num_calls: AtomicUsize::new(0),
+        responses: response_streams,
+    };
+
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .respond_with(seq_responder)
+        .expect(num_calls as u64)
+        .mount(&server)
+        .await;
+
+    let cwd = cwd.to_path_buf();
+    let uri = server.uri();
+    Command::cargo_bin("codex-exec")
+        .context("should find binary for codex-exec")
+        .expect("should find binary for codex-exec")
+        .current_dir(cwd.clone())
+        .env("CODEX_HOME", cwd.clone())
+        .env("OPENAI_API_KEY", "dummy")
+        .env("OPENAI_BASE_URL", format!("{uri}/v1"))
+        .arg("--skip-git-repo-check")
+        .arg("-s")
+        .arg("danger-full-access")
+        .arg("foo")
+        .assert()
+        .success();
+}

codex-rs/exec/tests/suite/mod.rs

@@ -0,0 +1,4 @@
+// Aggregates all former standalone integration tests as modules.
+mod apply_patch;
+mod common;
+mod sandbox;