non negative line validation for comment server (#429 )

* enforce non-negative validation for line in GH comment server * include .nonnegative() for startLine too
feat: add flexible bot access control with allowed_bots option (#117 )
2026-01-23 23:14:13 +08:00 · 2025-08-08 08:36:20 -07:00 · 2025-08-07 18:03:20 -07:00 · 2025-08-07 22:57:57 +00:00 · 2025-08-07 08:56:30 -07:00 · 2025-08-06 21:22:15 -07:00
41 changed files with 1832 additions and 1079 deletions
--- a/.github/workflows/claude.yml
+++ b/.github/workflows/claude.yml
@@ -36,4 +36,4 @@ jobs:
          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
          allowed_tools: "Bash(bun install),Bash(bun test:*),Bash(bun run format),Bash(bun typecheck)"
          custom_instructions: "You have also been granted tools for editing files and running bun commands (install, run, test, typecheck) for testing your changes: bun install, bun test, bun run format, bun typecheck."
-          model: "claude-opus-4-20250514"
+          model: "claude-opus-4-1-20250805"
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -53,7 +53,7 @@ Execution steps:
 #### Mode System (`src/modes/`)
 - **Tag Mode** (`tag/`): Responds to `@claude` mentions and issue assignments
- **Agent Mode** (`agent/`): Automated execution without trigger checking
+- **Agent Mode** (`agent/`): Automated execution for workflow_dispatch and schedule events only
 - Extensible registry pattern in `modes/registry.ts`
 #### GitHub Integration (`src/github/`)
@@ -118,7 +118,7 @@ src/
 - Modes implement `Mode` interface with `shouldTrigger()` and `prepare()` methods
 - Registry validates mode compatibility with GitHub event types
- Agent mode bypasses all trigger checking for automation scenarios
+- Agent mode only works with workflow_dispatch and schedule events
 ### Comment Threading
--- a/README.md
+++ b/README.md
@@ -23,963 +23,23 @@ This command will guide you through setting up the GitHub app and required secre
 **Note**:
 - You must be a repository admin to install the GitHub app and add secrets
- This quickstart method is only available for direct Anthropic API users. If you're using AWS Bedrock, please see the instructions below.
+- This quickstart method is only available for direct Anthropic API users. For AWS Bedrock or Google Vertex AI setup, see [docs/cloud-providers.md](./docs/cloud-providers.md).
-### Manual Setup (Direct API)
+## Documentation
-**Requirements**: You must be a repository admin to complete these steps.
+- [Setup Guide](./docs/setup.md) - Manual setup, custom GitHub apps, and security best practices
-
+- [Usage Guide](./docs/usage.md) - Basic usage, workflow configuration, and input parameters
-1. Install the Claude GitHub app to your repository: https://github.com/apps/claude
+- [Custom Automations](./docs/custom-automations.md) - Examples of automated workflows and custom prompts
-2. Add authentication to your repository secrets ([Learn how to use secrets in GitHub Actions](https://docs.github.com/en/actions/security-for-github-actions/security-guides/using-secrets-in-github-actions)):
+- [Configuration](./docs/configuration.md) - MCP servers, permissions, environment variables, and advanced settings
-   - Either `ANTHROPIC_API_KEY` for API key authentication
+- [Experimental Features](./docs/experimental.md) - Execution modes and network restrictions
-   - Or `CLAUDE_CODE_OAUTH_TOKEN` for OAuth token authentication (Pro and Max users can generate this by running `claude setup-token` locally)
+- [Cloud Providers](./docs/cloud-providers.md) - AWS Bedrock and Google Vertex AI setup
-3. Copy the workflow file from [`examples/claude.yml`](./examples/claude.yml) into your repository's `.github/workflows/`
+- [Capabilities & Limitations](./docs/capabilities-and-limitations.md) - What Claude can and cannot do
-
+- [Security](./docs/security.md) - Access control, permissions, and commit signing
-### Using a Custom GitHub App
+- [FAQ](./docs/faq.md) - Common questions and troubleshooting
 If you prefer not to install the official Claude app, you can create your own GitHub App to use with this action. This gives you complete control over permissions and access.
 **When you may want to use a custom GitHub App:**
 - You need more restrictive permissions than the official app
 - Organization policies prevent installing third-party apps
 - You're using AWS Bedrock or Google Vertex AI
 **Steps to create and use a custom GitHub App:**
 1. **Create a new GitHub App:**
   - Go to https://github.com/settings/apps (for personal apps) or your organization's settings
   - Click "New GitHub App"
   - Configure the app with these minimum permissions:
     - **Repository permissions:**
       - Contents: Read & Write
       - Issues: Read & Write
       - Pull requests: Read & Write
     - **Account permissions:** None required
   - Set "Where can this GitHub App be installed?" to your preference
   - Create the app
 2. **Generate and download a private key:**
   - After creating the app, scroll down to "Private keys"
   - Click "Generate a private key"
   - Download the `.pem` file (keep this secure!)
 3. **Install the app on your repository:**
   - Go to the app's settings page
   - Click "Install App"
   - Select the repositories where you want to use Claude
 4. **Add the app credentials to your repository secrets:**
   - Go to your repository's Settings → Secrets and variables → Actions
   - Add these secrets:
     - `APP_ID`: Your GitHub App's ID (found in the app settings)
     - `APP_PRIVATE_KEY`: The contents of the downloaded `.pem` file
 5. **Update your workflow to use the custom app:**
   ```yaml
   name: Claude with Custom App
   on:
     issue_comment:
       types: [created]
     # ... other triggers
   jobs:
     claude-response:
       runs-on: ubuntu-latest
       steps:
         # Generate a token from your custom app
         - name: Generate GitHub App token
           id: app-token
           uses: actions/create-github-app-token@v1
           with:
             app-id: ${{ secrets.APP_ID }}
             private-key: ${{ secrets.APP_PRIVATE_KEY }}
         # Use Claude with your custom app's token
         - uses: anthropics/claude-code-action@beta
           with:
             anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
             github_token: ${{ steps.app-token.outputs.token }}
             # ... other configuration
   ```
 **Important notes:**
 - The custom app must have read/write permissions for Issues, Pull Requests, and Contents
 - Your app's token will have the exact permissions you configured, nothing more
 For more information on creating GitHub Apps, see the [GitHub documentation](https://docs.github.com/en/apps/creating-github-apps).
 ## 📚 FAQ
-Having issues or questions? Check out our [Frequently Asked Questions](./FAQ.md) for solutions to common problems and detailed explanations of Claude's capabilities and limitations.
+Having issues or questions? Check out our [Frequently Asked Questions](./docs/faq.md) for solutions to common problems and detailed explanations of Claude's capabilities and limitations.
 ## Usage
 Add a workflow file to your repository (e.g., `.github/workflows/claude.yml`):
 ```yaml
 name: Claude Assistant
 on:
  issue_comment:
    types: [created]
  pull_request_review_comment:
    types: [created]
  issues:
    types: [opened, assigned, labeled]
  pull_request_review:
    types: [submitted]
 jobs:
  claude-response:
    runs-on: ubuntu-latest
    steps:
      - uses: anthropics/claude-code-action@beta
        with:
          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
          # Or use OAuth token instead:
          # claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
          github_token: ${{ secrets.GITHUB_TOKEN }}
          # Optional: set execution mode (default: tag)
          # mode: "tag"
          # Optional: add custom trigger phrase (default: @claude)
          # trigger_phrase: "/claude"
          # Optional: add assignee trigger for issues
          # assignee_trigger: "claude"
          # Optional: add label trigger for issues
          # label_trigger: "claude"
          # Optional: add custom environment variables (YAML format)
          # claude_env: |
          #   NODE_ENV: test
          #   DEBUG: true
          #   API_URL: https://api.example.com
          # Optional: limit the number of conversation turns
          # max_turns: "5"
          # Optional: grant additional permissions (requires corresponding GitHub token permissions)
          # additional_permissions: |
          #   actions: read
 ```
 ## Inputs
 | Input                          | Description                                                                                                                           | Required | Default   |
 | ------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------- | -------- | --------- |
 | `mode`                         | Execution mode: 'tag' (default - triggered by mentions/assignments), 'agent' (for automation), 'experimental-review' (for PR reviews) | No       | `tag`     |
 | `anthropic_api_key`            | Anthropic API key (required for direct API, not needed for Bedrock/Vertex)                                                            | No\*     | -         |
 | `claude_code_oauth_token`      | Claude Code OAuth token (alternative to anthropic_api_key)                                                                            | No\*     | -         |
 | `direct_prompt`                | Direct prompt for Claude to execute automatically without needing a trigger (for automated workflows)                                 | No       | -         |
 | `override_prompt`              | Complete replacement of Claude's prompt with custom template (supports variable substitution)                                         | No       | -         |
 | `base_branch`                  | The base branch to use for creating new branches (e.g., 'main', 'develop')                                                            | No       | -         |
 | `max_turns`                    | Maximum number of conversation turns Claude can take (limits back-and-forth exchanges)                                                | No       | -         |
 | `timeout_minutes`              | Timeout in minutes for execution                                                                                                      | No       | `30`      |
 | `use_sticky_comment`           | Use just one comment to deliver PR comments (only applies for pull_request event workflows)                                           | No       | `false`   |
 | `github_token`                 | GitHub token for Claude to operate with. **Only include this if you're connecting a custom GitHub app of your own!**                  | No       | -         |
 | `model`                        | Model to use (provider-specific format required for Bedrock/Vertex)                                                                   | No       | -         |
 | `fallback_model`               | Enable automatic fallback to specified model when primary model is unavailable                                                        | No       | -         |
 | `anthropic_model`              | **DEPRECATED**: Use `model` instead. Kept for backward compatibility.                                                                 | No       | -         |
 | `use_bedrock`                  | Use Amazon Bedrock with OIDC authentication instead of direct Anthropic API                                                           | No       | `false`   |
 | `use_vertex`                   | Use Google Vertex AI with OIDC authentication instead of direct Anthropic API                                                         | No       | `false`   |
 | `allowed_tools`                | Additional tools for Claude to use (the base GitHub tools will always be included)                                                    | No       | ""        |
 | `disallowed_tools`             | Tools that Claude should never use                                                                                                    | No       | ""        |
 | `custom_instructions`          | Additional custom instructions to include in the prompt for Claude                                                                    | No       | ""        |
 | `mcp_config`                   | Additional MCP configuration (JSON string) that merges with the built-in GitHub MCP servers                                           | No       | ""        |
 | `assignee_trigger`             | The assignee username that triggers the action (e.g. @claude). Only used for issue assignment                                         | No       | -         |
 | `label_trigger`                | The label name that triggers the action when applied to an issue (e.g. "claude")                                                      | No       | -         |
 | `trigger_phrase`               | The trigger phrase to look for in comments, issue/PR bodies, and issue titles                                                         | No       | `@claude` |
 | `branch_prefix`                | The prefix to use for Claude branches (defaults to 'claude/', use 'claude-' for dash format)                                          | No       | `claude/` |
 | `claude_env`                   | Custom environment variables to pass to Claude Code execution (YAML format)                                                           | No       | ""        |
 | `settings`                     | Claude Code settings as JSON string or path to settings JSON file                                                                     | No       | ""        |
 | `additional_permissions`       | Additional permissions to enable. Currently supports 'actions: read' for viewing workflow results                                     | No       | ""        |
 | `experimental_allowed_domains` | Restrict network access to these domains only (newline-separated).                                                                    | No       | ""        |
 | `use_commit_signing`           | Enable commit signing using GitHub's commit signature verification. When false, Claude uses standard git commands                     | No       | `false`   |
 \*Required when using direct Anthropic API (default and when not using Bedrock or Vertex)
 > **Note**: This action is currently in beta. Features and APIs may change as we continue to improve the integration.
 ## Execution Modes
 The action supports three execution modes, each optimized for different use cases:
 ### Tag Mode (Default)
 The traditional implementation mode that responds to @claude mentions, issue assignments, or labels.
 - **Triggers**: `@claude` mentions, issue assignment, label application
 - **Features**: Creates tracking comments with progress checkboxes, full implementation capabilities
 - **Use case**: General-purpose code implementation and Q&A
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    # mode: tag is the default
 ```
 ### Agent Mode
 For automation and scheduled tasks without trigger checking.
 - **Triggers**: Always runs (no trigger checking)
 - **Features**: Perfect for scheduled tasks, works with `override_prompt`
 - **Use case**: Maintenance tasks, automated reporting, scheduled checks
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    mode: agent
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    override_prompt: |
      Check for outdated dependencies and create an issue if any are found.
 ```
 ### Experimental Review Mode
 > **EXPERIMENTAL**: This mode is under active development and may change significantly. Use with caution in production workflows.
 Specialized mode for automated PR code reviews using GitHub's review API.
 - **Triggers**: Automatically on PR events (opened, synchronize, reopened) when configured in workflow
 - **Features**: Creates inline review comments with suggestions, batches feedback into a single review
 - **Use case**: Automated code reviews, security scanning, best practices enforcement
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    mode: experimental-review
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    custom_instructions: |
      Focus on security vulnerabilities, performance issues, and code quality.
 ```
 Review mode automatically includes GitHub MCP tools for creating pending reviews and inline comments. See [`examples/claude-experimental-review-mode.yml`](./examples/claude-experimental-review-mode.yml) for a complete example.
 See [`examples/claude-modes.yml`](./examples/claude-modes.yml) for complete examples of available modes.
 ### Using Custom MCP Configuration
 The `mcp_config` input allows you to add custom MCP (Model Context Protocol) servers to extend Claude's capabilities. These servers merge with the built-in GitHub MCP servers.
 #### Basic Example: Adding a Sequential Thinking Server
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    mcp_config: |
      {
        "mcpServers": {
          "sequential-thinking": {
            "command": "npx",
            "args": [
              "-y",
              "@modelcontextprotocol/server-sequential-thinking"
            ]
          }
        }
      }
    allowed_tools: "mcp__sequential-thinking__sequentialthinking" # Important: Each MCP tool from your server must be listed here, comma-separated
    # ... other inputs
 ```
 #### Passing Secrets to MCP Servers
 For MCP servers that require sensitive information like API keys or tokens, use GitHub Secrets in the environment variables:
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    mcp_config: |
      {
        "mcpServers": {
          "custom-api-server": {
            "command": "npx",
            "args": ["-y", "@example/api-server"],
            "env": {
              "API_KEY": "${{ secrets.CUSTOM_API_KEY }}",
              "BASE_URL": "https://api.example.com"
            }
          }
        }
      }
    # ... other inputs
 ```
 #### Using Python MCP Servers with uv
 For Python-based MCP servers managed with `uv`, you need to specify the directory containing your server:
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    mcp_config: |
      {
        "mcpServers": {
          "my-python-server": {
            "type": "stdio",
            "command": "uv",
            "args": [
              "--directory",
              "${{ github.workspace }}/path/to/server/",
              "run",
              "server_file.py"
            ]
          }
        }
      }
    allowed_tools: "my-python-server__<tool_name>" # Replace <tool_name> with your server's tool names
    # ... other inputs
 ```
 For example, if your Python MCP server is at `mcp_servers/weather.py`, you would use:
 ```yaml
 "args":
  ["--directory", "${{ github.workspace }}/mcp_servers/", "run", "weather.py"]
 ```
 **Important**:
 - Always use GitHub Secrets (`${{ secrets.SECRET_NAME }}`) for sensitive values like API keys, tokens, or passwords. Never hardcode secrets directly in the workflow file.
 - Your custom servers will override any built-in servers with the same name.
 ## Examples
 ### Ways to Tag @claude
 These examples show how to interact with Claude using comments in PRs and issues. By default, Claude will be triggered anytime you mention `@claude`, but you can customize the exact trigger phrase using the `trigger_phrase` input in the workflow.
 Claude will see the full PR context, including any comments.
 #### Ask Questions
 Add a comment to a PR or issue:
 ```
@claude What does this function do and how could we improve it?
 ```
 Claude will analyze the code and provide a detailed explanation with suggestions.
 #### Request Fixes
 Ask Claude to implement specific changes:
 ```
@claude Can you add error handling to this function?
 ```
 #### Code Review
 Get a thorough review:
 ```
@claude Please review this PR and suggest improvements
 ```
 Claude will analyze the changes and provide feedback.
 #### Fix Bugs from Screenshots
 Upload a screenshot of a bug and ask Claude to fix it:
 ```
@claude Here's a screenshot of a bug I'm seeing [upload screenshot]. Can you fix it?
 ```
 Claude can see and analyze images, making it easy to fix visual bugs or UI issues.
 ### Custom Automations
 These examples show how to configure Claude to act automatically based on GitHub events, without requiring manual @mentions.
 #### Supported GitHub Events
 This action supports the following GitHub events ([learn more GitHub event triggers](https://docs.github.com/en/actions/writing-workflows/choosing-when-your-workflow-runs/events-that-trigger-workflows)):
 - `pull_request` - When PRs are opened or synchronized
 - `issue_comment` - When comments are created on issues or PRs
 - `pull_request_comment` - When comments are made on PR diffs
 - `issues` - When issues are opened or assigned
 - `pull_request_review` - When PR reviews are submitted
 - `pull_request_review_comment` - When comments are made on PR reviews
 - `repository_dispatch` - Custom events triggered via API (coming soon)
 - `workflow_dispatch` - Manual workflow triggers (coming soon)
 #### Automated Documentation Updates
 Automatically update documentation when specific files change (see [`examples/claude-pr-path-specific.yml`](./examples/claude-pr-path-specific.yml)):
 ```yaml
 on:
  pull_request:
    paths:
      - "src/api/**/*.ts"
 steps:
  - uses: anthropics/claude-code-action@beta
    with:
      direct_prompt: |
        Update the API documentation in README.md to reflect
        the changes made to the API endpoints in this PR.
 ```
 When API files are modified, Claude automatically updates your README with the latest endpoint documentation and pushes the changes back to the PR, keeping your docs in sync with your code.
 #### Author-Specific Code Reviews
 Automatically review PRs from specific authors or external contributors (see [`examples/claude-review-from-author.yml`](./examples/claude-review-from-author.yml)):
 ```yaml
 on:
  pull_request:
    types: [opened, synchronize]
 jobs:
  review-by-author:
    if: |
      github.event.pull_request.user.login == 'developer1' ||
      github.event.pull_request.user.login == 'external-contributor'
    steps:
      - uses: anthropics/claude-code-action@beta
        with:
          direct_prompt: |
            Please provide a thorough review of this pull request.
            Pay extra attention to coding standards, security practices,
            and test coverage since this is from an external contributor.
 ```
 Perfect for automatically reviewing PRs from new team members, external contributors, or specific developers who need extra guidance.
 #### Custom Prompt Templates
 Use `override_prompt` for complete control over Claude's behavior with variable substitution:
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    override_prompt: |
      Analyze PR #$PR_NUMBER in $REPOSITORY for security vulnerabilities.
      Changed files:
      $CHANGED_FILES
      Focus on:
      - SQL injection risks
      - XSS vulnerabilities
      - Authentication bypasses
      - Exposed secrets or credentials
      Provide severity ratings (Critical/High/Medium/Low) for any issues found.
 ```
 The `override_prompt` feature supports these variables:
 - `$REPOSITORY`, `$PR_NUMBER`, `$ISSUE_NUMBER`
 - `$PR_TITLE`, `$ISSUE_TITLE`, `$PR_BODY`, `$ISSUE_BODY`
 - `$PR_COMMENTS`, `$ISSUE_COMMENTS`, `$REVIEW_COMMENTS`
 - `$CHANGED_FILES`, `$TRIGGER_COMMENT`, `$TRIGGER_USERNAME`
 - `$BRANCH_NAME`, `$BASE_BRANCH`, `$EVENT_TYPE`, `$IS_PR`
 ## How It Works
 1. **Trigger Detection**: Listens for comments containing the trigger phrase (default: `@claude`) or issue assignment to a specific user
 2. **Context Gathering**: Analyzes the PR/issue, comments, code changes
 3. **Smart Responses**: Either answers questions or implements changes
 4. **Branch Management**: Creates new PRs for human authors, pushes directly for Claude's own PRs
 5. **Communication**: Posts updates at every step to keep you informed
 This action is built on top of [`anthropics/claude-code-base-action`](https://github.com/anthropics/claude-code-base-action).
 ## Capabilities and Limitations
 ### What Claude Can Do
 - **Respond in a Single Comment**: Claude operates by updating a single initial comment with progress and results
 - **Answer Questions**: Analyze code and provide explanations
 - **Implement Code Changes**: Make simple to moderate code changes based on requests
 - **Prepare Pull Requests**: Creates commits on a branch and links back to a prefilled PR creation page
 - **Perform Code Reviews**: Analyze PR changes and provide detailed feedback
 - **Smart Branch Handling**:
  - When triggered on an **issue**: Always creates a new branch for the work
  - When triggered on an **open PR**: Always pushes directly to the existing PR branch
  - When triggered on a **closed PR**: Creates a new branch since the original is no longer active
 - **View GitHub Actions Results**: Can access workflow runs, job logs, and test results on the PR where it's tagged when `actions: read` permission is configured (see [Additional Permissions for CI/CD Integration](#additional-permissions-for-cicd-integration))
 ### What Claude Cannot Do
 - **Submit PR Reviews**: Claude cannot submit formal GitHub PR reviews
 - **Approve PRs**: For security reasons, Claude cannot approve pull requests
 - **Post Multiple Comments**: Claude only acts by updating its initial comment
 - **Execute Commands Outside Its Context**: Claude only has access to the repository and PR/issue context it's triggered in
 - **Run Arbitrary Bash Commands**: By default, Claude cannot execute Bash commands unless explicitly allowed using the `allowed_tools` configuration
 - **Perform Branch Operations**: Cannot merge branches, rebase, or perform other git operations beyond pushing commits
 ## Advanced Configuration
 ### Additional Permissions for CI/CD Integration
 The `additional_permissions` input allows Claude to access GitHub Actions workflow information when you grant the necessary permissions. This is particularly useful for analyzing CI/CD failures and debugging workflow issues.
 #### Enabling GitHub Actions Access
 To allow Claude to view workflow run results, job logs, and CI status:
 1. **Grant the necessary permission to your GitHub token**:
   - When using the default `GITHUB_TOKEN`, add the `actions: read` permission to your workflow:
   ```yaml
   permissions:
     contents: write
     pull-requests: write
     issues: write
     actions: read # Add this line
   ```
 2. **Configure the action with additional permissions**:
   ```yaml
   - uses: anthropics/claude-code-action@beta
     with:
       anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
       additional_permissions: |
         actions: read
       # ... other inputs
   ```
 3. **Claude will automatically get access to CI/CD tools**:
   When you enable `actions: read`, Claude can use the following MCP tools:
   - `mcp__github_ci__get_ci_status` - View workflow run statuses
   - `mcp__github_ci__get_workflow_run_details` - Get detailed workflow information
   - `mcp__github_ci__download_job_log` - Download and analyze job logs
 #### Example: Debugging Failed CI Runs
 ```yaml
 name: Claude CI Helper
 on:
  issue_comment:
    types: [created]
 permissions:
  contents: write
  pull-requests: write
  issues: write
  actions: read # Required for CI access
 jobs:
  claude-ci-helper:
    runs-on: ubuntu-latest
    steps:
      - uses: anthropics/claude-code-action@beta
        with:
          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
          additional_permissions: |
            actions: read
          # Now Claude can respond to "@claude why did the CI fail?"
 ```
 **Important Notes**:
 - The GitHub token must have the `actions: read` permission in your workflow
 - If the permission is missing, Claude will warn you and suggest adding it
 - Currently, only `actions: read` is supported, but the format allows for future extensions
 ### Custom Environment Variables
 You can pass custom environment variables to Claude Code execution using the `claude_env` input. This is useful for CI/test setups that require specific environment variables:
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    claude_env: |
      NODE_ENV: test
      CI: true
      DATABASE_URL: postgres://test:test@localhost:5432/test_db
    # ... other inputs
 ```
 The `claude_env` input accepts YAML format where each line defines a key-value pair. These environment variables will be available to Claude Code during execution, allowing it to run tests, build processes, or other commands that depend on specific environment configurations.
 ### Limiting Conversation Turns
 You can use the `max_turns` parameter to limit the number of back-and-forth exchanges Claude can have during task execution. This is useful for:
 - Controlling costs by preventing runaway conversations
 - Setting time boundaries for automated workflows
 - Ensuring predictable behavior in CI/CD pipelines
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    max_turns: "5" # Limit to 5 conversation turns
    # ... other inputs
 ```
 When the turn limit is reached, Claude will stop execution gracefully. Choose a value that gives Claude enough turns to complete typical tasks while preventing excessive usage.
 ### Custom Tools
 By default, Claude only has access to:
 - File operations (reading, committing, editing files, read-only git commands)
 - Comment management (creating/updating comments)
 - Basic GitHub operations
 Claude does **not** have access to execute arbitrary Bash commands by default. If you want Claude to run specific commands (e.g., npm install, npm test), you must explicitly allow them using the `allowed_tools` configuration:
 **Note**: If your repository has a `.mcp.json` file in the root directory, Claude will automatically detect and use the MCP server tools defined there. However, these tools still need to be explicitly allowed via the `allowed_tools` configuration.
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    allowed_tools: |
      Bash(npm install)
      Bash(npm run test)
      Edit
      Replace
      NotebookEditCell
    disallowed_tools: |
      TaskOutput
      KillTask
    # ... other inputs
 ```
 **Note**: The base GitHub tools are always included. Use `allowed_tools` to add additional tools (including specific Bash commands), and `disallowed_tools` to prevent specific tools from being used.
 ### Custom Model
 Use a specific Claude model:
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    # model: "claude-3-5-sonnet-20241022"  # Optional: specify a different model
    # ... other inputs
 ```
 ### Network Restrictions
 For enhanced security, you can restrict Claude's network access to specific domains only. This feature is particularly useful for:
 - Enterprise environments with strict security policies
 - Preventing access to external services
 - Limiting Claude to only your internal APIs and services
 When `experimental_allowed_domains` is set, Claude can only access the domains you explicitly list. You'll need to include the appropriate provider domains based on your authentication method.
 #### Provider-Specific Examples
 ##### If using Anthropic API or subscription
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    # Or: claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
    experimental_allowed_domains: |
      .anthropic.com
 ```
 ##### If using AWS Bedrock
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    use_bedrock: "true"
    experimental_allowed_domains: |
      bedrock.*.amazonaws.com
      bedrock-runtime.*.amazonaws.com
 ```
 ##### If using Google Vertex AI
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    use_vertex: "true"
    experimental_allowed_domains: |
      *.googleapis.com
      vertexai.googleapis.com
 ```
 #### Common GitHub Domains
 In addition to your provider domains, you may need to include GitHub-related domains. For GitHub.com users, common domains include:
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    experimental_allowed_domains: |
      .anthropic.com  # For Anthropic API
      .github.com
      .githubusercontent.com
      ghcr.io
      .blob.core.windows.net
 ```
 For GitHub Enterprise users, replace the GitHub.com domains above with your enterprise domains (e.g., `.github.company.com`, `packages.company.com`, etc.).
 To determine which domains your workflow needs, you can temporarily run without restrictions and monitor the network requests, or check your GitHub Enterprise configuration for the specific services you use.
 ### Claude Code Settings
 You can provide Claude Code settings to customize behavior such as model selection, environment variables, permissions, and hooks. Settings can be provided either as a JSON string or a path to a settings file.
 #### Option 1: Settings File
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    settings: "path/to/settings.json"
    # ... other inputs
 ```
 #### Option 2: Inline Settings
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    settings: |
      {
        "model": "claude-opus-4-20250514",
        "env": {
          "DEBUG": "true",
          "API_URL": "https://api.example.com"
        },
        "permissions": {
          "allow": ["Bash", "Read"],
          "deny": ["WebFetch"]
        },
        "hooks": {
          "PreToolUse": [{
            "matcher": "Bash",
            "hooks": [{
              "type": "command",
              "command": "echo Running bash command..."
            }]
          }]
        }
      }
    # ... other inputs
 ```
 The settings support all Claude Code settings options including:
 - `model`: Override the default model
 - `env`: Environment variables for the session
 - `permissions`: Tool usage permissions
 - `hooks`: Pre/post tool execution hooks
 - And more...
 For a complete list of available settings and their descriptions, see the [Claude Code settings documentation](https://docs.anthropic.com/en/docs/claude-code/settings).
 **Notes**:
 - The `enableAllProjectMcpServers` setting is always set to `true` by this action to ensure MCP servers work correctly.
 - If both the `model` input parameter and a `model` in settings are provided, the `model` input parameter takes precedence.
 - The `allowed_tools` and `disallowed_tools` input parameters take precedence over `permissions` in settings.
 - In a future version, we may deprecate individual input parameters in favor of using the settings file for all configuration.
 ## Cloud Providers
 You can authenticate with Claude using any of these three methods:
 1. Direct Anthropic API (default)
 2. Amazon Bedrock with OIDC authentication
 3. Google Vertex AI with OIDC authentication
 For detailed setup instructions for AWS Bedrock and Google Vertex AI, see the [official documentation](https://docs.anthropic.com/en/docs/claude-code/github-actions#using-with-aws-bedrock-%26-google-vertex-ai).
 **Note**:
 - Bedrock and Vertex use OIDC authentication exclusively
 - AWS Bedrock automatically uses cross-region inference profiles for certain models
 - For cross-region inference profile models, you need to request and be granted access to the Claude models in all regions that the inference profile uses
 ### Model Configuration
 Use provider-specific model names based on your chosen provider:
 ```yaml
 # For direct Anthropic API (default)
 - uses: anthropics/claude-code-action@beta
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    # ... other inputs
 # For Amazon Bedrock with OIDC
 - uses: anthropics/claude-code-action@beta
  with:
    model: "anthropic.claude-3-7-sonnet-20250219-beta:0" # Cross-region inference
    use_bedrock: "true"
    # ... other inputs
 # For Google Vertex AI with OIDC
 - uses: anthropics/claude-code-action@beta
  with:
    model: "claude-3-7-sonnet@20250219"
    use_vertex: "true"
    # ... other inputs
 ```
 ### OIDC Authentication for Bedrock and Vertex
 Both AWS Bedrock and GCP Vertex AI require OIDC authentication.
 ```yaml
 # For AWS Bedrock with OIDC
 - name: Configure AWS Credentials (OIDC)
  uses: aws-actions/configure-aws-credentials@v4
  with:
    role-to-assume: ${{ secrets.AWS_ROLE_TO_ASSUME }}
    aws-region: us-west-2
 - name: Generate GitHub App token
  id: app-token
  uses: actions/create-github-app-token@v2
  with:
    app-id: ${{ secrets.APP_ID }}
    private-key: ${{ secrets.APP_PRIVATE_KEY }}
 - uses: anthropics/claude-code-action@beta
  with:
    model: "anthropic.claude-3-7-sonnet-20250219-beta:0"
    use_bedrock: "true"
    # ... other inputs
  permissions:
    id-token: write # Required for OIDC
 ```
 ```yaml
 # For GCP Vertex AI with OIDC
 - name: Authenticate to Google Cloud
  uses: google-github-actions/auth@v2
  with:
    workload_identity_provider: ${{ secrets.GCP_WORKLOAD_IDENTITY_PROVIDER }}
    service_account: ${{ secrets.GCP_SERVICE_ACCOUNT }}
 - name: Generate GitHub App token
  id: app-token
  uses: actions/create-github-app-token@v2
  with:
    app-id: ${{ secrets.APP_ID }}
    private-key: ${{ secrets.APP_PRIVATE_KEY }}
 - uses: anthropics/claude-code-action@beta
  with:
    model: "claude-3-7-sonnet@20250219"
    use_vertex: "true"
    # ... other inputs
  permissions:
    id-token: write # Required for OIDC
 ```
 ## Security
 ### Access Control
 - **Repository Access**: The action can only be triggered by users with write access to the repository
 - **No Bot Triggers**: GitHub Apps and bots cannot trigger this action
 - **Token Permissions**: The GitHub app receives only a short-lived token scoped specifically to the repository it's operating in
 - **No Cross-Repository Access**: Each action invocation is limited to the repository where it was triggered
 - **Limited Scope**: The token cannot access other repositories or perform actions beyond the configured permissions
 ### GitHub App Permissions
 The [Claude Code GitHub app](https://github.com/apps/claude) requires these permissions:
 - **Pull Requests**: Read and write to create PRs and push changes
 - **Issues**: Read and write to respond to issues
 - **Contents**: Read and write to modify repository files
 ### Commit Signing
 All commits made by Claude through this action are automatically signed with commit signatures. This ensures the authenticity and integrity of commits, providing a verifiable trail of changes made by the action.
 ### ⚠️ Authentication Protection
 **CRITICAL: Never hardcode your Anthropic API key or OAuth token in workflow files!**
 Your authentication credentials must always be stored in GitHub secrets to prevent unauthorized access:
 ```yaml
 # CORRECT ✅
 anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
 # OR
 claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
 # NEVER DO THIS ❌
 anthropic_api_key: "sk-ant-api03-..." # Exposed and vulnerable!
 claude_code_oauth_token: "oauth_token_..." # Exposed and vulnerable!
 ```
 ### Setting Up GitHub Secrets
 1. Go to your repository's Settings
 2. Click on "Secrets and variables" → "Actions"
 3. Click "New repository secret"
 4. For authentication, choose one:
   - API Key: Name: `ANTHROPIC_API_KEY`, Value: Your Anthropic API key (starting with `sk-ant-`)
   - OAuth Token: Name: `CLAUDE_CODE_OAUTH_TOKEN`, Value: Your Claude Code OAuth token (Pro and Max users can generate this by running `claude setup-token` locally)
 5. Click "Add secret"
 ### Best Practices for Authentication
 1. ✅ Always use `${{ secrets.ANTHROPIC_API_KEY }}` or `${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}` in workflows
 2. ✅ Never commit API keys or tokens to version control
 3. ✅ Regularly rotate your API keys and tokens
 4. ✅ Use environment secrets for organization-wide access
 5. ❌ Never share API keys or tokens in pull requests or issues
 6. ❌ Avoid logging workflow variables that might contain keys
 ## Security Best Practices
 **⚠️ IMPORTANT: Never commit API keys directly to your repository! Always use GitHub Actions secrets.**
 To securely use your Anthropic API key:
 1. Add your API key as a repository secret:
   - Go to your repository's Settings
   - Navigate to "Secrets and variables" → "Actions"
   - Click "New repository secret"
   - Name it `ANTHROPIC_API_KEY`
   - Paste your API key as the value
 2. Reference the secret in your workflow:
   ```yaml
   anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
   ```
 **Never do this:**
 ```yaml
 # ❌ WRONG - Exposes your API key
 anthropic_api_key: "sk-ant-..."
 ```
 **Always do this:**
 ```yaml
 # ✅ CORRECT - Uses GitHub secrets
 anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
 ```
 This applies to all sensitive values including API keys, access tokens, and credentials.
 We also recommend that you always use short-lived tokens when possible
 ## License
--- a/ROADMAP.md
+++ b/ROADMAP.md
@@ -10,7 +10,7 @@ Thank you for trying out the beta of our GitHub Action! This document outlines o
 - **Support for workflow_dispatch and repository_dispatch events** - Dispatch Claude on events triggered via API from other workflows or from other services
 - **Ability to disable commit signing** - Option to turn off GPG signing for environments where it's not required. This will enable Claude to use normal `git` bash commands for committing. This will likely become the default behavior once added.
 - **Better code review behavior** - Support inline comments on specific lines, provide higher quality reviews with more actionable feedback
- **Support triggering @claude from bot users** - Allow automation and bot accounts to invoke Claude
+- ~**Support triggering @claude from bot users** - Allow automation and bot accounts to invoke Claude~
 - **Customizable base prompts** - Full control over Claude's initial context with template variables like `$PR_COMMENTS`, `$PR_FILES`, etc. Users can replace our default prompt entirely while still accessing key contextual data
 ---
--- a/action.yml
+++ b/action.yml
@@ -23,6 +23,10 @@ inputs:
    description: "The prefix to use for Claude branches (defaults to 'claude/', use 'claude-' for dash format)"
    required: false
    default: "claude/"
  allowed_bots:
    description: "Comma-separated list of allowed bot usernames, or '*' to allow all bots. Empty string (default) allows no bots."
    required: false
    default: ""
  # Mode configuration
  mode:
@@ -156,6 +160,7 @@ runs:
        OVERRIDE_PROMPT: ${{ inputs.override_prompt }}
        MCP_CONFIG: ${{ inputs.mcp_config }}
        OVERRIDE_GITHUB_TOKEN: ${{ inputs.github_token }}
        ALLOWED_BOTS: ${{ inputs.allowed_bots }}
        GITHUB_RUN_ID: ${{ github.run_id }}
        USE_STICKY_COMMENT: ${{ inputs.use_sticky_comment }}
        DEFAULT_WORKFLOW_TOKEN: ${{ github.token }}
@@ -172,7 +177,7 @@ runs:
        echo "Base-action dependencies installed"
        cd -
        # Install Claude Code globally
-        bun install -g @anthropic-ai/claude-code@1.0.66
+        bun install -g @anthropic-ai/claude-code@1.0.71
    - name: Setup Network Restrictions
      if: steps.prepare.outputs.contains_trigger == 'true' && inputs.experimental_allowed_domains != ''
@@ -201,10 +206,11 @@ runs:
        INPUT_MCP_CONFIG: ${{ steps.prepare.outputs.mcp_config }}
        INPUT_SETTINGS: ${{ inputs.settings }}
        INPUT_SYSTEM_PROMPT: ""
-        INPUT_APPEND_SYSTEM_PROMPT: ""
+        INPUT_APPEND_SYSTEM_PROMPT: ${{ env.APPEND_SYSTEM_PROMPT }}
        INPUT_TIMEOUT_MINUTES: ${{ inputs.timeout_minutes }}
        INPUT_CLAUDE_ENV: ${{ inputs.claude_env }}
        INPUT_FALLBACK_MODEL: ${{ inputs.fallback_model }}
        INPUT_EXPERIMENTAL_SLASH_COMMANDS_DIR: ${{ github.action_path }}/slash-commands
        # Model configuration
        ANTHROPIC_MODEL: ${{ inputs.model || inputs.anthropic_model }}
--- a/base-action/README.md
+++ b/base-action/README.md
@@ -69,7 +69,7 @@ Add the following to your workflow file:
  uses: anthropics/claude-code-base-action@beta
  with:
    prompt: "Review and fix TypeScript errors"
-    model: "claude-opus-4-20250514"
+    model: "claude-opus-4-1-20250805"
    fallback_model: "claude-sonnet-4-20250514"
    allowed_tools: "Bash(git:*),View,GlobTool,GrepTool,BatchTool"
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
@@ -217,7 +217,7 @@ Provide the settings configuration directly as a JSON string:
    prompt: "Your prompt here"
    settings: |
      {
-        "model": "claude-opus-4-20250514",
+        "model": "claude-opus-4-1-20250805",
        "env": {
          "DEBUG": "true",
          "API_URL": "https://api.example.com"
--- a/base-action/action.yml
+++ b/base-action/action.yml
@@ -61,6 +61,9 @@ inputs:
    description: "Timeout in minutes for Claude Code execution"
    required: false
    default: "10"
  experimental_slash_commands_dir:
    description: "Experimental: Directory containing slash command files to install"
    required: false
  # Authentication settings
  anthropic_api_key:
@@ -115,7 +118,7 @@ runs:
    - name: Install Claude Code
      shell: bash
-      run: bun install -g @anthropic-ai/claude-code@1.0.66
+      run: bun install -g @anthropic-ai/claude-code@1.0.71
    - name: Run Claude Code Action
      shell: bash
@@ -143,6 +146,7 @@ runs:
        INPUT_TIMEOUT_MINUTES: ${{ inputs.timeout_minutes }}
        INPUT_CLAUDE_ENV: ${{ inputs.claude_env }}
        INPUT_FALLBACK_MODEL: ${{ inputs.fallback_model }}
        INPUT_EXPERIMENTAL_SLASH_COMMANDS_DIR: ${{ inputs.experimental_slash_commands_dir }}
        # Provider configuration
        ANTHROPIC_API_KEY: ${{ inputs.anthropic_api_key }}
--- a/base-action/src/index.ts
+++ b/base-action/src/index.ts
@@ -10,7 +10,11 @@ async function run() {
  try {
    validateEnvironmentVariables();
-    await setupClaudeCodeSettings(process.env.INPUT_SETTINGS);
+    await setupClaudeCodeSettings(
      process.env.INPUT_SETTINGS,
      undefined, // homeDir
      process.env.INPUT_EXPERIMENTAL_SLASH_COMMANDS_DIR,
    );
    const promptConfig = await preparePrompt({
      prompt: process.env.INPUT_PROMPT || "",
--- a/base-action/src/setup-claude-code-settings.ts
+++ b/base-action/src/setup-claude-code-settings.ts
@@ -5,6 +5,7 @@ import { readFile } from "fs/promises";
 export async function setupClaudeCodeSettings(
  settingsInput?: string,
  homeDir?: string,
  slashCommandsDir?: string,
 ) {
  const home = homeDir ?? homedir();
  const settingsPath = `${home}/.claude/settings.json`;
@@ -65,4 +66,17 @@ export async function setupClaudeCodeSettings(
  await $`echo ${JSON.stringify(settings, null, 2)} > ${settingsPath}`.quiet();
  console.log(`Settings saved successfully`);
  if (slashCommandsDir) {
    console.log(
      `Copying slash commands from ${slashCommandsDir} to ${home}/.claude/`,
    );
    try {
      await $`test -d ${slashCommandsDir}`.quiet();
      await $`cp ${slashCommandsDir}/*.md ${home}/.claude/ 2>/dev/null || true`.quiet();
      console.log(`Slash commands copied successfully`);
    } catch (e) {
      console.log(`Slash commands directory not found or error copying: ${e}`);
    }
  }
 }
--- a/base-action/test/setup-claude-code-settings.test.ts
+++ b/base-action/test/setup-claude-code-settings.test.ts
@@ -3,7 +3,7 @@
 import { describe, test, expect, beforeEach, afterEach } from "bun:test";
 import { setupClaudeCodeSettings } from "../src/setup-claude-code-settings";
 import { tmpdir } from "os";
-import { mkdir, writeFile, readFile, rm } from "fs/promises";
+import { mkdir, writeFile, readFile, rm, readdir } from "fs/promises";
 import { join } from "path";
 const testHomeDir = join(
@@ -134,7 +134,7 @@ describe("setupClaudeCodeSettings", () => {
    // Then, add new settings
    const newSettings = JSON.stringify({
      newKey: "newValue",
-      model: "claude-opus-4-20250514",
+      model: "claude-opus-4-1-20250805",
    });
    await setupClaudeCodeSettings(newSettings, testHomeDir);
@@ -145,6 +145,74 @@ describe("setupClaudeCodeSettings", () => {
    expect(settings.enableAllProjectMcpServers).toBe(true);
    expect(settings.existingKey).toBe("existingValue");
    expect(settings.newKey).toBe("newValue");
-    expect(settings.model).toBe("claude-opus-4-20250514");
+    expect(settings.model).toBe("claude-opus-4-1-20250805");
  });
  test("should copy slash commands to .claude directory when path provided", async () => {
    const testSlashCommandsDir = join(testHomeDir, "test-slash-commands");
    await mkdir(testSlashCommandsDir, { recursive: true });
    await writeFile(
      join(testSlashCommandsDir, "test-command.md"),
      "---\ndescription: Test command\n---\nTest content",
    );
    await setupClaudeCodeSettings(undefined, testHomeDir, testSlashCommandsDir);
    const testCommandPath = join(testHomeDir, ".claude", "test-command.md");
    const content = await readFile(testCommandPath, "utf-8");
    expect(content).toContain("Test content");
  });
  test("should skip slash commands when no directory provided", async () => {
    await setupClaudeCodeSettings(undefined, testHomeDir);
    const settingsContent = await readFile(settingsPath, "utf-8");
    const settings = JSON.parse(settingsContent);
    expect(settings.enableAllProjectMcpServers).toBe(true);
  });
  test("should handle missing slash commands directory gracefully", async () => {
    const nonExistentDir = join(testHomeDir, "non-existent");
    await setupClaudeCodeSettings(undefined, testHomeDir, nonExistentDir);
    const settingsContent = await readFile(settingsPath, "utf-8");
    expect(JSON.parse(settingsContent).enableAllProjectMcpServers).toBe(true);
  });
  test("should skip non-.md files in slash commands directory", async () => {
    const testSlashCommandsDir = join(testHomeDir, "test-slash-commands");
    await mkdir(testSlashCommandsDir, { recursive: true });
    await writeFile(join(testSlashCommandsDir, "not-markdown.txt"), "ignored");
    await writeFile(join(testSlashCommandsDir, "valid.md"), "copied");
    await writeFile(join(testSlashCommandsDir, "another.md"), "also copied");
    await setupClaudeCodeSettings(undefined, testHomeDir, testSlashCommandsDir);
    const copiedFiles = await readdir(join(testHomeDir, ".claude"));
    expect(copiedFiles).toContain("valid.md");
    expect(copiedFiles).toContain("another.md");
    expect(copiedFiles).not.toContain("not-markdown.txt");
    expect(copiedFiles).toContain("settings.json"); // Settings should also exist
  });
  test("should handle slash commands path that is a file not directory", async () => {
    const testFile = join(testHomeDir, "not-a-directory.txt");
    await writeFile(testFile, "This is a file, not a directory");
    await setupClaudeCodeSettings(undefined, testHomeDir, testFile);
    const settingsContent = await readFile(settingsPath, "utf-8");
    expect(JSON.parse(settingsContent).enableAllProjectMcpServers).toBe(true);
  });
  test("should handle empty slash commands directory", async () => {
    const emptyDir = join(testHomeDir, "empty-slash-commands");
    await mkdir(emptyDir, { recursive: true });
    await setupClaudeCodeSettings(undefined, testHomeDir, emptyDir);
    const settingsContent = await readFile(settingsPath, "utf-8");
    expect(JSON.parse(settingsContent).enableAllProjectMcpServers).toBe(true);
  });
 });
--- a/docs/capabilities-and-limitations.md
+++ b/docs/capabilities-and-limitations.md
@@ -0,0 +1,33 @@
 # Capabilities and Limitations
 ## What Claude Can Do
 - **Respond in a Single Comment**: Claude operates by updating a single initial comment with progress and results
 - **Answer Questions**: Analyze code and provide explanations
 - **Implement Code Changes**: Make simple to moderate code changes based on requests
 - **Prepare Pull Requests**: Creates commits on a branch and links back to a prefilled PR creation page
 - **Perform Code Reviews**: Analyze PR changes and provide detailed feedback
 - **Smart Branch Handling**:
  - When triggered on an **issue**: Always creates a new branch for the work
  - When triggered on an **open PR**: Always pushes directly to the existing PR branch
  - When triggered on a **closed PR**: Creates a new branch since the original is no longer active
 - **View GitHub Actions Results**: Can access workflow runs, job logs, and test results on the PR where it's tagged when `actions: read` permission is configured (see [Additional Permissions for CI/CD Integration](./configuration.md#additional-permissions-for-cicd-integration))
 ## What Claude Cannot Do
 - **Submit PR Reviews**: Claude cannot submit formal GitHub PR reviews
 - **Approve PRs**: For security reasons, Claude cannot approve pull requests
 - **Post Multiple Comments**: Claude only acts by updating its initial comment
 - **Execute Commands Outside Its Context**: Claude only has access to the repository and PR/issue context it's triggered in
 - **Run Arbitrary Bash Commands**: By default, Claude cannot execute Bash commands unless explicitly allowed using the `allowed_tools` configuration
 - **Perform Branch Operations**: Cannot merge branches, rebase, or perform other git operations beyond pushing commits
 ## How It Works
 1. **Trigger Detection**: Listens for comments containing the trigger phrase (default: `@claude`) or issue assignment to a specific user
 2. **Context Gathering**: Analyzes the PR/issue, comments, code changes
 3. **Smart Responses**: Either answers questions or implements changes
 4. **Branch Management**: Creates new PRs for human authors, pushes directly for Claude's own PRs
 5. **Communication**: Posts updates at every step to keep you informed
 This action is built on top of [`anthropics/claude-code-base-action`](https://github.com/anthropics/claude-code-base-action).
--- a/docs/cloud-providers.md
+++ b/docs/cloud-providers.md
@@ -0,0 +1,95 @@
 # Cloud Providers
 You can authenticate with Claude using any of these three methods:
 1. Direct Anthropic API (default)
 2. Amazon Bedrock with OIDC authentication
 3. Google Vertex AI with OIDC authentication
 For detailed setup instructions for AWS Bedrock and Google Vertex AI, see the [official documentation](https://docs.anthropic.com/en/docs/claude-code/github-actions#using-with-aws-bedrock-%26-google-vertex-ai).
 **Note**:
 - Bedrock and Vertex use OIDC authentication exclusively
 - AWS Bedrock automatically uses cross-region inference profiles for certain models
 - For cross-region inference profile models, you need to request and be granted access to the Claude models in all regions that the inference profile uses
 ## Model Configuration
 Use provider-specific model names based on your chosen provider:
 ```yaml
 # For direct Anthropic API (default)
 - uses: anthropics/claude-code-action@beta
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    # ... other inputs
 # For Amazon Bedrock with OIDC
 - uses: anthropics/claude-code-action@beta
  with:
    model: "anthropic.claude-3-7-sonnet-20250219-beta:0" # Cross-region inference
    use_bedrock: "true"
    # ... other inputs
 # For Google Vertex AI with OIDC
 - uses: anthropics/claude-code-action@beta
  with:
    model: "claude-3-7-sonnet@20250219"
    use_vertex: "true"
    # ... other inputs
 ```
 ## OIDC Authentication for Bedrock and Vertex
 Both AWS Bedrock and GCP Vertex AI require OIDC authentication.
 ```yaml
 # For AWS Bedrock with OIDC
 - name: Configure AWS Credentials (OIDC)
  uses: aws-actions/configure-aws-credentials@v4
  with:
    role-to-assume: ${{ secrets.AWS_ROLE_TO_ASSUME }}
    aws-region: us-west-2
 - name: Generate GitHub App token
  id: app-token
  uses: actions/create-github-app-token@v2
  with:
    app-id: ${{ secrets.APP_ID }}
    private-key: ${{ secrets.APP_PRIVATE_KEY }}
 - uses: anthropics/claude-code-action@beta
  with:
    model: "anthropic.claude-3-7-sonnet-20250219-beta:0"
    use_bedrock: "true"
    # ... other inputs
  permissions:
    id-token: write # Required for OIDC
 ```
 ```yaml
 # For GCP Vertex AI with OIDC
 - name: Authenticate to Google Cloud
  uses: google-github-actions/auth@v2
  with:
    workload_identity_provider: ${{ secrets.GCP_WORKLOAD_IDENTITY_PROVIDER }}
    service_account: ${{ secrets.GCP_SERVICE_ACCOUNT }}
 - name: Generate GitHub App token
  id: app-token
  uses: actions/create-github-app-token@v2
  with:
    app-id: ${{ secrets.APP_ID }}
    private-key: ${{ secrets.APP_PRIVATE_KEY }}
 - uses: anthropics/claude-code-action@beta
  with:
    model: "claude-3-7-sonnet@20250219"
    use_vertex: "true"
    # ... other inputs
  permissions:
    id-token: write # Required for OIDC
 ```
--- a/docs/configuration.md
+++ b/docs/configuration.md
@@ -0,0 +1,292 @@
 # Advanced Configuration
 ## Using Custom MCP Configuration
 The `mcp_config` input allows you to add custom MCP (Model Context Protocol) servers to extend Claude's capabilities. These servers merge with the built-in GitHub MCP servers.
 ### Basic Example: Adding a Sequential Thinking Server
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    mcp_config: |
      {
        "mcpServers": {
          "sequential-thinking": {
            "command": "npx",
            "args": [
              "-y",
              "@modelcontextprotocol/server-sequential-thinking"
            ]
          }
        }
      }
    allowed_tools: "mcp__sequential-thinking__sequentialthinking" # Important: Each MCP tool from your server must be listed here, comma-separated
    # ... other inputs
 ```
 ### Passing Secrets to MCP Servers
 For MCP servers that require sensitive information like API keys or tokens, use GitHub Secrets in the environment variables:
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    mcp_config: |
      {
        "mcpServers": {
          "custom-api-server": {
            "command": "npx",
            "args": ["-y", "@example/api-server"],
            "env": {
              "API_KEY": "${{ secrets.CUSTOM_API_KEY }}",
              "BASE_URL": "https://api.example.com"
            }
          }
        }
      }
    # ... other inputs
 ```
 ### Using Python MCP Servers with uv
 For Python-based MCP servers managed with `uv`, you need to specify the directory containing your server:
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    mcp_config: |
      {
        "mcpServers": {
          "my-python-server": {
            "type": "stdio",
            "command": "uv",
            "args": [
              "--directory",
              "${{ github.workspace }}/path/to/server/",
              "run",
              "server_file.py"
            ]
          }
        }
      }
    allowed_tools: "my-python-server__<tool_name>" # Replace <tool_name> with your server's tool names
    # ... other inputs
 ```
 For example, if your Python MCP server is at `mcp_servers/weather.py`, you would use:
 ```yaml
 "args":
  ["--directory", "${{ github.workspace }}/mcp_servers/", "run", "weather.py"]
 ```
 **Important**:
 - Always use GitHub Secrets (`${{ secrets.SECRET_NAME }}`) for sensitive values like API keys, tokens, or passwords. Never hardcode secrets directly in the workflow file.
 - Your custom servers will override any built-in servers with the same name.
 ## Additional Permissions for CI/CD Integration
 The `additional_permissions` input allows Claude to access GitHub Actions workflow information when you grant the necessary permissions. This is particularly useful for analyzing CI/CD failures and debugging workflow issues.
 ### Enabling GitHub Actions Access
 To allow Claude to view workflow run results, job logs, and CI status:
 1. **Grant the necessary permission to your GitHub token**:
   - When using the default `GITHUB_TOKEN`, add the `actions: read` permission to your workflow:
   ```yaml
   permissions:
     contents: write
     pull-requests: write
     issues: write
     actions: read # Add this line
   ```
 2. **Configure the action with additional permissions**:
   ```yaml
   - uses: anthropics/claude-code-action@beta
     with:
       anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
       additional_permissions: |
         actions: read
       # ... other inputs
   ```
 3. **Claude will automatically get access to CI/CD tools**:
   When you enable `actions: read`, Claude can use the following MCP tools:
   - `mcp__github_ci__get_ci_status` - View workflow run statuses
   - `mcp__github_ci__get_workflow_run_details` - Get detailed workflow information
   - `mcp__github_ci__download_job_log` - Download and analyze job logs
 ### Example: Debugging Failed CI Runs
 ```yaml
 name: Claude CI Helper
 on:
  issue_comment:
    types: [created]
 permissions:
  contents: write
  pull-requests: write
  issues: write
  actions: read # Required for CI access
 jobs:
  claude-ci-helper:
    runs-on: ubuntu-latest
    steps:
      - uses: anthropics/claude-code-action@beta
        with:
          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
          additional_permissions: |
            actions: read
          # Now Claude can respond to "@claude why did the CI fail?"
 ```
 **Important Notes**:
 - The GitHub token must have the `actions: read` permission in your workflow
 - If the permission is missing, Claude will warn you and suggest adding it
 - Currently, only `actions: read` is supported, but the format allows for future extensions
 ## Custom Environment Variables
 You can pass custom environment variables to Claude Code execution using the `claude_env` input. This is useful for CI/test setups that require specific environment variables:
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    claude_env: |
      NODE_ENV: test
      CI: true
      DATABASE_URL: postgres://test:test@localhost:5432/test_db
    # ... other inputs
 ```
 The `claude_env` input accepts YAML format where each line defines a key-value pair. These environment variables will be available to Claude Code during execution, allowing it to run tests, build processes, or other commands that depend on specific environment configurations.
 ## Limiting Conversation Turns
 You can use the `max_turns` parameter to limit the number of back-and-forth exchanges Claude can have during task execution. This is useful for:
 - Controlling costs by preventing runaway conversations
 - Setting time boundaries for automated workflows
 - Ensuring predictable behavior in CI/CD pipelines
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    max_turns: "5" # Limit to 5 conversation turns
    # ... other inputs
 ```
 When the turn limit is reached, Claude will stop execution gracefully. Choose a value that gives Claude enough turns to complete typical tasks while preventing excessive usage.
 ## Custom Tools
 By default, Claude only has access to:
 - File operations (reading, committing, editing files, read-only git commands)
 - Comment management (creating/updating comments)
 - Basic GitHub operations
 Claude does **not** have access to execute arbitrary Bash commands by default. If you want Claude to run specific commands (e.g., npm install, npm test), you must explicitly allow them using the `allowed_tools` configuration:
 **Note**: If your repository has a `.mcp.json` file in the root directory, Claude will automatically detect and use the MCP server tools defined there. However, these tools still need to be explicitly allowed via the `allowed_tools` configuration.
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    allowed_tools: |
      Bash(npm install)
      Bash(npm run test)
      Edit
      Replace
      NotebookEditCell
    disallowed_tools: |
      TaskOutput
      KillTask
    # ... other inputs
 ```
 **Note**: The base GitHub tools are always included. Use `allowed_tools` to add additional tools (including specific Bash commands), and `disallowed_tools` to prevent specific tools from being used.
 ## Custom Model
 Use a specific Claude model:
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    # model: "claude-3-5-sonnet-20241022"  # Optional: specify a different model
    # ... other inputs
 ```
 ## Claude Code Settings
 You can provide Claude Code settings to customize behavior such as model selection, environment variables, permissions, and hooks. Settings can be provided either as a JSON string or a path to a settings file.
 ### Option 1: Settings File
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    settings: "path/to/settings.json"
    # ... other inputs
 ```
 ### Option 2: Inline Settings
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    settings: |
      {
        "model": "claude-opus-4-1-20250805",
        "env": {
          "DEBUG": "true",
          "API_URL": "https://api.example.com"
        },
        "permissions": {
          "allow": ["Bash", "Read"],
          "deny": ["WebFetch"]
        },
        "hooks": {
          "PreToolUse": [{
            "matcher": "Bash",
            "hooks": [{
              "type": "command",
              "command": "echo Running bash command..."
            }]
          }]
        }
      }
    # ... other inputs
 ```
 The settings support all Claude Code settings options including:
 - `model`: Override the default model
 - `env`: Environment variables for the session
 - `permissions`: Tool usage permissions
 - `hooks`: Pre/post tool execution hooks
 - And more...
 For a complete list of available settings and their descriptions, see the [Claude Code settings documentation](https://docs.anthropic.com/en/docs/claude-code/settings).
 **Notes**:
 - The `enableAllProjectMcpServers` setting is always set to `true` by this action to ensure MCP servers work correctly.
 - If both the `model` input parameter and a `model` in settings are provided, the `model` input parameter takes precedence.
 - The `allowed_tools` and `disallowed_tools` input parameters take precedence over `permissions` in settings.
 - In a future version, we may deprecate individual input parameters in favor of using the settings file for all configuration.
--- a/docs/custom-automations.md
+++ b/docs/custom-automations.md
@@ -0,0 +1,91 @@
 # Custom Automations
 These examples show how to configure Claude to act automatically based on GitHub events, without requiring manual @mentions.
 ## Supported GitHub Events
 This action supports the following GitHub events ([learn more GitHub event triggers](https://docs.github.com/en/actions/writing-workflows/choosing-when-your-workflow-runs/events-that-trigger-workflows)):
 - `pull_request` - When PRs are opened or synchronized
 - `issue_comment` - When comments are created on issues or PRs
 - `pull_request_comment` - When comments are made on PR diffs
 - `issues` - When issues are opened or assigned
 - `pull_request_review` - When PR reviews are submitted
 - `pull_request_review_comment` - When comments are made on PR reviews
 - `repository_dispatch` - Custom events triggered via API (coming soon)
 - `workflow_dispatch` - Manual workflow triggers (coming soon)
 ## Automated Documentation Updates
 Automatically update documentation when specific files change (see [`examples/claude-pr-path-specific.yml`](../examples/claude-pr-path-specific.yml)):
 ```yaml
 on:
  pull_request:
    paths:
      - "src/api/**/*.ts"
 steps:
  - uses: anthropics/claude-code-action@beta
    with:
      direct_prompt: |
        Update the API documentation in README.md to reflect
        the changes made to the API endpoints in this PR.
 ```
 When API files are modified, Claude automatically updates your README with the latest endpoint documentation and pushes the changes back to the PR, keeping your docs in sync with your code.
 ## Author-Specific Code Reviews
 Automatically review PRs from specific authors or external contributors (see [`examples/claude-review-from-author.yml`](../examples/claude-review-from-author.yml)):
 ```yaml
 on:
  pull_request:
    types: [opened, synchronize]
 jobs:
  review-by-author:
    if: |
      github.event.pull_request.user.login == 'developer1' ||
      github.event.pull_request.user.login == 'external-contributor'
    steps:
      - uses: anthropics/claude-code-action@beta
        with:
          direct_prompt: |
            Please provide a thorough review of this pull request.
            Pay extra attention to coding standards, security practices,
            and test coverage since this is from an external contributor.
 ```
 Perfect for automatically reviewing PRs from new team members, external contributors, or specific developers who need extra guidance.
 ## Custom Prompt Templates
 Use `override_prompt` for complete control over Claude's behavior with variable substitution:
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    override_prompt: |
      Analyze PR #$PR_NUMBER in $REPOSITORY for security vulnerabilities.
      Changed files:
      $CHANGED_FILES
      Focus on:
      - SQL injection risks
      - XSS vulnerabilities
      - Authentication bypasses
      - Exposed secrets or credentials
      Provide severity ratings (Critical/High/Medium/Low) for any issues found.
 ```
 The `override_prompt` feature supports these variables:
 - `$REPOSITORY`, `$PR_NUMBER`, `$ISSUE_NUMBER`
 - `$PR_TITLE`, `$ISSUE_TITLE`, `$PR_BODY`, `$ISSUE_BODY`
 - `$PR_COMMENTS`, `$ISSUE_COMMENTS`, `$REVIEW_COMMENTS`
 - `$CHANGED_FILES`, `$TRIGGER_COMMENT`, `$TRIGGER_USERNAME`
 - `$BRANCH_NAME`, `$BASE_BRANCH`, `$EVENT_TYPE`, `$IS_PR`
--- a/docs/experimental.md
+++ b/docs/experimental.md
@@ -0,0 +1,127 @@
 # Experimental Features
 **Note:** Experimental features are considered unstable and not supported for production use. They may change or be removed at any time.
 ## Execution Modes
 The action supports three execution modes, each optimized for different use cases:
 ### Tag Mode (Default)
 The traditional implementation mode that responds to @claude mentions, issue assignments, or labels.
 - **Triggers**: `@claude` mentions, issue assignment, label application
 - **Features**: Creates tracking comments with progress checkboxes, full implementation capabilities
 - **Use case**: General-purpose code implementation and Q&A
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    # mode: tag is the default
 ```
 ### Agent Mode
 **Note: Agent mode is currently in active development and may undergo breaking changes.**
 For automation with workflow_dispatch and scheduled events only.
 - **Triggers**: Only works with `workflow_dispatch` and `schedule` events - does NOT work with PR/issue events
 - **Features**: Perfect for scheduled tasks, works with `override_prompt`
 - **Use case**: Maintenance tasks, automated reporting, scheduled checks
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    mode: agent
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    override_prompt: |
      Check for outdated dependencies and create an issue if any are found.
 ```
 ### Experimental Review Mode
 **Warning: This is an experimental feature that may change or be removed at any time.**
 For automated code reviews on pull requests.
 - **Triggers**: Pull request events (`opened`, `synchronize`) or `@claude review` comments
 - **Features**: Provides detailed code reviews with inline comments and suggestions
 - **Use case**: Automated PR reviews, code quality checks
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    mode: experimental-review
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    custom_instructions: |
      Focus on code quality, security, and best practices.
 ```
 See [`examples/claude-modes.yml`](../examples/claude-modes.yml) and [`examples/claude-experimental-review-mode.yml`](../examples/claude-experimental-review-mode.yml) for complete examples of each mode.
 ## Network Restrictions
 For enhanced security, you can restrict Claude's network access to specific domains only. This feature is particularly useful for:
 - Enterprise environments with strict security policies
 - Preventing access to external services
 - Limiting Claude to only your internal APIs and services
 When `experimental_allowed_domains` is set, Claude can only access the domains you explicitly list. You'll need to include the appropriate provider domains based on your authentication method.
 ### Provider-Specific Examples
 #### If using Anthropic API or subscription
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    # Or: claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
    experimental_allowed_domains: |
      .anthropic.com
 ```
 #### If using AWS Bedrock
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    use_bedrock: "true"
    experimental_allowed_domains: |
      bedrock.*.amazonaws.com
      bedrock-runtime.*.amazonaws.com
 ```
 #### If using Google Vertex AI
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    use_vertex: "true"
    experimental_allowed_domains: |
      *.googleapis.com
      vertexai.googleapis.com
 ```
 ### Common GitHub Domains
 In addition to your provider domains, you may need to include GitHub-related domains. For GitHub.com users, common domains include:
 ```yaml
 - uses: anthropics/claude-code-action@beta
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    experimental_allowed_domains: |
      .anthropic.com  # For Anthropic API
      .github.com
      .githubusercontent.com
      ghcr.io
      .blob.core.windows.net
 ```
 For GitHub Enterprise users, replace the GitHub.com domains above with your enterprise domains (e.g., `.github.company.com`, `packages.company.com`, etc.).
 To determine which domains your workflow needs, you can temporarily run without restrictions and monitor the network requests, or check your GitHub Enterprise configuration for the specific services you use.
--- a/docs/faq.md
+++ b/docs/faq.md
@@ -135,6 +135,14 @@ allowed_tools: "Bash(npm:*),Bash(git:*)" # Allows only npm and git commands
 No, Claude's GitHub app token is sandboxed to the current repository only. It cannot push to any other repositories. It can, however, read public repositories, but to get access to this, you must configure it with tools to do so.
 ### Why aren't comments posted as claude[bot]?
 Comments appear as claude[bot] when the action uses its built-in authentication. However, if you provide a `github_token` in your workflow, the action will use that token's authentication instead, causing comments to appear under a different username.
 **Solution**: Remove `github_token` from your workflow file unless you're using a custom GitHub App.
 **Note**: The `use_sticky_comment` feature only works with claude[bot] authentication. If you're using a custom `github_token`, sticky comments won't update properly since they expect the claude[bot] username.
 ## MCP Servers and Extended Functionality
 ### What MCP servers are available by default?
--- a/docs/security.md
+++ b/docs/security.md
@@ -0,0 +1,38 @@
 # Security
 ## Access Control
 - **Repository Access**: The action can only be triggered by users with write access to the repository
 - **Bot User Control**: By default, GitHub Apps and bots cannot trigger this action for security reasons. Use the `allowed_bots` parameter to enable specific bots or all bots
 - **Token Permissions**: The GitHub app receives only a short-lived token scoped specifically to the repository it's operating in
 - **No Cross-Repository Access**: Each action invocation is limited to the repository where it was triggered
 - **Limited Scope**: The token cannot access other repositories or perform actions beyond the configured permissions
 ## GitHub App Permissions
 The [Claude Code GitHub app](https://github.com/apps/claude) requires these permissions:
 - **Pull Requests**: Read and write to create PRs and push changes
 - **Issues**: Read and write to respond to issues
 - **Contents**: Read and write to modify repository files
 ## Commit Signing
 All commits made by Claude through this action are automatically signed with commit signatures. This ensures the authenticity and integrity of commits, providing a verifiable trail of changes made by the action.
 ## ⚠️ Authentication Protection
 **CRITICAL: Never hardcode your Anthropic API key or OAuth token in workflow files!**
 Your authentication credentials must always be stored in GitHub secrets to prevent unauthorized access:
 ```yaml
 # CORRECT ✅
 anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
 # OR
 claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
 # NEVER DO THIS ❌
 anthropic_api_key: "sk-ant-api03-..." # Exposed and vulnerable!
 claude_code_oauth_token: "oauth_token_..." # Exposed and vulnerable!
 ```
--- a/docs/setup.md
+++ b/docs/setup.md
@@ -0,0 +1,146 @@
 # Setup Guide
 ## Manual Setup (Direct API)
 **Requirements**: You must be a repository admin to complete these steps.
 1. Install the Claude GitHub app to your repository: https://github.com/apps/claude
 2. Add authentication to your repository secrets ([Learn how to use secrets in GitHub Actions](https://docs.github.com/en/actions/security-for-github-actions/security-guides/using-secrets-in-github-actions)):
   - Either `ANTHROPIC_API_KEY` for API key authentication
   - Or `CLAUDE_CODE_OAUTH_TOKEN` for OAuth token authentication (Pro and Max users can generate this by running `claude setup-token` locally)
 3. Copy the workflow file from [`examples/claude.yml`](../examples/claude.yml) into your repository's `.github/workflows/`
 ## Using a Custom GitHub App
 If you prefer not to install the official Claude app, you can create your own GitHub App to use with this action. This gives you complete control over permissions and access.
 **When you may want to use a custom GitHub App:**
 - You need more restrictive permissions than the official app
 - Organization policies prevent installing third-party apps
 - You're using AWS Bedrock or Google Vertex AI
 **Steps to create and use a custom GitHub App:**
 1. **Create a new GitHub App:**
   - Go to https://github.com/settings/apps (for personal apps) or your organization's settings
   - Click "New GitHub App"
   - Configure the app with these minimum permissions:
     - **Repository permissions:**
       - Contents: Read & Write
       - Issues: Read & Write
       - Pull requests: Read & Write
     - **Account permissions:** None required
   - Set "Where can this GitHub App be installed?" to your preference
   - Create the app
 2. **Generate and download a private key:**
   - After creating the app, scroll down to "Private keys"
   - Click "Generate a private key"
   - Download the `.pem` file (keep this secure!)
 3. **Install the app on your repository:**
   - Go to the app's settings page
   - Click "Install App"
   - Select the repositories where you want to use Claude
 4. **Add the app credentials to your repository secrets:**
   - Go to your repository's Settings → Secrets and variables → Actions
   - Add these secrets:
     - `APP_ID`: Your GitHub App's ID (found in the app settings)
     - `APP_PRIVATE_KEY`: The contents of the downloaded `.pem` file
 5. **Update your workflow to use the custom app:**
   ```yaml
   name: Claude with Custom App
   on:
     issue_comment:
       types: [created]
     # ... other triggers
   jobs:
     claude-response:
       runs-on: ubuntu-latest
       steps:
         # Generate a token from your custom app
         - name: Generate GitHub App token
           id: app-token
           uses: actions/create-github-app-token@v1
           with:
             app-id: ${{ secrets.APP_ID }}
             private-key: ${{ secrets.APP_PRIVATE_KEY }}
         # Use Claude with your custom app's token
         - uses: anthropics/claude-code-action@beta
           with:
             anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
             github_token: ${{ steps.app-token.outputs.token }}
             # ... other configuration
   ```
 **Important notes:**
 - The custom app must have read/write permissions for Issues, Pull Requests, and Contents
 - Your app's token will have the exact permissions you configured, nothing more
 For more information on creating GitHub Apps, see the [GitHub documentation](https://docs.github.com/en/apps/creating-github-apps).
 ## Security Best Practices
 **⚠️ IMPORTANT: Never commit API keys directly to your repository! Always use GitHub Actions secrets.**
 To securely use your Anthropic API key:
 1. Add your API key as a repository secret:
   - Go to your repository's Settings
   - Navigate to "Secrets and variables" → "Actions"
   - Click "New repository secret"
   - Name it `ANTHROPIC_API_KEY`
   - Paste your API key as the value
 2. Reference the secret in your workflow:
   ```yaml
   anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
   ```
 **Never do this:**
 ```yaml
 # ❌ WRONG - Exposes your API key
 anthropic_api_key: "sk-ant-..."
 ```
 **Always do this:**
 ```yaml
 # ✅ CORRECT - Uses GitHub secrets
 anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
 ```
 This applies to all sensitive values including API keys, access tokens, and credentials.
 We also recommend that you always use short-lived tokens when possible
 ## Setting Up GitHub Secrets
 1. Go to your repository's Settings
 2. Click on "Secrets and variables" → "Actions"
 3. Click "New repository secret"
 4. For authentication, choose one:
   - API Key: Name: `ANTHROPIC_API_KEY`, Value: Your Anthropic API key (starting with `sk-ant-`)
   - OAuth Token: Name: `CLAUDE_CODE_OAUTH_TOKEN`, Value: Your Claude Code OAuth token (Pro and Max users can generate this by running `claude setup-token` locally)
 5. Click "Add secret"
 ### Best Practices for Authentication
 1. ✅ Always use `${{ secrets.ANTHROPIC_API_KEY }}` or `${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}` in workflows
 2. ✅ Never commit API keys or tokens to version control
 3. ✅ Regularly rotate your API keys and tokens
 4. ✅ Use environment secrets for organization-wide access
 5. ❌ Never share API keys or tokens in pull requests or issues
 6. ❌ Avoid logging workflow variables that might contain keys
--- a/docs/usage.md
+++ b/docs/usage.md
@@ -0,0 +1,129 @@
 # Usage
 Add a workflow file to your repository (e.g., `.github/workflows/claude.yml`):
 ```yaml
 name: Claude Assistant
 on:
  issue_comment:
    types: [created]
  pull_request_review_comment:
    types: [created]
  issues:
    types: [opened, assigned, labeled]
  pull_request_review:
    types: [submitted]
 jobs:
  claude-response:
    runs-on: ubuntu-latest
    steps:
      - uses: anthropics/claude-code-action@beta
        with:
          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
          # Or use OAuth token instead:
          # claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
          github_token: ${{ secrets.GITHUB_TOKEN }}
          # Optional: set execution mode (default: tag)
          # mode: "tag"
          # Optional: add custom trigger phrase (default: @claude)
          # trigger_phrase: "/claude"
          # Optional: add assignee trigger for issues
          # assignee_trigger: "claude"
          # Optional: add label trigger for issues
          # label_trigger: "claude"
          # Optional: add custom environment variables (YAML format)
          # claude_env: |
          #   NODE_ENV: test
          #   DEBUG: true
          #   API_URL: https://api.example.com
          # Optional: limit the number of conversation turns
          # max_turns: "5"
          # Optional: grant additional permissions (requires corresponding GitHub token permissions)
          # additional_permissions: |
          #   actions: read
          # Optional: allow bot users to trigger the action
          # allowed_bots: "dependabot[bot],renovate[bot]"
 ```
 ## Inputs
 | Input                          | Description                                                                                                                           | Required | Default   |
 | ------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------- | -------- | --------- |
 | `mode`                         | Execution mode: 'tag' (default - triggered by mentions/assignments), 'agent' (for automation), 'experimental-review' (for PR reviews) | No       | `tag`     |
 | `anthropic_api_key`            | Anthropic API key (required for direct API, not needed for Bedrock/Vertex)                                                            | No\*     | -         |
 | `claude_code_oauth_token`      | Claude Code OAuth token (alternative to anthropic_api_key)                                                                            | No\*     | -         |
 | `direct_prompt`                | Direct prompt for Claude to execute automatically without needing a trigger (for automated workflows)                                 | No       | -         |
 | `override_prompt`              | Complete replacement of Claude's prompt with custom template (supports variable substitution)                                         | No       | -         |
 | `base_branch`                  | The base branch to use for creating new branches (e.g., 'main', 'develop')                                                            | No       | -         |
 | `max_turns`                    | Maximum number of conversation turns Claude can take (limits back-and-forth exchanges)                                                | No       | -         |
 | `timeout_minutes`              | Timeout in minutes for execution                                                                                                      | No       | `30`      |
 | `use_sticky_comment`           | Use just one comment to deliver PR comments (only applies for pull_request event workflows)                                           | No       | `false`   |
 | `github_token`                 | GitHub token for Claude to operate with. **Only include this if you're connecting a custom GitHub app of your own!**                  | No       | -         |
 | `model`                        | Model to use (provider-specific format required for Bedrock/Vertex)                                                                   | No       | -         |
 | `fallback_model`               | Enable automatic fallback to specified model when primary model is unavailable                                                        | No       | -         |
 | `anthropic_model`              | **DEPRECATED**: Use `model` instead. Kept for backward compatibility.                                                                 | No       | -         |
 | `use_bedrock`                  | Use Amazon Bedrock with OIDC authentication instead of direct Anthropic API                                                           | No       | `false`   |
 | `use_vertex`                   | Use Google Vertex AI with OIDC authentication instead of direct Anthropic API                                                         | No       | `false`   |
 | `allowed_tools`                | Additional tools for Claude to use (the base GitHub tools will always be included)                                                    | No       | ""        |
 | `disallowed_tools`             | Tools that Claude should never use                                                                                                    | No       | ""        |
 | `custom_instructions`          | Additional custom instructions to include in the prompt for Claude                                                                    | No       | ""        |
 | `mcp_config`                   | Additional MCP configuration (JSON string) that merges with the built-in GitHub MCP servers                                           | No       | ""        |
 | `assignee_trigger`             | The assignee username that triggers the action (e.g. @claude). Only used for issue assignment                                         | No       | -         |
 | `label_trigger`                | The label name that triggers the action when applied to an issue (e.g. "claude")                                                      | No       | -         |
 | `trigger_phrase`               | The trigger phrase to look for in comments, issue/PR bodies, and issue titles                                                         | No       | `@claude` |
 | `branch_prefix`                | The prefix to use for Claude branches (defaults to 'claude/', use 'claude-' for dash format)                                          | No       | `claude/` |
 | `claude_env`                   | Custom environment variables to pass to Claude Code execution (YAML format)                                                           | No       | ""        |
 | `settings`                     | Claude Code settings as JSON string or path to settings JSON file                                                                     | No       | ""        |
 | `additional_permissions`       | Additional permissions to enable. Currently supports 'actions: read' for viewing workflow results                                     | No       | ""        |
 | `experimental_allowed_domains` | Restrict network access to these domains only (newline-separated).                                                                    | No       | ""        |
 | `use_commit_signing`           | Enable commit signing using GitHub's commit signature verification. When false, Claude uses standard git commands                     | No       | `false`   |
 | `allowed_bots`                 | Comma-separated list of allowed bot usernames, or '\*' to allow all bots. Empty string (default) allows no bots                       | No       | ""        |
 \*Required when using direct Anthropic API (default and when not using Bedrock or Vertex)
 > **Note**: This action is currently in beta. Features and APIs may change as we continue to improve the integration.
 ## Ways to Tag @claude
 These examples show how to interact with Claude using comments in PRs and issues. By default, Claude will be triggered anytime you mention `@claude`, but you can customize the exact trigger phrase using the `trigger_phrase` input in the workflow.
 Claude will see the full PR context, including any comments.
 ### Ask Questions
 Add a comment to a PR or issue:
 ```
@claude What does this function do and how could we improve it?
 ```
 Claude will analyze the code and provide a detailed explanation with suggestions.
 ### Request Fixes
 Ask Claude to implement specific changes:
 ```
@claude Can you add error handling to this function?
 ```
 ### Code Review
 Get a thorough review:
 ```
@claude Please review this PR and suggest improvements
 ```
 Claude will analyze the changes and provide feedback.
 ### Fix Bugs from Screenshots
 Upload a screenshot of a bug and ask Claude to fix it:
 ```
@claude Here's a screenshot of a bug I'm seeing [upload screenshot]. Can you fix it?
 ```
 Claude can see and analyze images, making it easy to fix visual bugs or UI issues.
--- a/examples/claude-modes.yml
+++ b/examples/claude-modes.yml
@@ -1,13 +1,17 @@
 name: Claude Mode Examples
 on:
-  # Common events for both modes
+  # Events for tag mode
  issue_comment:
    types: [created]
  issues:
    types: [opened, labeled]
  pull_request:
    types: [opened]
  # Events for agent mode (only these work with agent mode)
  workflow_dispatch:
  schedule:
    - cron: "0 0 * * 0" # Weekly on Sunday
 jobs:
  # Tag Mode (Default) - Traditional implementation
@@ -28,13 +32,12 @@ jobs:
          # - Creates tracking comments with progress checkboxes
          # - Perfect for: Interactive Q&A, on-demand code changes
-  # Agent Mode - Automation without triggers
+  # Agent Mode - Automation for workflow_dispatch and schedule events
-  agent-mode-auto-review:
+  agent-mode-scheduled-task:
-    # Automatically review every new PR
+    # Only works with workflow_dispatch or schedule events
    if: github.event_name == 'pull_request' && github.event.action == 'opened'
    runs-on: ubuntu-latest
    permissions:
-      contents: read
+      contents: write
      pull-requests: write
      issues: write
      id-token: write
@@ -44,13 +47,10 @@ jobs:
          mode: agent
          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
          override_prompt: |
-            Review this PR for code quality. Focus on:
+            Check for outdated dependencies and security vulnerabilities.
-            - Potential bugs or logic errors
+            Create an issue if any critical problems are found.
            - Security concerns
            - Performance issues
            Provide specific, actionable feedback.
          # Agent mode behavior:
-          # - NO @claude mention needed - runs immediately
+          # - ONLY works with workflow_dispatch and schedule events
-          # - Enables true automation (impossible with tag mode)
+          # - Does NOT work with pull_request, issues, or issue_comment events
-          # - Perfect for: CI/CD integration, automatic reviews, label-based workflows
+          # - No @claude mention needed for supported events
          # - Perfect for: scheduled maintenance, manual automation runs
--- a/scripts/install-hooks.sh
+++ b/scripts/install-hooks.sh
@@ -6,8 +6,8 @@ echo "Installing git hooks..."
 # Make sure hooks directory exists
 mkdir -p .git/hooks
-# Install pre-push hook
+# Install pre-commit hook
-cp scripts/pre-push .git/hooks/pre-push
+cp scripts/pre-commit .git/hooks/pre-commit
-chmod +x .git/hooks/pre-push
+chmod +x .git/hooks/pre-commit
 echo "Git hooks installed successfully!"
--- a/scripts/pre-commit
+++ b/scripts/pre-commit
--- a/src/create-prompt/index.ts
+++ b/src/create-prompt/index.ts
@@ -60,8 +60,6 @@ export function buildAllowedToolsString(
      "Bash(git diff:*)",
      "Bash(git log:*)",
      "Bash(git rm:*)",
      "Bash(git config user.name:*)",
      "Bash(git config user.email:*)",
    );
  }
--- a/src/entrypoints/prepare.ts
+++ b/src/entrypoints/prepare.ts
@@ -81,6 +81,19 @@ async function run() {
    // Set the MCP config output
    core.setOutput("mcp_config", result.mcpConfig);
    // Step 6: Get system prompt from mode if available
    if (mode.getSystemPrompt) {
      const modeContext = mode.prepareContext(context, {
        commentId: result.commentId,
        baseBranch: result.branchInfo.baseBranch,
        claudeBranch: result.branchInfo.claudeBranch,
      });
      const systemPrompt = mode.getSystemPrompt(modeContext);
      if (systemPrompt) {
        core.exportVariable("APPEND_SYSTEM_PROMPT", systemPrompt);
      }
    }
  } catch (error) {
    const errorMessage = error instanceof Error ? error.message : String(error);
    core.setFailed(`Prepare step failed with error: ${errorMessage}`);
--- a/src/github/context.ts
+++ b/src/github/context.ts
@@ -77,6 +77,7 @@ type BaseContext = {
    useStickyComment: boolean;
    additionalPermissions: Map<string, string>;
    useCommitSigning: boolean;
    allowedBots: string;
  };
 };
@@ -136,6 +137,7 @@ export function parseGitHubContext(): GitHubContext {
        process.env.ADDITIONAL_PERMISSIONS ?? "",
      ),
      useCommitSigning: process.env.USE_COMMIT_SIGNING === "true",
      allowedBots: process.env.ALLOWED_BOTS ?? "",
    },
  };
--- a/src/github/utils/image-downloader.ts
+++ b/src/github/utils/image-downloader.ts
@@ -3,11 +3,17 @@ import path from "path";
 import type { Octokits } from "../api/client";
 import { GITHUB_SERVER_URL } from "../api/config";
 const escapedUrl = GITHUB_SERVER_URL.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
 const IMAGE_REGEX = new RegExp(
-  `!\\[[^\\]]*\\]\\((${GITHUB_SERVER_URL.replace(/[.*+?^${}()|[\]\\]/g, "\\$&")}\\/user-attachments\\/assets\\/[^)]+)\\)`,
+  `!\\[[^\\]]*\\]\\((${escapedUrl}\\/user-attachments\\/assets\\/[^)]+)\\)`,
  "g",
 );
 const HTML_IMG_REGEX = new RegExp(
  `<img[^>]+src=["']([^"']*${escapedUrl}\\/user-attachments\\/assets\\/[^"']+)["'][^>]*>`,
  "gi",
 );
 type IssueComment = {
  type: "issue_comment";
  id: string;
@@ -63,8 +69,16 @@ export async function downloadCommentImages(
  }> = [];
  for (const comment of comments) {
-    const imageMatches = [...comment.body.matchAll(IMAGE_REGEX)];
+    // Extract URLs from Markdown format
-    const urls = imageMatches.map((match) => match[1] as string);
+    const markdownMatches = [...comment.body.matchAll(IMAGE_REGEX)];
    const markdownUrls = markdownMatches.map((match) => match[1] as string);
    // Extract URLs from HTML format
    const htmlMatches = [...comment.body.matchAll(HTML_IMG_REGEX)];
    const htmlUrls = htmlMatches.map((match) => match[1] as string);
    // Combine and deduplicate URLs
    const urls = [...new Set([...markdownUrls, ...htmlUrls])];
    if (urls.length > 0) {
      commentsWithImages.push({ comment, urls });
--- a/src/github/validation/actor.ts
+++ b/src/github/validation/actor.ts
@@ -21,9 +21,42 @@ export async function checkHumanActor(
  console.log(`Actor type: ${actorType}`);
  // Check bot permissions if actor is not a User
  if (actorType !== "User") {
    const allowedBots = githubContext.inputs.allowedBots;
    // Check if all bots are allowed
    if (allowedBots.trim() === "*") {
      console.log(
        `All bots are allowed, skipping human actor check for: ${githubContext.actor}`,
      );
      return;
    }
    // Parse allowed bots list
    const allowedBotsList = allowedBots
      .split(",")
      .map((bot) =>
        bot
          .trim()
          .toLowerCase()
          .replace(/\[bot\]$/, ""),
      )
      .filter((bot) => bot.length > 0);
    const botName = githubContext.actor.toLowerCase().replace(/\[bot\]$/, "");
    // Check if specific bot is allowed
    if (allowedBotsList.includes(botName)) {
      console.log(
        `Bot ${botName} is in allowed list, skipping human actor check`,
      );
      return;
    }
    // Bot not allowed
    throw new Error(
-      `Workflow initiated by non-human actor: ${githubContext.actor} (type: ${actorType}).`,
+      `Workflow initiated by non-human actor: ${botName} (type: ${actorType}). Add bot to allowed_bots list or use '*' to allow all bots.`,
    );
  }
--- a/src/github/validation/permissions.ts
+++ b/src/github/validation/permissions.ts
@@ -17,6 +17,12 @@ export async function checkWritePermissions(
  try {
    core.info(`Checking permissions for actor: ${actor}`);
    // Check if the actor is a GitHub App (bot user)
    if (actor.endsWith("[bot]")) {
      core.info(`Actor is a GitHub App: ${actor}`);
      return true;
    }
    // Check permissions directly using the permission endpoint
    const response = await octokit.repos.getCollaboratorPermissionLevel({
      owner: repository.owner,
--- a/src/mcp/github-inline-comment-server.ts
+++ b/src/mcp/github-inline-comment-server.ts
@@ -0,0 +1,180 @@
 #!/usr/bin/env node
 import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
 import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
 import { z } from "zod";
 import { createOctokit } from "../github/api/client";
 // Get repository and PR information from environment variables
 const REPO_OWNER = process.env.REPO_OWNER;
 const REPO_NAME = process.env.REPO_NAME;
 const PR_NUMBER = process.env.PR_NUMBER;
 if (!REPO_OWNER || !REPO_NAME || !PR_NUMBER) {
  console.error(
    "Error: REPO_OWNER, REPO_NAME, and PR_NUMBER environment variables are required",
  );
  process.exit(1);
 }
 // GitHub Inline Comment MCP Server - Provides inline PR comment functionality
 // Provides an inline comment tool without exposing full PR review capabilities, so that
 // Claude can't accidentally approve a PR
 const server = new McpServer({
  name: "GitHub Inline Comment Server",
  version: "0.0.1",
 });
 server.tool(
  "create_inline_comment",
  "Create an inline comment on a specific line or lines in a PR file",
  {
    path: z
      .string()
      .describe("The file path to comment on (e.g., 'src/index.js')"),
    body: z
      .string()
      .describe(
        "The comment text (supports markdown and GitHub code suggestion blocks). " +
          "For code suggestions, use: ```suggestion\\nreplacement code\\n```. " +
          "IMPORTANT: The suggestion block will REPLACE the ENTIRE line range (single line or startLine to line). " +
          "Ensure the replacement is syntactically complete and valid - it must work as a drop-in replacement for the selected lines.",
      ),
    line: z
      .number()
      .nonnegative()
      .optional()
      .describe(
        "Line number for single-line comments (required if startLine is not provided)",
      ),
    startLine: z
      .number()
      .nonnegative()
      .optional()
      .describe(
        "Start line for multi-line comments (use with line parameter for the end line)",
      ),
    side: z
      .enum(["LEFT", "RIGHT"])
      .optional()
      .default("RIGHT")
      .describe(
        "Side of the diff to comment on: LEFT (old code) or RIGHT (new code)",
      ),
    commit_id: z
      .string()
      .optional()
      .describe(
        "Specific commit SHA to comment on (defaults to latest commit)",
      ),
  },
  async ({ path, body, line, startLine, side, commit_id }) => {
    try {
      const githubToken = process.env.GITHUB_TOKEN;
      if (!githubToken) {
        throw new Error("GITHUB_TOKEN environment variable is required");
      }
      const owner = REPO_OWNER;
      const repo = REPO_NAME;
      const pull_number = parseInt(PR_NUMBER, 10);
      const octokit = createOctokit(githubToken).rest;
      // Validate that either line or both startLine and line are provided
      if (!line && !startLine) {
        throw new Error(
          "Either 'line' for single-line comments or both 'startLine' and 'line' for multi-line comments must be provided",
        );
      }
      // If only line is provided, it's a single-line comment
      // If both startLine and line are provided, it's a multi-line comment
      const isSingleLine = !startLine;
      const pr = await octokit.pulls.get({
        owner,
        repo,
        pull_number,
      });
      const params: Parameters<
        typeof octokit.rest.pulls.createReviewComment
      >[0] = {
        owner,
        repo,
        pull_number,
        body,
        path,
        side: side || "RIGHT",
        commit_id: commit_id || pr.data.head.sha,
      };
      if (isSingleLine) {
        // Single-line comment
        params.line = line;
      } else {
        // Multi-line comment
        params.start_line = startLine;
        params.start_side = side || "RIGHT";
        params.line = line;
      }
      const result = await octokit.rest.pulls.createReviewComment(params);
      return {
        content: [
          {
            type: "text",
            text: JSON.stringify(
              {
                success: true,
                comment_id: result.data.id,
                html_url: result.data.html_url,
                path: result.data.path,
                line: result.data.line || result.data.original_line,
                message: `Inline comment created successfully on ${path}${isSingleLine ? ` at line ${line}` : ` from line ${startLine} to ${line}`}`,
              },
              null,
              2,
            ),
          },
        ],
      };
    } catch (error) {
      const errorMessage =
        error instanceof Error ? error.message : String(error);
      // Provide more helpful error messages for common issues
      let helpMessage = "";
      if (errorMessage.includes("Validation Failed")) {
        helpMessage =
          "\n\nThis usually means the line number doesn't exist in the diff or the file path is incorrect. Make sure you're commenting on lines that are part of the PR's changes.";
      } else if (errorMessage.includes("Not Found")) {
        helpMessage =
          "\n\nThis usually means the PR number, repository, or file path is incorrect.";
      }
      return {
        content: [
          {
            type: "text",
            text: `Error creating inline comment: ${errorMessage}${helpMessage}`,
          },
        ],
        error: errorMessage,
        isError: true,
      };
    }
  },
 );
 async function runServer() {
  const transport = new StdioServerTransport();
  await server.connect(transport);
  process.on("exit", () => {
    server.close();
  });
 }
 runServer().catch(console.error);
--- a/src/mcp/install-mcp-server.ts
+++ b/src/mcp/install-mcp-server.ts
@@ -111,6 +111,24 @@ export async function prepareMcpConfig(
      };
    }
    // Include inline comment server for experimental review mode
    if (context.inputs.mode === "experimental-review" && context.isPR) {
      baseMcpConfig.mcpServers.github_inline_comment = {
        command: "bun",
        args: [
          "run",
          `${process.env.GITHUB_ACTION_PATH}/src/mcp/github-inline-comment-server.ts`,
        ],
        env: {
          GITHUB_TOKEN: githubToken,
          REPO_OWNER: owner,
          REPO_NAME: repo,
          PR_NUMBER: context.entityNumber?.toString() || "",
          GITHUB_API_URL: GITHUB_API_URL,
        },
      };
    }
    // Only add CI server if we have actions:read permission and we're in a PR context
    const hasActionsReadPermission =
      context.inputs.additionalPermissions.get("actions") === "read";
@@ -118,7 +136,7 @@ export async function prepareMcpConfig(
    if (context.isPR && hasActionsReadPermission) {
      // Verify the token actually has actions:read permission
      const actuallyHasPermission = await checkActionsReadPermission(
-        process.env.ACTIONS_TOKEN || "",
+        process.env.DEFAULT_WORKFLOW_TOKEN || "",
        owner,
        repo,
      );
@@ -138,7 +156,7 @@ export async function prepareMcpConfig(
        ],
        env: {
          // Use workflow github token, not app token
-          GITHUB_TOKEN: process.env.ACTIONS_TOKEN,
+          GITHUB_TOKEN: process.env.DEFAULT_WORKFLOW_TOKEN,
          REPO_OWNER: owner,
          REPO_NAME: repo,
          PR_NUMBER: context.entityNumber?.toString() || "",
--- a/src/modes/agent/index.ts
+++ b/src/modes/agent/index.ts
@@ -1,4 +1,5 @@
 import * as core from "@actions/core";
 import { mkdir, writeFile } from "fs/promises";
 import type { Mode, ModeOptions, ModeResult } from "../types";
 import { isAutomationContext } from "../../github/context";
 import type { PreparedContext } from "../../create-prompt/types";
@@ -42,7 +43,23 @@ export const agentMode: Mode = {
  async prepare({ context }: ModeOptions): Promise<ModeResult> {
    // Agent mode handles automation events (workflow_dispatch, schedule) only
-    // Agent mode doesn't need to create prompt files here - handled by createPrompt
+    // TODO: handle by createPrompt (similar to tag and review modes)
    // Create prompt directory
    await mkdir(`${process.env.RUNNER_TEMP}/claude-prompts`, {
      recursive: true,
    });
    // Write the prompt file - the base action requires a prompt_file parameter,
    // so we must create this file even though agent mode typically uses
    // override_prompt or direct_prompt. If neither is provided, we write
    // a minimal prompt with just the repository information.
    const promptContent =
      context.inputs.overridePrompt ||
      context.inputs.directPrompt ||
      `Repository: ${context.repository.owner}/${context.repository.repo}`;
    await writeFile(
      `${process.env.RUNNER_TEMP}/claude-prompts/claude-prompt.txt`,
      promptContent,
    );
    // Export tool environment variables for agent mode
    const baseTools = [
@@ -112,4 +129,9 @@ export const agentMode: Mode = {
    // Minimal fallback - repository is a string in PreparedContext
    return `Repository: ${context.repository}`;
  },
  getSystemPrompt() {
    // Agent mode doesn't need additional system prompts
    return undefined;
  },
 };
--- a/src/modes/review/index.ts
+++ b/src/modes/review/index.ts
@@ -60,20 +60,8 @@ export const reviewMode: Mode = {
  getAllowedTools() {
    return [
-      // Context tools - to know who the current user is
+      "Bash(gh issue comment:*)",
-      "mcp__github__get_me",
+      "mcp__github_inline_comment__create_inline_comment",
      // Core review tools
      "mcp__github__create_pending_pull_request_review",
      "mcp__github__add_comment_to_pending_review",
      "mcp__github__submit_pending_pull_request_review",
      "mcp__github__delete_pending_pull_request_review",
      "mcp__github__create_and_submit_pull_request_review",
      // Comment tools
      "mcp__github__add_issue_comment",
      // PR information tools
      "mcp__github__get_pull_request",
      "mcp__github__get_pull_request_reviews",
      "mcp__github__get_pull_request_status",
    ];
  },
@@ -115,6 +103,9 @@ export const reviewMode: Mode = {
      ? formatBody(contextData.body, imageUrlMap)
      : "No description provided";
    // Using a variable for code blocks to avoid escaping backticks in the template string
    const codeBlock = "```";
    return `You are Claude, an AI assistant specialized in code reviews for GitHub pull requests. You are operating in REVIEW MODE, which means you should focus on providing thorough code review feedback using GitHub MCP tools for inline comments and suggestions.
 <formatted_context>
@@ -163,68 +154,50 @@ REVIEW MODE WORKFLOW:
 1. First, understand the PR context:
   - You are reviewing PR #${eventData.isPR && eventData.prNumber ? eventData.prNumber : "[PR number]"} in ${context.repository}
   - Use mcp__github__get_pull_request to get PR metadata
   - Use the Read, Grep, and Glob tools to examine the modified files directly from disk
   - This provides the full context and latest state of the code
   - Look at the changed_files section above to see which files were modified
-2. Create a pending review:
+2. Create review comments using GitHub MCP tools:
-   - Use mcp__github__create_pending_pull_request_review to start your review
+   - Use Bash(gh issue comment:*) for general PR-level comments
-   - This allows you to batch comments before submitting
+   - Use mcp__github_inline_comment__create_inline_comment for line-specific feedback (strongly preferred)
-3. Add inline comments:
+3. When creating inline comments with suggestions:
-   - Use mcp__github__add_comment_to_pending_review for each issue or suggestion
+   CRITICAL: GitHub's suggestion blocks REPLACE the ENTIRE line range you select
-   - Parameters:
+   - For single-line comments: Use 'line' parameter only
-     * path: The file path (e.g., "src/index.js")
+   - For multi-line comments: Use both 'startLine' and 'line' parameters
-     * line: Line number for single-line comments
+   - The 'body' parameter should contain your comment and/or suggestion block
     * startLine & line: For multi-line comments (startLine is the first line, line is the last)
     * side: "LEFT" (old code) or "RIGHT" (new code)
     * subjectType: "line" for line-level comments
     * body: Your comment text
-   - When to use multi-line comments:
+   How to write code suggestions correctly:
-     * When replacing multiple consecutive lines
+   a) To remove a line (e.g., removing console.log on line 22):
-     * When the fix requires changes across several lines
+      - Set line: 22
-     * Example: To replace lines 19-20, use startLine: 19, line: 20
+      - Body: ${codeBlock}suggestion
      ${codeBlock}
      (Empty suggestion block removes the line)
-   - For code suggestions, use this EXACT format in the body:
+   b) To modify a single line (e.g., fixing line 22):
-     \`\`\`suggestion
+      - Set line: 22  
-     corrected code here
+      - Body: ${codeBlock}suggestion
-     \`\`\`
+      await this.emailInput.fill(email);
      ${codeBlock}
-   CRITICAL: GitHub suggestion blocks must ONLY contain the replacement for the specific line(s) being commented on:
+   c) To replace multiple lines (e.g., lines 21-23):
-   - For single-line comments: Replace ONLY that line
+      - Set startLine: 21, line: 23
-   - For multi-line comments: Replace ONLY the lines in the range
+      - Body must include ALL lines being replaced:
-   - Do NOT include surrounding context or function signatures
+      ${codeBlock}suggestion
-   - Do NOT suggest changes that span beyond the commented lines
+      async typeEmail(email: string): Promise<void> {
          await this.emailInput.fill(email);
      }
      ${codeBlock}
-   Example for line 19 \`var name = user.name;\`:
+   COMMON MISTAKE TO AVOID:
-   WRONG:
+   Never duplicate code in suggestions. For example, DON'T do this:
-   \\\`\\\`\\\`suggestion
+   ${codeBlock}suggestion
-   function processUser(user) {
+   async typeEmail(email: string): Promise<void> {
-       if (!user) throw new Error('Invalid user');
+   async typeEmail(email: string): Promise<void> {  // WRONG: Duplicate signature!
-       const name = user.name;
+       await this.emailInput.fill(email);
-   \\\`\\\`\\\`
+   }
-   
+   ${codeBlock}
   CORRECT:
   \\\`\\\`\\\`suggestion
   const name = user.name;
   \\\`\\\`\\\`
   For validation suggestions, comment on the function declaration line or create separate comments for each concern.
 4. Submit your review:
   - Use mcp__github__submit_pending_pull_request_review
   - Parameters:
     * event: "COMMENT" (general feedback), "REQUEST_CHANGES" (issues found), or "APPROVE" (if appropriate)
     * body: Write a comprehensive review summary that includes:
       - Overview of what was reviewed (files, scope, focus areas)
       - Summary of all issues found (with counts by severity if applicable)
       - Key recommendations and action items
       - Highlights of good practices observed
       - Overall assessment and recommendation
   - The body should be detailed and informative since it's the main review content
   - Structure the body with clear sections using markdown headers
 REVIEW GUIDELINES:
@@ -238,13 +211,11 @@ REVIEW GUIDELINES:
 - Provide:
  * Specific, actionable feedback
-  * Code suggestions when possible (following GitHub's format exactly)
+  * Code suggestions using the exact format described above
-  * Clear explanations of issues
+  * Clear explanations of issues found
-  * Constructive criticism
+  * Constructive criticism with solutions
  * Recognition of good practices
-  * For complex changes that require multiple modifications:
+  * For complex changes: Create separate inline comments for each logical change
    - Create separate comments for each logical change
    - Or explain the full solution in text without a suggestion block
 - Communication:
  * All feedback goes through GitHub's review system
@@ -349,4 +320,10 @@ This ensures users get value from the review even before checking individual inl
      mcpConfig,
    };
  },
  getSystemPrompt() {
    // Review mode doesn't need additional system prompts
    // The review-specific instructions are included in the main prompt
    return undefined;
  },
 };
--- a/src/modes/tag/index.ts
+++ b/src/modes/tag/index.ts
@@ -130,4 +130,9 @@ export const tagMode: Mode = {
  ): string {
    return generateDefaultPrompt(context, githubData, useCommitSigning);
  },
  getSystemPrompt() {
    // Tag mode doesn't need additional system prompts
    return undefined;
  },
 };
--- a/src/modes/types.ts
+++ b/src/modes/types.ts
@@ -73,6 +73,13 @@ export type Mode = {
   * @returns PrepareResult with commentId, branchInfo, and mcpConfig
   */
  prepare(options: ModeOptions): Promise<ModeResult>;
  /**
   * Returns an optional system prompt to append to Claude's base system prompt.
   * This allows modes to add mode-specific instructions.
   * @returns The system prompt string or undefined if no additional prompt is needed
   */
  getSystemPrompt?(context: ModeContext): string | undefined;
 };
 // Define types for mode prepare method
--- a/test/actor.test.ts
+++ b/test/actor.test.ts
@@ -0,0 +1,96 @@
 #!/usr/bin/env bun
 import { describe, test, expect } from "bun:test";
 import { checkHumanActor } from "../src/github/validation/actor";
 import type { Octokit } from "@octokit/rest";
 import { createMockContext } from "./mockContext";
 function createMockOctokit(userType: string): Octokit {
  return {
    users: {
      getByUsername: async () => ({
        data: {
          type: userType,
        },
      }),
    },
  } as unknown as Octokit;
 }
 describe("checkHumanActor", () => {
  test("should pass for human actor", async () => {
    const mockOctokit = createMockOctokit("User");
    const context = createMockContext();
    context.actor = "human-user";
    await expect(
      checkHumanActor(mockOctokit, context),
    ).resolves.toBeUndefined();
  });
  test("should throw error for bot actor when not allowed", async () => {
    const mockOctokit = createMockOctokit("Bot");
    const context = createMockContext();
    context.actor = "test-bot[bot]";
    context.inputs.allowedBots = "";
    await expect(checkHumanActor(mockOctokit, context)).rejects.toThrow(
      "Workflow initiated by non-human actor: test-bot (type: Bot). Add bot to allowed_bots list or use '*' to allow all bots.",
    );
  });
  test("should pass for bot actor when all bots allowed", async () => {
    const mockOctokit = createMockOctokit("Bot");
    const context = createMockContext();
    context.actor = "test-bot[bot]";
    context.inputs.allowedBots = "*";
    await expect(
      checkHumanActor(mockOctokit, context),
    ).resolves.toBeUndefined();
  });
  test("should pass for specific bot when in allowed list", async () => {
    const mockOctokit = createMockOctokit("Bot");
    const context = createMockContext();
    context.actor = "dependabot[bot]";
    context.inputs.allowedBots = "dependabot[bot],renovate[bot]";
    await expect(
      checkHumanActor(mockOctokit, context),
    ).resolves.toBeUndefined();
  });
  test("should pass for specific bot when in allowed list (without [bot])", async () => {
    const mockOctokit = createMockOctokit("Bot");
    const context = createMockContext();
    context.actor = "dependabot[bot]";
    context.inputs.allowedBots = "dependabot,renovate";
    await expect(
      checkHumanActor(mockOctokit, context),
    ).resolves.toBeUndefined();
  });
  test("should throw error for bot not in allowed list", async () => {
    const mockOctokit = createMockOctokit("Bot");
    const context = createMockContext();
    context.actor = "other-bot[bot]";
    context.inputs.allowedBots = "dependabot[bot],renovate[bot]";
    await expect(checkHumanActor(mockOctokit, context)).rejects.toThrow(
      "Workflow initiated by non-human actor: other-bot (type: Bot). Add bot to allowed_bots list or use '*' to allow all bots.",
    );
  });
  test("should throw error for bot not in allowed list (without [bot])", async () => {
    const mockOctokit = createMockOctokit("Bot");
    const context = createMockContext();
    context.actor = "other-bot[bot]";
    context.inputs.allowedBots = "dependabot,renovate";
    await expect(checkHumanActor(mockOctokit, context)).rejects.toThrow(
      "Workflow initiated by non-human actor: other-bot (type: Bot). Add bot to allowed_bots list or use '*' to allow all bots.",
    );
  });
 });
--- a/test/create-prompt.test.ts
+++ b/test/create-prompt.test.ts
@@ -1041,8 +1041,6 @@ describe("buildAllowedToolsString", () => {
    expect(result).toContain("Bash(git diff:*)");
    expect(result).toContain("Bash(git log:*)");
    expect(result).toContain("Bash(git rm:*)");
    expect(result).toContain("Bash(git config user.name:*)");
    expect(result).toContain("Bash(git config user.email:*)");
    // Comment tool from minimal server should be included
    expect(result).toContain("mcp__github_comment__update_claude_comment");
--- a/test/image-downloader.test.ts
+++ b/test/image-downloader.test.ts
@@ -662,4 +662,255 @@ describe("downloadCommentImages", () => {
    );
    expect(result.get(imageUrl2)).toBeUndefined();
  });
  test("should detect and download images from HTML img tags", async () => {
    const mockOctokit = createMockOctokit();
    const imageUrl =
      "https://github.com/user-attachments/assets/html-image.png";
    const signedUrl =
      "https://private-user-images.githubusercontent.com/html.png?jwt=token";
    // Mock octokit response
    // @ts-expect-error Mock implementation doesn't match full type signature
    mockOctokit.rest.issues.getComment = jest.fn().mockResolvedValue({
      data: {
        body_html: `<img src="${signedUrl}">`,
      },
    });
    // Mock fetch for image download
    const mockArrayBuffer = new ArrayBuffer(8);
    fetchSpy = spyOn(global, "fetch").mockResolvedValue({
      ok: true,
      arrayBuffer: async () => mockArrayBuffer,
    } as Response);
    const comments: CommentWithImages[] = [
      {
        type: "issue_comment",
        id: "777",
        body: `Here's an HTML image: <img src="${imageUrl}" alt="test">`,
      },
    ];
    const result = await downloadCommentImages(
      mockOctokit,
      "owner",
      "repo",
      comments,
    );
    expect(mockOctokit.rest.issues.getComment).toHaveBeenCalledWith({
      owner: "owner",
      repo: "repo",
      comment_id: 777,
      mediaType: { format: "full+json" },
    });
    expect(fetchSpy).toHaveBeenCalledWith(signedUrl);
    expect(fsWriteFileSpy).toHaveBeenCalledWith(
      "/tmp/github-images/image-1704067200000-0.png",
      Buffer.from(mockArrayBuffer),
    );
    expect(result.size).toBe(1);
    expect(result.get(imageUrl)).toBe(
      "/tmp/github-images/image-1704067200000-0.png",
    );
    expect(consoleLogSpy).toHaveBeenCalledWith(
      "Found 1 image(s) in issue_comment 777",
    );
    expect(consoleLogSpy).toHaveBeenCalledWith(`Downloading ${imageUrl}...`);
    expect(consoleLogSpy).toHaveBeenCalledWith(
      "✓ Saved: /tmp/github-images/image-1704067200000-0.png",
    );
  });
  test("should handle HTML img tags with different quote styles", async () => {
    const mockOctokit = createMockOctokit();
    const imageUrl1 =
      "https://github.com/user-attachments/assets/single-quote.jpg";
    const imageUrl2 =
      "https://github.com/user-attachments/assets/double-quote.png";
    const signedUrl1 =
      "https://private-user-images.githubusercontent.com/single.jpg?jwt=token1";
    const signedUrl2 =
      "https://private-user-images.githubusercontent.com/double.png?jwt=token2";
    // @ts-expect-error Mock implementation doesn't match full type signature
    mockOctokit.rest.issues.getComment = jest.fn().mockResolvedValue({
      data: {
        body_html: `<img src="${signedUrl1}"><img src="${signedUrl2}">`,
      },
    });
    fetchSpy = spyOn(global, "fetch").mockResolvedValue({
      ok: true,
      arrayBuffer: async () => new ArrayBuffer(8),
    } as Response);
    const comments: CommentWithImages[] = [
      {
        type: "issue_comment",
        id: "888",
        body: `Single quote: <img src='${imageUrl1}' alt="test"> and double quote: <img src="${imageUrl2}" alt="test">`,
      },
    ];
    const result = await downloadCommentImages(
      mockOctokit,
      "owner",
      "repo",
      comments,
    );
    expect(fetchSpy).toHaveBeenCalledTimes(2);
    expect(result.size).toBe(2);
    expect(result.get(imageUrl1)).toBe(
      "/tmp/github-images/image-1704067200000-0.jpg",
    );
    expect(result.get(imageUrl2)).toBe(
      "/tmp/github-images/image-1704067200000-1.png",
    );
    expect(consoleLogSpy).toHaveBeenCalledWith(
      "Found 2 image(s) in issue_comment 888",
    );
  });
  test("should handle mixed Markdown and HTML images", async () => {
    const mockOctokit = createMockOctokit();
    const markdownUrl =
      "https://github.com/user-attachments/assets/markdown.png";
    const htmlUrl = "https://github.com/user-attachments/assets/html.jpg";
    const signedUrl1 =
      "https://private-user-images.githubusercontent.com/md.png?jwt=token1";
    const signedUrl2 =
      "https://private-user-images.githubusercontent.com/html.jpg?jwt=token2";
    // @ts-expect-error Mock implementation doesn't match full type signature
    mockOctokit.rest.issues.getComment = jest.fn().mockResolvedValue({
      data: {
        body_html: `<img src="${signedUrl1}"><img src="${signedUrl2}">`,
      },
    });
    fetchSpy = spyOn(global, "fetch").mockResolvedValue({
      ok: true,
      arrayBuffer: async () => new ArrayBuffer(8),
    } as Response);
    const comments: CommentWithImages[] = [
      {
        type: "issue_comment",
        id: "999",
        body: `Markdown: ![test](${markdownUrl}) and HTML: <img src="${htmlUrl}" alt="test">`,
      },
    ];
    const result = await downloadCommentImages(
      mockOctokit,
      "owner",
      "repo",
      comments,
    );
    expect(fetchSpy).toHaveBeenCalledTimes(2);
    expect(result.size).toBe(2);
    expect(result.get(markdownUrl)).toBe(
      "/tmp/github-images/image-1704067200000-0.png",
    );
    expect(result.get(htmlUrl)).toBe(
      "/tmp/github-images/image-1704067200000-1.jpg",
    );
    expect(consoleLogSpy).toHaveBeenCalledWith(
      "Found 2 image(s) in issue_comment 999",
    );
  });
  test("should deduplicate identical URLs from Markdown and HTML", async () => {
    const mockOctokit = createMockOctokit();
    const imageUrl = "https://github.com/user-attachments/assets/duplicate.png";
    const signedUrl =
      "https://private-user-images.githubusercontent.com/dup.png?jwt=token";
    // @ts-expect-error Mock implementation doesn't match full type signature
    mockOctokit.rest.issues.getComment = jest.fn().mockResolvedValue({
      data: {
        body_html: `<img src="${signedUrl}">`,
      },
    });
    fetchSpy = spyOn(global, "fetch").mockResolvedValue({
      ok: true,
      arrayBuffer: async () => new ArrayBuffer(8),
    } as Response);
    const comments: CommentWithImages[] = [
      {
        type: "issue_comment",
        id: "1000",
        body: `Same image twice: ![test](${imageUrl}) and <img src="${imageUrl}" alt="test">`,
      },
    ];
    const result = await downloadCommentImages(
      mockOctokit,
      "owner",
      "repo",
      comments,
    );
    expect(fetchSpy).toHaveBeenCalledTimes(1); // Only downloaded once
    expect(result.size).toBe(1);
    expect(result.get(imageUrl)).toBe(
      "/tmp/github-images/image-1704067200000-0.png",
    );
    expect(consoleLogSpy).toHaveBeenCalledWith(
      "Found 1 image(s) in issue_comment 1000",
    );
  });
  test("should handle HTML img tags with additional attributes", async () => {
    const mockOctokit = createMockOctokit();
    const imageUrl =
      "https://github.com/user-attachments/assets/complex-tag.webp";
    const signedUrl =
      "https://private-user-images.githubusercontent.com/complex.webp?jwt=token";
    // @ts-expect-error Mock implementation doesn't match full type signature
    mockOctokit.rest.issues.getComment = jest.fn().mockResolvedValue({
      data: {
        body_html: `<img src="${signedUrl}">`,
      },
    });
    fetchSpy = spyOn(global, "fetch").mockResolvedValue({
      ok: true,
      arrayBuffer: async () => new ArrayBuffer(8),
    } as Response);
    const comments: CommentWithImages[] = [
      {
        type: "issue_comment",
        id: "1001",
        body: `Complex tag: <img class="image" src="${imageUrl}" alt="test image" width="100" height="200">`,
      },
    ];
    const result = await downloadCommentImages(
      mockOctokit,
      "owner",
      "repo",
      comments,
    );
    expect(fetchSpy).toHaveBeenCalledTimes(1);
    expect(result.size).toBe(1);
    expect(result.get(imageUrl)).toBe(
      "/tmp/github-images/image-1704067200000-0.webp",
    );
    expect(consoleLogSpy).toHaveBeenCalledWith(
      "Found 1 image(s) in issue_comment 1001",
    );
  });
 });
--- a/test/install-mcp-server.test.ts
+++ b/test/install-mcp-server.test.ts
@@ -37,6 +37,7 @@ describe("prepareMcpConfig", () => {
      useStickyComment: false,
      additionalPermissions: new Map(),
      useCommitSigning: false,
      allowedBots: "",
    },
  };
@@ -547,8 +548,8 @@ describe("prepareMcpConfig", () => {
  });
  test("should include github_ci server when context.isPR is true and actions:read permission is granted", async () => {
-    const oldEnv = process.env.ACTIONS_TOKEN;
+    const oldEnv = process.env.DEFAULT_WORKFLOW_TOKEN;
-    process.env.ACTIONS_TOKEN = "workflow-token";
+    process.env.DEFAULT_WORKFLOW_TOKEN = "workflow-token";
    const contextWithPermissions = {
      ...mockPRContext,
@@ -575,7 +576,7 @@ describe("prepareMcpConfig", () => {
    expect(parsed.mcpServers.github_ci.env.PR_NUMBER).toBe("456");
    expect(parsed.mcpServers.github_file_ops).toBeDefined();
-    process.env.ACTIONS_TOKEN = oldEnv;
+    process.env.DEFAULT_WORKFLOW_TOKEN = oldEnv;
  });
  test("should not include github_ci server when context.isPR is false", async () => {
@@ -595,8 +596,8 @@ describe("prepareMcpConfig", () => {
  });
  test("should not include github_ci server when actions:read permission is not granted", async () => {
-    const oldTokenEnv = process.env.ACTIONS_TOKEN;
+    const oldTokenEnv = process.env.DEFAULT_WORKFLOW_TOKEN;
-    process.env.ACTIONS_TOKEN = "workflow-token";
+    process.env.DEFAULT_WORKFLOW_TOKEN = "workflow-token";
    const result = await prepareMcpConfig({
      githubToken: "test-token",
@@ -612,12 +613,12 @@ describe("prepareMcpConfig", () => {
    expect(parsed.mcpServers.github_ci).not.toBeDefined();
    expect(parsed.mcpServers.github_file_ops).toBeDefined();
-    process.env.ACTIONS_TOKEN = oldTokenEnv;
+    process.env.DEFAULT_WORKFLOW_TOKEN = oldTokenEnv;
  });
  test("should parse additional_permissions with multiple lines correctly", async () => {
-    const oldTokenEnv = process.env.ACTIONS_TOKEN;
+    const oldTokenEnv = process.env.DEFAULT_WORKFLOW_TOKEN;
-    process.env.ACTIONS_TOKEN = "workflow-token";
+    process.env.DEFAULT_WORKFLOW_TOKEN = "workflow-token";
    const contextWithPermissions = {
      ...mockPRContext,
@@ -644,12 +645,12 @@ describe("prepareMcpConfig", () => {
    expect(parsed.mcpServers.github_ci).toBeDefined();
    expect(parsed.mcpServers.github_ci.env.GITHUB_TOKEN).toBe("workflow-token");
-    process.env.ACTIONS_TOKEN = oldTokenEnv;
+    process.env.DEFAULT_WORKFLOW_TOKEN = oldTokenEnv;
  });
  test("should warn when actions:read is requested but token lacks permission", async () => {
-    const oldTokenEnv = process.env.ACTIONS_TOKEN;
+    const oldTokenEnv = process.env.DEFAULT_WORKFLOW_TOKEN;
-    process.env.ACTIONS_TOKEN = "invalid-token";
+    process.env.DEFAULT_WORKFLOW_TOKEN = "invalid-token";
    const contextWithPermissions = {
      ...mockPRContext,
@@ -677,6 +678,6 @@ describe("prepareMcpConfig", () => {
      ),
    );
-    process.env.ACTIONS_TOKEN = oldTokenEnv;
+    process.env.DEFAULT_WORKFLOW_TOKEN = oldTokenEnv;
  });
 });
--- a/test/mockContext.ts
+++ b/test/mockContext.ts
@@ -28,6 +28,7 @@ const defaultInputs = {
  useStickyComment: false,
  additionalPermissions: new Map<string, string>(),
  useCommitSigning: false,
  allowedBots: "",
 };
 const defaultRepository = {
--- a/test/permissions.test.ts
+++ b/test/permissions.test.ts
@@ -73,6 +73,7 @@ describe("checkWritePermissions", () => {
      useStickyComment: false,
      additionalPermissions: new Map(),
      useCommitSigning: false,
      allowedBots: "",
    },
  });
@@ -126,6 +127,16 @@ describe("checkWritePermissions", () => {
    );
  });
  test("should return true for bot user", async () => {
    const mockOctokit = createMockOctokit("none");
    const context = createContext();
    context.actor = "test-bot[bot]";
    const result = await checkWritePermissions(mockOctokit, context);
    expect(result).toBe(true);
  });
  test("should throw error when permission check fails", async () => {
    const error = new Error("API error");
    const mockOctokit = {
--- a/test/trigger-validation.test.ts
+++ b/test/trigger-validation.test.ts
@@ -41,6 +41,7 @@ describe("checkContainsTrigger", () => {
          useStickyComment: false,
          additionalPermissions: new Map(),
          useCommitSigning: false,
          allowedBots: "",
        },
      });
      expect(checkContainsTrigger(context)).toBe(true);
@@ -74,6 +75,7 @@ describe("checkContainsTrigger", () => {
          useStickyComment: false,
          additionalPermissions: new Map(),
          useCommitSigning: false,
          allowedBots: "",
        },
      });
      expect(checkContainsTrigger(context)).toBe(false);
@@ -291,6 +293,7 @@ describe("checkContainsTrigger", () => {
          useStickyComment: false,
          additionalPermissions: new Map(),
          useCommitSigning: false,
          allowedBots: "",
        },
      });
      expect(checkContainsTrigger(context)).toBe(true);
@@ -325,6 +328,7 @@ describe("checkContainsTrigger", () => {
          useStickyComment: false,
          additionalPermissions: new Map(),
          useCommitSigning: false,
          allowedBots: "",
        },
      });
      expect(checkContainsTrigger(context)).toBe(true);
@@ -359,6 +363,7 @@ describe("checkContainsTrigger", () => {
          useStickyComment: false,
          additionalPermissions: new Map(),
          useCommitSigning: false,
          allowedBots: "",
        },
      });
      expect(checkContainsTrigger(context)).toBe(false);
Author	SHA1	Message	Date
Steve	6d5c92076b	non negative line validation for comment server (#429 ) * enforce non-negative validation for line in GH comment server * include .nonnegative() for startLine too	2025-08-08 08:36:20 -07:00
Yuku Kotani	fec554fc7c	feat: add flexible bot access control with allowed_bots option (#117 ) * feat: skip permission check for GitHub App bot users GitHub Apps (users ending with [bot]) now bypass permission checks as they have their own authorization mechanism. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> * feat: add allow_bot_users option to control bot user access - Add allow_bot_users input parameter (default: false) - Modify checkHumanActor to optionally allow bot users - Add comprehensive tests for bot user handling - Improve security by blocking bot users by default This change prevents potential prompt injection attacks from bot users while providing flexibility for trusted bot integrations. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> * docs: mark bot user support feature as completed in roadmap 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> * refactor: move allowedBots parameter to context object Move allowedBots from function parameter to context.inputs to maintain consistency with other input handling throughout the codebase. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> * docs: update README for bot user support feature Add documentation for the new allowed_bots parameter that enables bot users to trigger Claude actions with granular control. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> * fix: add missing allowedBots property in permissions test 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> * fix: update bot name format to include [bot] suffix in tests and docs - Update test cases to use correct bot actor names with [bot] suffix - Update documentation example to show correct bot name format - Align with GitHub's actual bot naming convention 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> * feat: normalize bot names for allowed_bots validation - Strip [bot] suffix from both actor names and allowed bot list for comparison - Allow both "dependabot" and "dependabot[bot]" formats in allowed_bots input - Display normalized bot names in error messages for consistency - Add comprehensive test coverage for both naming formats 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com>	2025-08-07 18:03:20 -07:00
GitHub Actions	59ca6e42d9	chore: bump Claude Code version to 1.0.71	2025-08-07 22:57:57 +00:00
Aner Cohen	7afc848186	fix: improve GitHub suggestion guidelines in review mode to prevent code duplication (#422 ) * fix: prevent duplicate function signatures in review mode suggestions This fixes a critical bug in the experimental review mode where GitHub suggestions could create duplicate function signatures when applied. The issue occurred because: - GitHub suggestions REPLACE the entire selected line range - Claude wasn't aware of this behavior and would include the function signature in multi-line suggestions, causing duplication Changes: - Added detailed instructions about GitHub's line replacement behavior - Provided clear examples for single-line vs multi-line suggestions - Added explicit warnings about common mistakes (duplicate signatures) - Improved code readability by using a codeBlock variable instead of escaped backticks in template strings This ensures Claude creates syntactically correct suggestions that won't break code when applied through GitHub's suggestion feature. * chore: format	2025-08-07 08:56:30 -07:00
Graham Campbell	6debac392b	Go with Opus 4.1 (#420 )	2025-08-06 21:22:15 -07:00
GitHub Actions	55fb6a96d0	chore: bump Claude Code version to 1.0.70	2025-08-06 19:59:40 +00:00
Ashwin Bhat	15db2b3c79	feat: add inline comment MCP server for experimental review mode (#414 ) * feat: add inline comment MCP server for experimental review mode - Create standalone inline PR comments without review workflow - Support single-line and multi-line comments - Auto-install server when in experimental review mode - Uses octokit.rest.pulls.createReviewComment() directly * docs: clarify GitHub code suggestion syntax in inline comment server Add clear documentation that suggestion blocks replace the entire selected line range and must be syntactically complete drop-in replacements. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com>	2025-08-06 08:21:29 -07:00
Ashwin Bhat	188d526721	refactor: change git hook from pre-push to pre-commit (#401 ) - Renamed scripts/pre-push to scripts/pre-commit - Updated install-hooks.sh to install pre-commit hook - Hook now runs formatting, type checking, and tests before commit	2025-08-05 17:02:34 -07:00
Ashwin Bhat	a519840051	fix: remove git config user.name and user.email from allowed tools (#410 ) These git config commands are no longer needed as allowed tools since Claude should not be modifying git configuration settings. Updated the corresponding test to reflect this intentional change. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-authored-by: Claude <noreply@anthropic.com>	2025-08-05 11:32:46 -07:00
yoshikouki	85287e957d	fix: restore prompt file creation in agent mode (#405 ) - Restore prompt file creation logic that was accidentally removed in PR #374 - Agent mode now creates the prompt file directly in prepare() method - Uses override_prompt or direct_prompt if available, falls back to minimal prompt - Fixes 'Prompt file does not exist' error for workflow_dispatch and schedule events - Add TODO comment to refactor this to use createPrompt in the future Fixes #403	2025-08-05 11:14:28 -07:00
GitHub Actions	c6a07895d7	chore: bump Claude Code version to 1.0.69	2025-08-05 16:50:23 +00:00
atsushi-ishibashi	0c5d54472f	feat: Add HTML img tag support to GitHub image downloader (#402 ) * feat: support html img tag * rm files * refactor	2025-08-04 19:37:50 -07:00
GitHub Actions	2845685880	chore: bump Claude Code version to 1.0.68	2025-08-04 23:29:44 +00:00
Ashwin Bhat	b39377f9bc	feat: add getSystemPrompt method to mode interface (#400 ) Allows modes to provide custom system prompts that are appended to Claude's base system prompt. This enables mode-specific instructions without modifying the core action logic. - Add optional getSystemPrompt method to Mode interface - Implement method in all existing modes (tag, agent, review) - Update prepare.ts to call getSystemPrompt and export as env var - Wire up APPEND_SYSTEM_PROMPT in action.yml to pass to base-action All modes currently return undefined (no additional prompts), but the infrastructure is now in place for future modes to provide custom instructions.	2025-08-04 10:51:30 -07:00
Matthew Burke	618565bc0e	Update documentation incorrectly reverted after refactor (#399 )	2025-08-04 09:00:22 -07:00
Ashwin Bhat	0d9513b3b3	refactor: restructure documentation into organized docs directory (#383 ) - Move FAQ.md to docs/faq.md - Create structured documentation files: - setup.md: Manual setup and custom GitHub app instructions - usage.md: Basic usage and workflow configuration - custom-automations.md: Automation examples - configuration.md: MCP servers and advanced settings - experimental.md: Execution modes and network restrictions - cloud-providers.md: AWS Bedrock and Google Vertex setup - capabilities-and-limitations.md: Features and constraints - security.md: Security information - Condense README.md to overview with links to detailed docs - Keep CONTRIBUTING.md, SECURITY.md, CODE_OF_CONDUCT.md at top level	2025-08-03 21:16:50 -07:00
km-anthropic	458e4b9e7f	feat: ship slash commands with GitHub Action (#381 ) * feat: add slash command shipping infrastructure - Created /slash-commands/ directory to store bundled slash commands - Added code-review.md slash command for automated PR reviews - Modified setup-claude-code-settings.ts to copy slash commands to ~/.claude/ - Added test coverage for slash command installation - Commands are automatically installed when the GitHub Action runs * fix: simplify slash command implementation to match codebase patterns - Reverted to using Bun's $ shell syntax consistently with the rest of the codebase - Simplified slash command copying to basic shell commands - Removed unnecessary fs/promises complexity - Maintained all functionality and test coverage - More appropriate for GitHub Action context where inputs are trusted * remove test slash command * fix: rename slash_commands_dir to experimental_slash_commands_dir - Added 'experimental' prefix as suggested by Ashwin - Updated all references in action.yml and base-action - Restored accidentally removed code-review.md file --------- Co-authored-by: km-anthropic <km-anthropic@users.noreply.github.com>	2025-08-03 21:05:33 -07:00
Ashwin Bhat	d66adfb7fa	refactor: rename ACTIONS_TOKEN to DEFAULT_WORKFLOW_TOKEN (#385 ) Updated all references from ACTIONS_TOKEN to DEFAULT_WORKFLOW_TOKEN to match the naming convention used in action.yml where the GitHub token is passed as DEFAULT_WORKFLOW_TOKEN environment variable.	2025-08-02 21:26:52 -07:00
GitHub Actions	d829b4d14b	chore: bump Claude Code version to 1.0.67	2025-08-01 22:56:22 +00:00
km-anthropic	0a78530f89	docs: clarify agent mode only works with workflow_dispatch and schedule events (#378 ) * docs: clarify agent mode only works with workflow_dispatch and schedule events Updates documentation to match the current implementation where agent mode is restricted to workflow_dispatch and schedule events only. This addresses the confusion reported in issues #364 and #376. Changes: - Updated README to clearly state agent mode limitations - Added explicit note that agent mode does NOT work with PR/issue events - Updated example workflows to only show supported event types - Updated CLAUDE.md internal documentation Fixes #364 Fixes #376 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> * minor formatting update * update agent mode docs --------- Co-authored-by: km-anthropic <km-anthropic@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com>	2025-08-01 15:47:53 -07:00
GitHub Actions	20e09ef881	chore: bump Claude Code version to 1.0.65	2025-08-01 22:28:48 +00:00