GitHub PRs

Overview

Guidelines for AI agents handling pull request interactions. Agents must identify themselves and use MCP tools for GitHub operations.

When to Use

• Responding to PR review comments
• Replying to PR discussion comments
• Updating PR based on feedback
• Checking PR status or reviews

Agent Identification

Always identify yourself as an AI agent when commenting on PRs.

Use this signature at the end of PR comments:

— Claude

MCP Tools for PRs

Use the GitHub MCP server tools instead of gh CLI when available:

| Task | MCP Tool | |------|----------| | Read PR details | mcp__github__pull_request_read with method: "get" | | Get PR diff | mcp__github__pull_request_read with method: "get_diff" | | Get PR comments | mcp__github__pull_request_read with method: "get_comments" | | Get review comments | mcp__github__pull_request_read with method: "get_review_comments" | | Get PR checks/status | mcp__github__pull_request_read with method: "get_status" | | Add PR comment | mcp__github__add_issue_comment (PRs are issues) | | Create review | mcp__github__pull_request_review_write with method: "create" | | Update PR | mcp__github__update_pull_request |

Reporting PR Checks

When asked about PR checks or status, use pull_request_read with method: "get_status" and display results in a table:

| Check | Status | Details |
|-------|--------|---------|
| Validate PR Title | :white_check_mark: Passed | - |
| Build | :white_check_mark: Passed | - |
| Lint | :x: Failed | `src/auth.py:42` - Line too long |
| Tests | :hourglass: Pending | Running... |

Status icons:

• :white_check_mark: - Passed/Success
• :x: - Failed
• :hourglass: - Pending/In Progress
• :warning: - Skipped or neutral

Responding to PR Comments

1. Read the comment using pull_request_read with method: "get_comments" or "get_review_comments"
2. Understand the feedback - what change is being requested?
3. Make the fix if it's a code change request
4. Reply to acknowledge using add_issue_comment
5. Include agent signature in your response

Example Response

Thanks for the feedback! I've updated the code to address your concern.

Changes made:
- Fixed the validation logic in `auth.py:45`
- Added the missing error handling

— Claude

Handling Review Feedback

When a PR review requests changes:

1. Read review comments with pull_request_read method "get_review_comments"
2. Address each comment with code fixes
3. Commit and push fixes
4. Reply to each review thread if possible, or add a summary comment

Summary Comment Template

## Addressed Review Feedback

| Comment | Resolution |
|---------|------------|
| [Brief description] | [What was fixed] |
| [Brief description] | [What was fixed] |

All feedback has been addressed. Ready for re-review.

— Claude

Key Principles

1. Always identify as AI - transparency about agent-generated responses
2. Use MCP tools - prefer MCP GitHub tools over CLI when available
3. Acknowledge feedback - always respond to comments, even if no code change needed
4. Be concise - keep responses focused and actionable
5. Reference changes - point to specific files/lines when discussing fixes