| name | pr-respond |
| description | Draft brief, professional responses to GitHub PR review comments. Use /pr-respond with a PR number to fetch unresolved comments and craft replies. |
| allowed-tools | ["Bash","Read","Grep","Glob"] |
PR Review Response
Draft brief, professional responses to PR review comments.
Workflow
- Run the fetch script (auto-detects PR from current branch, or accepts PR number):
./fetch-unresolved-comments.sh [PR_NUMBER]
- Read relevant code files for context
- Draft responses for each comment
Response Style
Human reviewers: Brief and professional. 1-2 sentences max.
- "Fixed." / "Good catch, fixed."
- "Done, switched to [approach]."
- "Intentional - [one line reason]."
- "Can you clarify what you mean by X?"
AI bot reviewers (CodeRabbit, Copilot, etc.):
- Fixing: Ultra-brief - "Fixed." / "Done."
- Not fixing: Brief explanation - "Won't fix - [reason]" or "Intentional - [reason]"
- Nitpicks: Skip or "N/A"
Output Format
**[file:line]** (@reviewer)
> [their comment summary]
[your response]
---
Task Management
For PRs with multiple review comments:
- Use TodoWrite to create a task for each unresolved comment thread
- Mark tasks as in_progress when drafting each response
- Mark completed after drafting (user will copy/post responses)
Example todo structure:
- "Respond to @reviewer on src/file.ts:42"
- "Respond to @coderabbit on src/utils.ts:15"
Sub-Agents
Use Task tool with Explore agent when you need to:
- Understand unfamiliar code referenced in review comments
- Find related implementations to justify design decisions
- Search for patterns/conventions used elsewhere in the codebase
Keep exploration focused - only spawn agents when context from a simple file read isn't sufficient.
Guidelines
- Keep responses short - respect reviewer's time
- AI bots get minimal acknowledgment unless the suggestion is substantive
- If disagreeing, one sentence of reasoning is enough
- Don't over-explain