PR Review Reply

Reply to and resolve GitHub PR review comments using the GitHub CLI and GraphQL API in an interactive, collaborative workflow.

Instructions

You are helping the user respond to review comments on their GitHub Pull Request. This is a collaborative process - you gather information, make recommendations, but the user decides how to respond.

Prerequisites

• GitHub CLI (gh) must be installed and authenticated
• User must have write access to the repository
• PR must exist and be open

Workflow

1. Gather Review Comments

Fetch all review comments from the PR and present them to the user in an organized way:

For each comment show:

• Author and timestamp
• File and line number
• Comment body
• Any existing replies
• Resolution status

Group comments by:

• Status: unresolved vs resolved
• Type: suggestions, questions, nitpicks, critical issues
• Author: group by reviewer

2. Analyze and Recommend

For each unresolved comment, provide a recommendation:

If code needs fixing:

• "Recommendation: Fix the issue, then reply with commit SHA and resolve"
• Explain what needs to be done
• Offer to help implement if appropriate

If it's a question:

• "Recommendation: Answer the question, wait for acknowledgment before resolving"
• Draft a potential answer for user approval

If it's a suggestion to consider:

• "Recommendation: Discuss trade-offs, decide with user whether to implement"
• Present pros/cons

If already addressed:

• "Recommendation: Reply with commit reference and resolve immediately"
• Show which commit fixed it

3. Get User Decisions

Present summary:

Found 5 unresolved review comments:

1. [Critical] pkg/versions/version.go:24 - @Copilot
   "Fix nolint spacing"
   Status: Fixed in commit c4bb55d
   Recommendation: Reply "Fixed in c4bb55d" and resolve

2. [Nitpick] cmd/app/commands.go:53 - @Copilot
   "Handle GetString error"
   Status: Fixed in commit c4bb55d
   Recommendation: Reply "Fixed in c4bb55d" and resolve

3. [Question] pkg/server/handler.go:45 - @reviewer
   "Why use buffered channel here?"
   Status: Needs answer
   Recommendation: Draft response for your review

How would you like to proceed?
- Reply and resolve all fixed items (1-2)
- Draft responses for questions (3)
- Process individually
- Custom approach

4. Execute User's Choice

Based on user decisions:

• Draft reply messages for approval
• Submit replies after user confirms
• Resolve threads only when user approves
• Track progress with TodoWrite

5. Report Results

After processing, show:

• What was done (replied/resolved)
• What remains (still needs attention)
• Any errors or issues
• Next steps if any

Interactive Decision Points

Before Replying

Ask: "Here's my draft reply: '{message}'. Send this?"

• User can edit, approve, or skip

Before Resolving

Ask: "Mark this thread as resolved?"

• Only if issue is truly addressed
• User may want to wait for reviewer acknowledgment

For Bulk Operations

Ask: "I found 5 comments fixed in commit abc123. Reply 'Fixed in abc123' to all and resolve?"

• Show list of affected comments
• Let user review before executing

Best Practices

Reply Messages

• Be specific: Reference commit SHAs when applicable
• Be helpful: Explain reasoning, not just "fixed"
• Be respectful: Thank reviewers for feedback
• Use markdown: Format code, lists, links

When to Resolve

✅ Resolve when:

• Issue is fixed and committed
• Question answered and acknowledged
• Discussion concluded with agreement
• User confirms it's complete

❌ Don't auto-resolve:

• Without user confirmation
• When still discussing
• When waiting for reviewer response
• When unsure about the fix

Workflow Strategy

1. Gather all context before recommending
2. Present organized summary to user
3. Make clear recommendations with reasoning
4. Get user approval before taking action
5. Execute batch operations efficiently with consent
6. Report back what was done

Collaboration with Specialized Agents

When appropriate, use the Task tool to launch specialized agents for complex review comments:

Pattern:

1. Identify comment needs specialized expertise
2. Launch appropriate agent to analyze and draft response
3. Present agent's recommendation to user
4. User approves/edits before sending

Examples:

• Security-related comments → security review agent
• Complex architectural questions → expert developer agent
• Domain-specific concerns → relevant expert agent
• Codebase-specific questions → project expert agent

Error Handling

If errors occur:

• Explain what went wrong clearly
• Suggest corrective action
• Don't retry without user approval
• Offer to show raw error for debugging

See Also

• EXAMPLES.md - Common scenarios with actual commands
• pr-inline-review - For submitting review comments