PR Review Response

Input

https://github.com/{owner}/{repo}/pull/{number}

• owner

  : Repository owner

• repo

  : Repository name

• number

  : PR number

Extracting Comments

gh

1. Review Comments (inline comments on code)

gh api repos/{owner}/{repo}/pulls/{number}/comments

• path

  : File path the comment references

• line

  or

  original_line

  : Line number

• body

  : Comment text

• diff_hunk

  : Code context

2. PR Reviews (approval/request changes with summary)

gh api repos/{owner}/{repo}/pulls/{number}/reviews

• body

  : Review summary

• state

  : APPROVED, CHANGES_REQUESTED, COMMENTED

3. General PR Comments (conversation thread)

gh api repos/{owner}/{repo}/issues/{number}/comments

Processing Comments

1. Group by file: Organize inline comments by their

   path

   field
2. Classify feedback by intent:
   • Change requests — explicit directives to modify code (e.g. "please change", "should be", "fix", "update")
   • Open questions — inquiries that require investigation or clarification
   • Enhancement suggestions — non-blocking recommendations for improvement
3. Prioritize: Address blocking comments (from CHANGES_REQUESTED reviews) first

Acting on Feedback

1. Read the referenced file to understand the current code
2. Fix problems at the root: Don't apply superficial fixes. Investigate and address the underlying cause of the issue, not just the symptom mentioned in the comment
3. Apply requested changes using the Edit tool
4. For questions: Investigate and provide answers or make clarifying changes
5. For suggestions: Evaluate and implement if appropriate
6. Add tests for edge cases: If the feedback identifies an edge case and the project has testing already set up, add a test to cover that edge case to prevent regression

Example Workflow

https://github.com/acme/app/pull/42

1. Extract: owner=

   acme

   , repo=

   app

   , number=

   42

2. Fetch comments:
   bash

   gh api repos/acme/app/pulls/42/comments
   gh api repos/acme/app/pulls/42/reviews
   gh api repos/acme/app/issues/42/comments

3. Parse the JSON responses to identify feedback
4. For each comment with a file path, read that file and apply changes
5. Summarize what was addressed

Output

• Number of comments found
• Actions taken for each comment
• Any comments that need clarification or manual review