jamesc/resolve-pr

PR Resolve Workflow

When activated, execute this workflow to systematically address all PR review comments:

Steps

1. Determine Issue ID per pick-issue step 1 (branch name → worktree name → ask user).

2. Enumerate ALL unresolved threads — never declare "no comments" without evidence:

   Fetch from all three sources and print a numbered list of everything found:

   Review threads (code-level, includes CodeRabbit and Copilot): Use the GitHub MCP pull_request_read tool with method get_review_comments. Capture threadId (node ID, e.g. PRRT_kwDO…) and isResolved/isOutdated for each.

   General PR comments (conversation-level): Use method get_comments. Includes bot summary comments from CodeRabbit and Copilot.

   Bot reviews explicitly: Also fetch reviews from named bots:

   gh api repos/{owner}/{repo}/pulls/{pr}/reviews --jq '.[] | select(.user.login | test("copilot|coderabbit|github-copilot"; "i")) | {id, state, user: .user.login}'
   

   RULE: You MUST show the raw count from each source before saying there is nothing to do. "0 threads, 0 general comments, 0 bot reviews" is the only acceptable "nothing to do" output.

3. Analyze and plan: For each review comment:

   • Understand what the reviewer is asking for
   • Determine if it needs a code fix, documentation, Linear issue, or just clarification
   • Create a todo list with all items to address

4. Run tests first: Verify current state passes checks:

   just test
   

5. Write tests for behavioral changes: For code fixes that change logic or behavior:

   • Write a failing test that demonstrates the bug or missing behavior
   • Run the test to confirm it fails as expected
   • This ensures the fix is verifiable and prevents regressions

   Skip tests for: Naming changes, documentation updates, style/formatting fixes, comment improvements — these don't need TDD.

6. Address each comment: For each item in the plan:

   • Make the necessary code changes to make the test pass
   • Run tests after each significant change to catch regressions early
   • If a comment requires a follow-up Linear issue (e.g., "TODO for later"):
     • Create the Linear issue with full context
     • Add a TODO comment in the code referencing the issue number
   • Mark the todo item complete

7. Run full test suite: After all changes:

   just ci
   

   (Use just test for quick iteration; just ci is the final gate before push.)

8. Commit changes: Stage and commit with a descriptive message (using issue ID from step 1):

   git add -A
   git commit -m "fix: address PR review comments BT-{number}
   
   - Summary of each fix
   - Reference any Linear issues created"
   

9. Push changes:

   git push
   

10. Reply to each comment: For every review comment that was addressed, add a reply explaining what was done:

gh api repos/{owner}/{repo}/pulls/{pr}/comments/{comment_id}/replies -f body="<explanation of fix, commit hash, any Linear issues created>"

Include: - Commit hash where the fix was made - Brief description of the change - Links to any Linear issues created for follow-up work

11. Resolve review threads: After replying, resolve each addressed review thread using the GraphQL API:

    gh api graphql -f query='mutation { resolveReviewThread(input: {threadId: "<thread_node_id>"}) { thread { isResolved } } }'
    

    The thread node ID (e.g., PRRT_kwDO...) is available from the get_review_comments response in step 2. Only resolve threads where the fix has been committed and pushed.

12. Final verification pass — re-fetch all threads after pushing and confirm completeness:

    # Re-fetch review threads and check none are still unresolved
    gh api repos/{owner}/{repo}/pulls/{pr}/comments --jq '[.[] | {id, path, resolved: false}] | length'
    # Re-fetch bot reviews
    gh api repos/{owner}/{repo}/pulls/{pr}/reviews --jq '[.[] | select(.state == "CHANGES_REQUESTED")] | length'
    

    Also re-run the GitHub MCP get_review_comments to confirm isResolved: true on every thread addressed.

    Do not proceed to step 13 until every thread is verified resolved or explicitly acknowledged as a known skip (e.g. outdated).

13. Report summary: Provide a summary table of all comments and how they were resolved, with thread IDs and commit hashes.

14. Auto-chain to done: If all review comments have been successfully resolved (no failures, no pending issues), automatically activate the done skill:

    • Inform the user that all PR comments have been addressed
    • Activate the done skill without waiting for user confirmation
    • The done skill's bot-review gate (step 12) will re-verify that no unresolved Copilot / CodeRabbit findings remain before reporting success. If something was missed here, the gate will halt there.

    If there are any issues or manual steps needed, report them and wait for user input instead.