/address-review
Respond to the review comments on a PR — yours or the one for the current branch — without letting a design disagreement get silently coded away.
What it does
Invokes the address-review skill, which:
- Resolves the PR (current branch by default, or
#N) and fetches all review comments via the GitHub MCP — every reviewer, human and bot. - Classifies each comment: convergent (a clear fix) vs divergent (a challenge to the approach). When uncertain → divergent.
- Shows you the classification table so you can re-bucket anything.
- Convergent → drafts the minimal fix (one apply-agent per comment), runs
/verify, applies after your sign-off. - Divergent → escalates back to you as a question; on your decision, either applies the change or replies on the reviewer's thread with the rationale — never a silent edit.
- Closes the loop with a reply on each reviewer thread; reports per repo. Does not merge, does not auto-resolve threads.
Arguments
$ARGUMENTS — optional PR selector:
- nothing → the open PR for the current branch.
#N→ a specific PR.- a comment/thread reference → restrict to that one comment.
When to use vs. siblings
- This — you have review comments to respond to. It covers both cases: a human-reviewed PR with possible design disagreements, and the bulk "just apply the bot's accepted suggestions" case (the fast path — bot comments classify convergent and apply in one batch; no separate command).
/pr-review— the producing side: review someone's PR. This command is the consuming side: respond to a review of yours.
Cross-references
- Underlying skill:
address-review - Methodology: Human Review in an AI Workflow — the convergent/divergent discipline this command enforces.
/continuestep 11 — the loop step this command wires.