Skill: gh-address-comments
Purpose
Turn PR review comments (automated or human) into fixes, with exactly one human decision point: the triage. The human decides what deserves a fix; the execution is autonomous.
Step 1: Collect and assess
Run the comment-triage subagent for the current branch's PR. It fetches
top-level and inline comments, drops what's already resolved or addressed
by a later commit, and reads the actual code behind each remaining one to
form its own assessment. That reading is the expensive part; it happens in
the subagent's own context, not this session's. It returns a compact list:
comment, assessment, recommended keep/drop.
Step 2: Triage (the human decision)
Present the subagent's list to the user: for each open comment, its one-line summary and the assessment (agree/disagree/unclear, with why).
Then ask the user in ONE question (multiSelect) which findings to address. Recommend a default selection based on the subagent's assessment. Do not start fixing anything before this triage, even findings you are sure about.
Step 3: Fix (autonomous)
For each retained finding:
- Fix the cause, tests included when the finding revealed a coverage gap.
- Never weaken a gate, a test or the spec to satisfy a comment.
- Conventional commits: one commit per logical fix, or one grouped
fix(review): …commit for small related items.
Then push (flux-gate runs), and follow the CI watch loop from the gh-pr
skill until green.
Step 4: Close the loop
- Reply briefly on each addressed comment (what was done, commit sha) and on declined ones (why not, per the triage decision).
- Report: what was fixed, what was declined and why, CI status. Nothing
re-reviews on its own: run
/flux:reviewagain for a fresh pass on the new commits if one is wanted before merge, which stays with the human either way.