review-pr-comments-loop

Purpose

Combine PR-comment triage with a closing self-review: investigate each reviewer comment, implement fixes for the valid ones directly (no plan-mode gate — invoking this skill is the authorization), compile the rejection list for the reviewer, then run session-review-loop on the resulting diff so any bug the fixes themselves introduce gets caught and closed out before you're done.

Input

Same as review-pr-comments: comments passed as arguments after the slash command, or the most recent numbered/bulleted block of text in the conversation. Preserve the original numbering or bullet labels throughout.

Status line

The status line shows this skill's progress while it runs. Write the state file after every step below:

mkdir -p ~/.claude/skill-status
jq -n --arg skill "review-pr-comments-loop" --arg detail "<detail>" --argjson ts "$(date +%s)" \
  '{skill:$skill, detail:$detail, updated_at:$ts}' \
  > ~/.claude/skill-status/"$CLAUDE_CODE_SESSION_ID".json

Step 5 hands off to session-review-loop, which overwrites this same file with its own progress — don't clear it before that handoff. If this skill stops before reaching step 5 (nothing to fix, or aborted), clear it yourself: rm -f ~/.claude/skill-status/"$CLAUDE_CODE_SESSION_ID".json.

Steps

1. Parse and investigate

Write status: detail="triaging <N> comment(s)".

Extract each distinct review comment, assigning/preserving its original number. For every comment, read the relevant files and grep for the symbols, patterns, or logic the reviewer mentions — do not rely on memory or assumptions. Verdict per comment:

• Valid — the issue the reviewer describes genuinely exists in the current code. Fix it.
• Invalid — the code already handles this correctly, the reviewer misread the code, the concern doesn't apply, or the suggested change would be incorrect/harmful. Write a one- or two-sentence rebuttal referencing the specific file:line that disproves the comment.

2. Fix valid comments directly

Write status: detail="fixing <k>/<N> valid comments".

For each Valid comment, implement the fix in the repo now — no plan-mode gate, this skill runs to completion. Keep each fix scoped to that comment; no drive-by refactors.

3. Verify

Write status: detail="verifying fixes".

Run typecheck/lint/build and the relevant test suite for files touched in step 2. Note anything that fails for reasons unrelated to these fixes rather than blocking on it.

4. Save the reviewer rejection list

Save only the Invalid comments — numbered per their original labels, each with its rebuttal — to ~/.claude/plans/pr-review-triage-<YYYY-MM-DD>.md. This file is for copy-pasting back to the reviewer, not an execution plan. State the filename.

If every comment was Valid, note "All comments were valid — no rejections" instead of writing an empty file.

5. Hand off to session-review-loop

Write status: detail="handing off to session-review-loop" (it will immediately overwrite this with its own progress).

Invoke the session-review-loop skill now via the Skill tool. It independently re-reviews the full session diff — including the fixes just made in step 2 — and iterates, fixing any confirmed issue it finds, until a round comes back clean or its own round cap is hit. Do not duplicate its logic here; let it own its full flow, including clearing the status file when it finishes.

6. Final report

Report: comment-by-comment verdicts, the filename of the saved rejection list, and session-review-loop's own summary (rounds run, fixes applied per round, remaining advisory notes).

Constraints

• Do not invent fixes for comments you have not investigated.
• Keep rebuttals factual and collegial; cite file paths and line numbers where they strengthen the case.
• Adhere to the user's %% annotation convention: if %% markers appear in any file this skill touches, address them before finalising.