code-review-loop

code-review-loop

Independence

This skill MUST NOT invoke or delegate to any superpowers:* skill.

Purpose

Handles the full code-review cycle in one skill: request → receive → verify → fix → re-review. Replaces the former separate requesting-code-review and receiving-code-review skills.

Applicability

• T0/T1: Do NOT trigger unless the user explicitly requests review.
• T2/T3: Run spec-coexist:pre-review-self-check before this skill.

Procedure

Phase 1 — Request review

1. Run scripts/collect-review-context.sh to get BASE_SHA / HEAD_SHA.
2. Prepare WHAT_WAS_IMPLEMENTED, PLAN_OR_REQUIREMENTS, DESCRIPTION.
3. Dispatch a fresh subagent using the template in code-reviewer.md.
4. Wait for the reviewer's structured response.

Phase 2 — Process feedback

5. Read the entire feedback without reacting.
6. Verify each item against the actual codebase — does the problem exist?
7. Evaluate technical correctness for this codebase.
8. Respond with a fix or reasoned pushback (see references/pushback-rules.md).
9. Implement one item at a time in severity order (Critical → Important → Minor/Suggestion). Run tests after each fix.

Phase 3 — Re-review if needed

10. After Critical/Important fixes, re-dispatch on the new HEAD_SHA.
11. Record the outcome with a Review: <verdict> line.

Forbidden Responses

Do NOT respond with performative agreement ("Great point!", "You're right!", "Thanks for catching!"). Fix the code — that is the acknowledgment.

Flow

flowchart TD
    Start([Code ready]) --> Collect[collect-review-context.sh]
    Collect --> Dispatch[Dispatch reviewer subagent]
    Dispatch --> Read[Read feedback]
    Read --> Verify[Verify each item]
    Verify --> Fix[Fix by severity order]
    Fix --> Test[Run tests after each fix]
    Test --> More{Critical/Important<br/>remain?}
    More -- Yes --> Fix
    More -- No --> Recheck{Re-review needed?}
    Recheck -- Yes --> Dispatch
    Recheck -- No --> Done([Review: verdict])

References

• references/why-fresh-subagent.md — rationale for fresh subagent
• references/mandatory-triggers.md — events that trigger review
• references/severity-policy.md — severity thresholds
• references/pushback-rules.md — when and how to push back
• references/review-response-protocol.md — severity-ordering, YAGNI check
• references/source-handling.md — trust levels per feedback source

Scripts

• scripts/collect-review-context.sh [--base <sha>] — emits BASE_SHA and HEAD_SHA
• scripts/sort_feedback_items.sh <feedback-file> — classifies items by severity