Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Triage and resolve GitHub PR review comments one by one, interactively. Use when the user asks to handle, address, respond to, or resolve PR review comments, or mentions reviewer feedback on a pull request.
.claude/skills/storybookjs-handle-pr-comments/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -46% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -61% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 5% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -55% | 0% |
| case-08 | ✓→✗ | ▼ Worse | -65% | 0% |
Evaluate review comments on the PR (from the current branch, or a PR number/URL if given). For each unresolved comment: summarize the feedback, suggest a resolution, ask the user how to resolve it, apply their choice, commit if needed, and resolve the thread.
gh pr view). Stop if no PR exists.reviewThreads field on the pull request — request each thread's id, isResolved, isOutdated, path, line, and comments. Filter to isResolved == false. (Use GraphQL, not REST: only it exposes thread ids and resolution state.)resolveReviewThread mutation. | Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 4,709 | 7,313 | +55% | 1 | 1 | 0% | 569 | 794 | +40% | 0 | 0 | — |
case-02 | fail→fail | 6,419 | 8,546 | +33% | 1 | 1 | 0% | 776 | 807 | +4% | 0 | 0 | — |
case-03 | fail→fail | 5,318 | 9,780 | +84% | 1 | 1 | 0% | 358 | 778 | +117% | 0 | 0 | — |
case-04 | pass→pass | 12,211 | 10,157 | -17% | 1 | 1 | 0% | 2,006 | 2,077 | +4% | 0 | 0 | — |
case-05 | pass→pass | 11,576 | 7,526 | -35% | 1 | 1 | 0% | 2,130 | 1,529 | -28% | 0 | 0 | — |
case-06 | pass→pass | 7,576 | 4,567 | -40% | 1 | 1 | 0% | 1,255 | 1,035 | -18% | 0 | 0 | — |
case-07 | fail→pass | 18,693 | 7,631 | -59% | 1 | 1 | 0% | 2,561 | 1,376 | -46% | 0 | 0 | — |
case-08 | pass→fail | 12,960 | 3,691 | -72% | 1 | 1 | 0% | 1,961 | 687 | -65% | 0 | 0 | — |
case-09 | pass→pass | 13,567 | 9,539 | -30% | 1 | 1 | 0% | 2,058 | 1,598 | -22% | 0 | 0 | — |
case-10 | pass→pass | 13,715 | 3,591 | -74% | 1 | 1 | 0% | 1,950 | 813 | -58% | 0 | 0 | — |
case-11 | fail→pass | 13,038 | 4,336 | -67% | 1 | 1 | 0% | 1,873 | 736 | -61% | 0 | 0 | — |
case-12 | pass→pass | 9,290 | 2,582 | -72% | 1 | 1 | 0% | 1,439 | 656 | -54% | 0 | 0 | — |
case-13 | pass→pass | 13,883 | 4,025 | -71% | 1 | 1 | 0% | 2,053 | 879 | -57% | 0 | 0 | — |
case-14 | pass→pass | 17,430 | 5,600 | -68% | 1 | 1 | 0% | 2,912 | 1,183 | -59% | 0 | 0 | — |
case-15 | fail→pass | 6,861 | 3,766 | -45% | 1 | 1 | 0% | 821 | 859 | +5% | 0 | 0 | — |
case-16 | fail→pass | 10,558 | 3,682 | -65% | 1 | 1 | 0% | 1,703 | 772 | -55% | 0 | 0 | — |
case-17 | pass→pass | 10,734 | 3,849 | -64% | 1 | 1 | 0% | 1,596 | 693 | -57% | 0 | 0 | — |
case-18 | pass→pass | 12,448 | 3,992 | -68% | 1 | 1 | 0% | 1,750 | 951 | -46% | 0 | 0 | — |
case-19 | pass→pass | 11,035 | 9,114 | -17% | 1 | 1 | 0% | 1,880 | 1,732 | -8% | 0 | 0 | — |
case-20 | pass→fail | 10,742 | 9,481 | -12% | 1 | 1 | 0% | 1,733 | 624 | -64% | 0 | 0 | — |
case-21 | pass→pass | 10,895 | 12,549 | +15% | 1 | 1 | 0% | 1,974 | 1,677 | -15% | 0 | 0 | — |
case-22 | pass→pass | 8,980 | 6,372 | -29% | 1 | 1 | 0% | 1,427 | 1,389 | -3% | 0 | 0 | — |
case-23 | pass→pass | 9,107 | 7,136 | -22% | 1 | 1 | 0% | 1,476 | 1,502 | +2% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 19 counted toward the lift figure. The other 4 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +9 percentage points is the difference between those two pass rates over the 19 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.