Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →超圧縮PRレビューコメント。1行1指摘: 位置・問題・修正。前置き削除、シグナル優先。 日本語対応。「PRレビューして」「コードレビュー」「/review」「/genshijin-review」で起動。 プルリクエストレビュー時に自動起動候補。
.claude/skills/interfacex-co-jp-genshijin-review/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | -24% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -20% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 136% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -17% | 0% |
レビューコメントは簡潔かつ行動可能に。1行1指摘。位置・問題・修正。前置き禁止。
形式: L<line>: <問題>。<修正>。 — 複数ファイル時 <file>:L<line>: ...
重大度プレフィックス（混在時）:
バグ: — 壊れている。インシデント直結リスク: — 動くが脆い（race, null未チェック, 握り潰しerror）nit: — スタイル・命名・ミクロ最適化。著者無視可質問: — 純粋な疑問。提案ではない削除:
nit: 使う質問:保持:
❌ 「L42 で user オブジェクトが null かどうかをチェックせずに email プロパティにアクセスしているように見えます。DBで user が見つからなかった場合にクラッシュする可能性があります。null チェックを追加することを検討してみてください。」
✅ L42: 🔴 バグ: .find() 後 user null 可。.email 前にガード追加。
❌ 「この関数はいろいろやっていて、小さな関数に分割すると読みやすくなるかもしれません。」
✅ L88-140: 🔵 nit: 50行fn 4責務。validate/normalize/persist 抽出。
❌ 「APIが 429 を返した場合の処理は考慮されていますか？対応したほうがよいと思います。」
✅ L23: 🟡 リスク: 429 リトライなし。withBackoff(3) で包む。
以下は簡潔モード解除・通常の段落で記述:
該当指摘後 即復帰。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | fail→pass | 7,639 | 2,307 | -70% | 1 | 1 | 0% | 1,323 | 1,004 | -24% | 0 | 0 | — |
case-01 | fail→fail | 5,875 | 17,325 | +195% | 1 | 1 | 0% | 963 | 855 | -11% | 0 | 0 | — |
case-02 | fail→pass | 7,210 | 1,976 | -73% | 1 | 1 | 0% | 1,235 | 983 | -20% | 0 | 0 | — |
case-03 | fail→pass | 2,696 | 1,890 | -30% | 1 | 1 | 0% | 393 | 926 | +136% | 0 | 0 | — |
case-04 | fail→pass | 7,298 | 2,112 | -71% | 1 | 1 | 0% | 1,183 | 973 | -18% | 0 | 0 | — |
case-06 | fail→fail | 11,859 | 5,189 | -56% | 1 | 1 | 0% | 1,878 | 1,542 | -18% | 0 | 0 | — |
case-07 | pass→pass | 2,706 | 3,065 | +13% | 1 | 1 | 0% | 370 | 1,130 | +205% | 0 | 0 | — |
case-08 | pass→pass | 11,292 | 2,147 | -81% | 1 | 1 | 0% | 1,868 | 997 | -47% | 0 | 0 | — |
case-09 | pass→pass | 26,033 | 4,288 | -84% | 1 | 1 | 0% | 2,685 | 1,417 | -47% | 0 | 0 | — |
case-10 | fail→pass | 15,554 | 8,889 | -43% | 1 | 1 | 0% | 2,394 | 1,991 | -17% | 0 | 0 | — |
case-11 | pass→pass | 12,218 | 4,387 | -64% | 1 | 1 | 0% | 2,042 | 1,331 | -35% | 0 | 0 | — |
case-12 | pass→fail | 3,551 | 5,350 | +51% | 1 | 1 | 0% | 609 | 915 | +50% | 0 | 0 | — |
case-13 | fail→pass | 9,184 | 3,627 | -61% | 1 | 1 | 0% | 1,466 | 1,270 | -13% | 0 | 0 | — |
case-14 | pass→pass | 11,608 | 2,614 | -77% | 1 | 1 | 0% | 2,121 | 1,043 | -51% | 0 | 0 | — |
case-15 | pass→pass | 8,526 | 4,358 | -49% | 1 | 1 | 0% | 1,535 | 1,335 | -13% | 0 | 0 | — |
case-16 | fail→pass | 7,435 | 2,775 | -63% | 1 | 1 | 0% | 1,427 | 1,092 | -23% | 0 | 0 | — |
case-17 | pass→pass | 6,323 | 1,737 | -73% | 1 | 1 | 0% | 1,218 | 919 | -25% | 0 | 0 | — |
case-18 | fail→pass | 5,689 | 2,105 | -63% | 1 | 1 | 0% | 998 | 993 | -1% | 0 | 0 | — |
case-19 | fail→pass | 9,187 | 2,499 | -73% | 1 | 1 | 0% | 1,589 | 1,055 | -34% | 0 | 0 | — |
case-20 | pass→fail | 6,140 | 4,043 | -34% | 1 | 1 | 0% | 1,230 | 1,384 | +13% | 0 | 0 | — |
case-21 | pass→fail | 4,588 | 4,774 | +4% | 1 | 1 | 0% | 858 | 1,291 | +50% | 0 | 0 | — |
case-22 | pass→pass | 7,979 | 2,354 | -70% | 1 | 1 | 0% | 1,449 | 1,029 | -29% | 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. 22 cases were attempted. The headline lift of +27 percentage points is the difference between those two pass rates over the 22 comparable cases. 3 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.