Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Review implementation against provided markdown requirements, specs, PRDs, or phase docs; optionally apply focused fixes. Use when a user supplies .md file paths and asks to verify correctness/completeness, check PRD phase completion, find missing or overbuilt work, assess critical issues, refactor long files/functions, or update code to match the docs.
.claude/skills/golbin-review-implementation/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 75% | 17 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 45% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 1% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 10% | 0% |
| case-02 | ✓→✓ | = Same ✓ | 4% | 0% |
| case-03 | ✓→✓ | = Same ✓ | 16% | 0% |
Review from evidence. Protect the purpose first, then simplify repeatedly.
Follow the Unix and X11 philosophies:
Simplicity is not fewer lines at the expense of correctness. Keep everything required by the goal and nothing else.
Work in this order. Do not collapse all concerns into one pass.
Run several small passes:
After each pass, preserve intended behavior with proportionate checks. Then repeat the purpose review and refinement passes until a full pass finds no meaningful problem or simplification. Do not stop after the first acceptable result.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 10,796 | 10,404 | -4% | 1 | 1 | 0% | 1,924 | 2,793 | +45% | 0 | 0 | — |
case-02 | pass→pass | 13,219 | 7,984 | -40% | 1 | 1 | 0% | 2,062 | 2,144 | +4% | 0 | 0 | — |
case-03 | pass→pass | 12,932 | 9,616 | -26% | 1 | 1 | 0% | 2,267 | 2,641 | +16% | 0 | 0 | — |
case-04 | pass→pass | 13,757 | 13,870 | +1% | 1 | 1 | 0% | 2,463 | 3,173 | +29% | 0 | 0 | — |
case-05 | pass→pass | 11,896 | 7,685 | -35% | 1 | 1 | 0% | 1,922 | 2,133 | +11% | 0 | 0 | — |
case-06 | pass→pass | 9,385 | 4,404 | -53% | 1 | 1 | 0% | 1,560 | 1,611 | +3% | 0 | 0 | — |
case-07 | fail→pass | 11,808 | 7,075 | -40% | 1 | 1 | 0% | 2,072 | 2,092 | +1% | 0 | 0 | — |
case-08 | pass→pass | 11,773 | 5,412 | -54% | 1 | 1 | 0% | 2,004 | 1,675 | -16% | 0 | 0 | — |
case-09 | pass→pass | 10,414 | 7,902 | -24% | 1 | 1 | 0% | 1,703 | 2,170 | +27% | 0 | 0 | — |
case-10 | pass→pass | 12,747 | 9,083 | -29% | 1 | 1 | 0% | 1,961 | 2,279 | +16% | 0 | 0 | — |
case-11 | pass→pass | 9,352 | 5,326 | -43% | 1 | 1 | 0% | 1,557 | 1,807 | +16% | 0 | 0 | — |
case-12 | pass→pass | 6,372 | 3,734 | -41% | 1 | 1 | 0% | 1,106 | 1,516 | +37% | 0 | 0 | — |
case-13 | fail→pass | 9,458 | 3,702 | -61% | 1 | 1 | 0% | 1,477 | 1,619 | +10% | 0 | 0 | — |
case-14 | pass→pass | 11,446 | 10,296 | -10% | 1 | 1 | 0% | 1,999 | 2,579 | +29% | 0 | 0 | — |
case-15 | pass→pass | 16,744 | 9,591 | -43% | 1 | 1 | 0% | 2,404 | 2,450 | +2% | 0 | 0 | — |
case-16 | pass→pass | 6,074 | 4,147 | -32% | 1 | 1 | 0% | 996 | 1,558 | +56% | 0 | 0 | — |
case-17 | pass→pass | 18,331 | 6,615 | -64% | 1 | 1 | 0% | 2,130 | 2,019 | -5% | 0 | 0 | — |
case-18 | pass→pass | 9,364 | 5,288 | -44% | 1 | 1 | 0% | 1,512 | 1,753 | +16% | 0 | 0 | — |
case-19 | pass→pass | 15,042 | 4,609 | -69% | 1 | 1 | 0% | 1,760 | 1,671 | -5% | 0 | 0 | — |
case-20 | pass→pass | 15,861 | 12,711 | -20% | 1 | 1 | 0% | 2,659 | 2,998 | +13% | 0 | 0 | — |
case-21 | pass→pass | 18,718 | 82,615 | +341% | 1 | 1 | 0% | 2,500 | 2,956 | +18% | 0 | 0 | — |
case-22 | pass→pass | 7,618 | 32,700 | +329% | 1 | 1 | 0% | 1,282 | 1,322 | +3% | 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 +14 percentage points is the difference between those two pass rates over the 22 comparable cases.
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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/22/2026 | — |
Other measured skills in the registry, with their headline benchmark lift.