Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Run Karpathy's 4-principle review on staged changes or the last commit. Checks complexity, diff noise, hidden assumptions, and goal verification. Usage /karpathy-check [--last-commit]
.claude/skills/alirezarezvani-karpathy-check/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -14% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -65% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -66% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -58% | 0% |
| case-20 | ✗→✓ | ▲ Improved | -57% | 0% |
<!-- canonical copy: engineering/karpathy-coder/commands/karpathy-check.md — keep in sync (root copy uses repo-root-relative script paths) -->
Review your staged changes (or last commit) against Karpathy's 4 coding principles.
/karpathy-check # review staged changes
/karpathy-check --last-commit # review the most recent commitengineering/karpathy-coder/skills/karpathy-coder/scripts/complexity_checker.py on all changed files — detects over-engineering, premature abstractions, deep nesting, long functionsengineering/karpathy-coder/skills/karpathy-coder/scripts/diff_surgeon.py on the diff — detects comment-only changes, whitespace noise, style drift, drive-by refactorskarpathy-reviewer agent reads the diff and applies human-judgment checks — hidden assumptions, missing verificationA structured report with per-principle verdicts and specific line-level fix recommendations.
Dispatches the karpathy-reviewer agent. See agents/karpathy-reviewer.md.
engineering/karpathy-coder/skills/karpathy-coder/scripts/complexity_checker.pyengineering/karpathy-coder/skills/karpathy-coder/scripts/diff_surgeon.pyengineering/karpathy-coder/skills/karpathy-coder/scripts/assumption_linter.pyengineering/karpathy-coder/skills/karpathy-coder/scripts/goal_verifier.py→ engineering/karpathy-coder/skills/karpathy-coder/SKILL.md
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 7,457 | 2,602 | -65% | 1 | 1 | 0% | 1,219 | 850 | -30% | 0 | 0 | — |
case-02 | fail→fail | 3,828 | 5,250 | +37% | 1 | 1 | 0% | 585 | 738 | +26% | 0 | 0 | — |
case-03 | fail→fail | 5,454 | 3,409 | -37% | 1 | 1 | 0% | 778 | 970 | +25% | 0 | 0 | — |
case-04 | pass→pass | 13,498 | 10,217 | -24% | 1 | 1 | 0% | 2,667 | 2,396 | -10% | 0 | 0 | — |
case-05 | pass→pass | 9,185 | 7,796 | -15% | 1 | 1 | 0% | 1,674 | 2,032 | +21% | 0 | 0 | — |
case-06 | pass→pass | 15,284 | 13,845 | -9% | 1 | 1 | 0% | 3,146 | 3,430 | +9% | 0 | 0 | — |
case-07 | fail→pass | 6,516 | 2,429 | -63% | 1 | 1 | 0% | 1,041 | 897 | -14% | 0 | 0 | — |
case-08 | fail→pass | 13,011 | 1,599 | -88% | 1 | 1 | 0% | 2,027 | 714 | -65% | 0 | 0 | — |
case-09 | fail→pass | 12,473 | 1,756 | -86% | 1 | 1 | 0% | 2,311 | 794 | -66% | 0 | 0 | — |
case-10 | fail→pass | 9,818 | 1,503 | -85% | 1 | 1 | 0% | 1,785 | 742 | -58% | 0 | 0 | — |
case-20 | fail→pass | 11,933 | 2,049 | -83% | 1 | 1 | 0% | 1,876 | 810 | -57% | 0 | 0 | — |
case-11 | fail→pass | 10,766 | 1,447 | -87% | 1 | 1 | 0% | 1,702 | 714 | -58% | 0 | 0 | — |
case-12 | pass→pass | 6,450 | 2,078 | -68% | 1 | 1 | 0% | 1,062 | 778 | -27% | 0 | 0 | — |
case-13 | fail→pass | 10,134 | 1,593 | -84% | 1 | 1 | 0% | 1,686 | 743 | -56% | 0 | 0 | — |
case-14 | fail→pass | 9,439 | 2,313 | -75% | 1 | 1 | 0% | 1,586 | 892 | -44% | 0 | 0 | — |
case-15 | fail→pass | 7,268 | 1,511 | -79% | 1 | 1 | 0% | 1,199 | 590 | -51% | 0 | 0 | — |
case-16 | fail→pass | 11,117 | 2,747 | -75% | 1 | 1 | 0% | 1,739 | 964 | -45% | 0 | 0 | — |
case-17 | fail→pass | 14,924 | 1,565 | -90% | 1 | 1 | 0% | 2,693 | 710 | -74% | 0 | 0 | — |
case-18 | fail→pass | 17,086 | 2,022 | -88% | 1 | 1 | 0% | 3,211 | 750 | -77% | 0 | 0 | — |
case-19 | fail→pass | 14,910 | 2,233 | -85% | 1 | 1 | 0% | 2,324 | 767 | -67% | 0 | 0 | — |
case-21 | fail→pass | 11,222 | 2,155 | -81% | 1 | 1 | 0% | 1,845 | 858 | -53% | 0 | 0 | — |
case-22 | pass→pass | 11,045 | 3,717 | -66% | 1 | 1 | 0% | 1,795 | 1,092 | -39% | 0 | 0 | — |
case-23 | fail→pass | 9,289 | 1,748 | -81% | 1 | 1 | 0% | 1,565 | 713 | -54% | 0 | 0 | — |
case-24 | pass→pass | 5,688 | 2,318 | -59% | 1 | 1 | 0% | 1,044 | 883 | -15% | 0 | 0 | — |
case-25 | fail→pass | 16,356 | 11,888 | -27% | 1 | 1 | 0% | 2,658 | 2,386 | -10% | 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. 25 cases were attempted. The headline lift of +64 percentage points is the difference between those two pass rates over the 25 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.
Other measured skills in the registry, with their headline benchmark lift.