Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Version control discipline for agent work. Atomic commits, meaningful messages, safe operations, and zero destructive surprises. Use this whenever committing, branching, merging, rebasing, resolving conflicts, or preparing pull requests, before any git command that rewrites or discards history, and when the user says "commit this", "clean up the history", or "undo". Consult it before every force push, reset, or clean, without exception.
.claude/skills/adityaarakeri-git-hygiene/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | 273% | 0% |
| case-07 | ✓→✗ | ▼ Worse | 24% | 0% |
| case-17 | ✓→✗ | ▼ Worse | -13% | 0% |
| case-18 | ✓→✗ | ▼ Worse | 16% | 0% |
| case-03 | ✓→✓ | ＝ Same ✓ | 15% | 0% |
Treat version-control state as user data. Mutate only the parts required by the requested git workflow.
--force-with-lease over an authorized force push and revert over resetting shared history.reflog, a backup branch, stash, or patch before destructive recovery, but do not create or apply them without authority.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 2,788 | 10,541 | +278% | 1 | 1 | 0% | 221 | 672 | +204% | 0 | 0 | — |
case-02 | fail→fail | 8,099 | 5,598 | -31% | 1 | 1 | 0% | 1,539 | 839 | -45% | 0 | 0 | — |
case-03 | pass→pass | 5,251 | 4,746 | -10% | 1 | 1 | 0% | 1,031 | 1,190 | +15% | 0 | 0 | — |
case-04 | pass→pass | 4,433 | 2,982 | -33% | 1 | 1 | 0% | 841 | 917 | +9% | 0 | 0 | — |
case-05 | pass→pass | 1,722 | 9,133 | +430% | 1 | 1 | 0% | 260 | 615 | +137% | 0 | 0 | — |
case-06 | pass→pass | 4,948 | 6,747 | +36% | 1 | 1 | 0% | 953 | 1,661 | +74% | 0 | 0 | — |
case-07 | pass→fail | 3,499 | 3,108 | -11% | 1 | 1 | 0% | 674 | 833 | +24% | 0 | 0 | — |
case-08 | fail→fail | 4,682 | 3,490 | -25% | 1 | 1 | 0% | 731 | 884 | +21% | 0 | 0 | — |
case-09 | fail→pass | 8,229 | 8,164 | -1% | 1 | 1 | 0% | 533 | 1,987 | +273% | 0 | 0 | — |
case-10 | pass→pass | 4,052 | 7,307 | +80% | 1 | 1 | 0% | 769 | 1,869 | +143% | 0 | 0 | — |
case-11 | fail→fail | 6,430 | 11,773 | +83% | 1 | 1 | 0% | 1,088 | 1,762 | +62% | 0 | 0 | — |
case-12 | pass→pass | 4,341 | 3,585 | -17% | 1 | 1 | 0% | 822 | 1,038 | +26% | 0 | 0 | — |
case-13 | pass→pass | 3,548 | 8,485 | +139% | 1 | 1 | 0% | 572 | 787 | +38% | 0 | 0 | — |
case-14 | pass→pass | 5,500 | 5,058 | -8% | 1 | 1 | 0% | 985 | 1,193 | +21% | 0 | 0 | — |
case-15 | pass→pass | 8,650 | 6,711 | -22% | 1 | 1 | 0% | 1,604 | 1,635 | +2% | 0 | 0 | — |
case-16 | fail→fail | 5,303 | 7,428 | +40% | 1 | 1 | 0% | 1,091 | 1,657 | +52% | 0 | 0 | — |
case-17 | pass→fail | 3,752 | 3,966 | +6% | 1 | 1 | 0% | 729 | 634 | -13% | 0 | 0 | — |
case-18 | pass→fail | 3,409 | 3,875 | +14% | 1 | 1 | 0% | 533 | 620 | +16% | 0 | 0 | — |
case-19 | pass→pass | 9,609 | 7,779 | -19% | 1 | 1 | 0% | 1,690 | 1,728 | +2% | 0 | 0 | — |
case-20 | pass→pass | 5,405 | 5,353 | -1% | 1 | 1 | 0% | 925 | 1,323 | +43% | 0 | 0 | — |
case-21 | pass→pass | 2,849 | 2,991 | +5% | 1 | 1 | 0% | 548 | 894 | +63% | 0 | 0 | — |
case-22 | fail→fail | 5,117 | 5,364 | +5% | 1 | 1 | 0% | 371 | 607 | +64% | 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, and 18 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 -100 percentage points is the difference between those two pass rates over the 18 comparable cases. 3 cases got worse with the skill loaded, and they are included in that figure.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/3/2026 | +33% |
| gemini-3.6-flash | verified | 8/3/2026 | +18% |
Other measured skills in the registry, with their headline benchmark lift.