Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Commit staged work, push the branch, and open a PR. Use when wrapping up changes that need to ship.
.claude/skills/itsbencodes-commit/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | 64% | 0% |
| case-09 | ✓→✗ | ▼ Worse | -62% | 0% |
| case-15 | ✓→✗ | ▼ Worse | -53% | 0% |
| case-16 | ✓→✓ | ＝ Same ✓ | -22% | 0% |
| case-17 | ✓→✓ | ＝ Same ✓ | 22% | 0% |
Commit → push → open PR. Never on main.
Stage only files relevant to this change. Prefer explicit paths over git add -A unless every modified file belongs to the change.
git log). Concise, focused on the why.--no-verify. If a hook fails, fix the cause and make a new commit.bashgit commit -m "<message>"
bashgit push -u origin HEAD
Title under 70 chars. Body via HEREDOC. Default base is main.
bashgh pr create --title "<title>" --body "$(cat <<'EOF' ## Summary - ... ## Test plan - [ ] ... EOF )"
Return the PR URL. Use gh pr merge --auto --squash only when the caller asks for auto-merge (e.g. release PRs).
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 7,226 | 11,212 | +55% | 1 | 1 | 0% | 317 | 489 | +54% | 0 | 0 | — |
case-02 | fail→fail | 7,267 | 6,298 | -13% | 1 | 1 | 0% | 1,024 | 579 | -43% | 0 | 0 | — |
case-03 | fail→fail | 11,770 | 5,467 | -54% | 1 | 1 | 0% | 462 | 460 | -0% | 0 | 0 | — |
case-04 | fail→fail | 10,771 | 5,707 | -47% | 1 | 1 | 0% | 1,405 | 466 | -67% | 0 | 0 | — |
case-05 | fail→fail | 9,943 | 8,446 | -15% | 1 | 1 | 0% | 1,313 | 564 | -57% | 0 | 0 | — |
case-06 | fail→pass | 7,432 | 12,756 | +72% | 1 | 1 | 0% | 1,273 | 2,084 | +64% | 0 | 0 | — |
case-07 | fail→fail | 11,651 | 5,319 | -54% | 1 | 1 | 0% | 1,952 | 462 | -76% | 0 | 0 | — |
case-20 | fail→fail | 5,624 | 5,509 | -2% | 1 | 1 | 0% | 356 | 584 | +64% | 0 | 0 | — |
case-08 | fail→fail | 8,683 | 11,866 | +37% | 1 | 1 | 0% | 1,223 | 1,364 | +12% | 0 | 0 | — |
case-09 | pass→fail | 9,504 | 4,938 | -48% | 1 | 1 | 0% | 1,240 | 470 | -62% | 0 | 0 | — |
case-10 | fail→fail | 9,956 | 5,269 | -47% | 1 | 1 | 0% | 819 | 553 | -32% | 0 | 0 | — |
case-11 | fail→fail | 3,403 | 4,897 | +44% | 1 | 1 | 0% | 309 | 517 | +67% | 0 | 0 | — |
case-21 | fail→fail | 6,336 | 8,219 | +30% | 1 | 1 | 0% | 977 | 765 | -22% | 0 | 0 | — |
case-12 | fail→fail | 4,929 | 4,396 | -11% | 1 | 1 | 0% | 721 | 396 | -45% | 0 | 0 | — |
case-13 | fail→fail | 5,532 | 5,360 | -3% | 1 | 1 | 0% | 441 | 555 | +26% | 0 | 0 | — |
case-14 | fail→fail | 5,900 | 4,790 | -19% | 1 | 1 | 0% | 175 | 513 | +193% | 0 | 0 | — |
case-15 | pass→fail | 12,181 | 7,792 | -36% | 1 | 1 | 0% | 1,862 | 867 | -53% | 0 | 0 | — |
case-22 | fail→fail | 5,405 | 6,370 | +18% | 1 | 1 | 0% | 319 | 411 | +29% | 0 | 0 | — |
case-16 | pass→pass | 6,951 | 7,522 | +8% | 1 | 1 | 0% | 1,268 | 994 | -22% | 0 | 0 | — |
case-17 | pass→pass | 3,700 | 3,677 | -1% | 1 | 1 | 0% | 617 | 754 | +22% | 0 | 0 | — |
case-18 | pass→pass | 5,646 | 6,628 | +17% | 1 | 1 | 0% | 1,038 | 1,184 | +14% | 0 | 0 | — |
case-19 | fail→fail | 3,225 | 5,425 | +68% | 1 | 1 | 0% | 257 | 474 | +84% | 0 | 0 | — |
case-23 | fail→fail | 13,763 | 6,121 | -56% | 1 | 1 | 0% | 1,792 | 621 | -65% | 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 4 counted toward the lift figure. The other 19 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 -4 percentage points is the difference between those two pass rates over the 4 comparable cases. 7 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.