Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Writes Git commit messages. Activates when the user asks to write a commit message, draft a commit message, or similar.
.claude/skills/herdrdev-writing-commit-messages/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -2% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -28% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -16% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -50% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -39% | 0% |
Write commit messages that follow commit style guidelines for the project.
<subsystem>: <summary>
<reference issues/PRs/etc.>
<long form description>area of code changed (e.g., terminal, vt, lib, config, font). Determine this from the file paths in the diff. If changes span the macOS app, use macos. For GTK, use gtk. For build system, use build. Use nested subsystems with / when helpful and exclusive (e.g., terminal/osc).
no trailing period. Keep it concise—ideally under 60 characters total for the whole subject line.
the relevant numbers on their own lines after the subject, separated by a blank line. E.g. #1234
line).
and how the new behavior works at a high level.
.jj is present, use jj instead of git for all commands.branch name.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 9,934 | 17,423 | +75% | 1 | 1 | 0% | 507 | 768 | +51% | 0 | 0 | — |
case-02 | fail→fail | 8,699 | 16,415 | +89% | 1 | 1 | 0% | 218 | 710 | +226% | 0 | 0 | — |
case-03 | fail→fail | 3,536 | 16,419 | +364% | 1 | 1 | 0% | 436 | 632 | +45% | 0 | 0 | — |
case-04 | pass→pass | 13,895 | 18,042 | +30% | 1 | 1 | 0% | 2,269 | 2,084 | -8% | 0 | 0 | — |
case-05 | pass→pass | 15,632 | 15,010 | -4% | 1 | 1 | 0% | 1,696 | 1,829 | +8% | 0 | 0 | — |
case-06 | pass→fail | 18,033 | 11,377 | -37% | 1 | 1 | 0% | 1,654 | 728 | -56% | 0 | 0 | — |
case-07 | fail→pass | 13,694 | 5,187 | -62% | 1 | 1 | 0% | 1,155 | 1,133 | -2% | 0 | 0 | — |
case-08 | pass→pass | 10,231 | 8,490 | -17% | 1 | 1 | 0% | 753 | 972 | +29% | 0 | 0 | — |
case-09 | pass→fail | 11,150 | 7,897 | -29% | 1 | 1 | 0% | 858 | 740 | -14% | 0 | 0 | — |
case-10 | fail→pass | 19,815 | 9,377 | -53% | 1 | 1 | 0% | 1,339 | 960 | -28% | 0 | 0 | — |
case-11 | pass→pass | 11,108 | 8,591 | -23% | 1 | 1 | 0% | 694 | 842 | +21% | 0 | 0 | — |
case-12 | fail→pass | 18,131 | 11,782 | -35% | 1 | 1 | 0% | 1,586 | 1,327 | -16% | 0 | 0 | — |
case-13 | pass→pass | 14,589 | 5,180 | -64% | 1 | 1 | 0% | 1,424 | 1,101 | -23% | 0 | 0 | — |
case-14 | pass→pass | 8,712 | 8,081 | -7% | 1 | 1 | 0% | 1,255 | 830 | -34% | 0 | 0 | — |
case-15 | fail→pass | 14,403 | 2,463 | -83% | 1 | 1 | 0% | 1,366 | 685 | -50% | 0 | 0 | — |
case-16 | pass→pass | 8,601 | 4,636 | -46% | 1 | 1 | 0% | 1,364 | 1,171 | -14% | 0 | 0 | — |
case-17 | fail→pass | 16,839 | 3,559 | -79% | 1 | 1 | 0% | 1,478 | 908 | -39% | 0 | 0 | — |
case-18 | pass→pass | 13,186 | 7,483 | -43% | 1 | 1 | 0% | 1,023 | 675 | -34% | 0 | 0 | — |
case-19 | pass→pass | 20,637 | 11,079 | -46% | 1 | 1 | 0% | 2,071 | 1,232 | -41% | 0 | 0 | — |
case-20 | fail→fail | 13,558 | 3,580 | -74% | 1 | 1 | 0% | 1,192 | 988 | -17% | 0 | 0 | — |
case-21 | fail→pass | 13,005 | 12,141 | -7% | 1 | 1 | 0% | 1,082 | 861 | -20% | 0 | 0 | — |
case-22 | pass→pass | 18,259 | 9,423 | -48% | 1 | 1 | 0% | 1,530 | 824 | -46% | 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 +18 percentage points is the difference between those two pass rates over the 18 comparable cases. 2 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/10/2026 | +14% |
Other measured skills in the registry, with their headline benchmark lift.