Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Selectively stages and commits only the changes related to the current session, skipping unrelated modifications.
.claude/skills/dcouple-commit/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-13 | ✗→✓ | ▲ Improved | 2% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -38% | 0% |
| case-06 | ✓→✗ | ▼ Worse | -17% | 0% |
| case-08 | ✓→✗ | ▼ Worse | 112% | 0% |
| case-09 | ✓→✗ | ▼ Worse | -54% | 0% |
Commit only the changes made in the current session to the local branch. Ignore all other changes. Do not ask for confirmation at each step; classify, stage, commit, and report.
./tmp/done-plans/ and ./tmp/ready-plans/$ARGUMENTS is provided and is not already a type: description commitmessage, use it as extra classification context
git statusgit diff and git diff --cachedInclude:
./tmp/done-plans/ files and related context artifactsExclude:
./tmp/ filesWhen in doubt, include the file rather than leaving it out.
If zero files are clearly in scope, stop and say so.
git add . or git add -A$ARGUMENTS already matches type: description, use it verbatimReport:
/prepare-pr if appropriate| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 3,234 | 4,414 | +36% | 1 | 1 | 0% | 371 | 644 | +74% | 0 | 0 | — |
case-02 | fail→fail | 4,068 | 4,538 | +12% | 1 | 1 | 0% | 144 | 613 | +326% | 0 | 0 | — |
case-03 | fail→fail | 2,783 | 4,902 | +76% | 1 | 1 | 0% | 286 | 655 | +129% | 0 | 0 | — |
case-04 | fail→fail | 7,738 | 5,241 | -32% | 1 | 1 | 0% | 1,201 | 792 | -34% | 0 | 0 | — |
case-05 | pass→pass | 12,863 | 4,548 | -65% | 1 | 1 | 0% | 2,142 | 1,235 | -42% | 0 | 0 | — |
case-06 | pass→fail | 4,784 | 5,978 | +25% | 1 | 1 | 0% | 794 | 660 | -17% | 0 | 0 | — |
case-07 | fail→fail | 7,865 | 5,407 | -31% | 1 | 1 | 0% | 1,354 | 761 | -44% | 0 | 0 | — |
case-08 | pass→fail | 2,095 | 4,951 | +136% | 1 | 1 | 0% | 345 | 733 | +112% | 0 | 0 | — |
case-09 | pass→fail | 10,993 | 5,660 | -49% | 1 | 1 | 0% | 1,698 | 777 | -54% | 0 | 0 | — |
case-10 | pass→fail | 7,138 | 3,064 | -57% | 1 | 1 | 0% | 1,181 | 790 | -33% | 0 | 0 | — |
case-11 | pass→fail | 8,580 | 4,669 | -46% | 1 | 1 | 0% | 1,311 | 717 | -45% | 0 | 0 | — |
case-12 | fail→fail | 8,446 | 4,136 | -51% | 1 | 1 | 0% | 1,407 | 744 | -47% | 0 | 0 | — |
case-13 | fail→pass | 6,931 | 3,875 | -44% | 1 | 1 | 0% | 1,089 | 1,107 | +2% | 0 | 0 | — |
case-14 | fail→pass | 6,342 | 1,431 | -77% | 1 | 1 | 0% | 1,046 | 651 | -38% | 0 | 0 | — |
case-15 | pass→pass | 8,446 | 1,750 | -79% | 1 | 1 | 0% | 1,260 | 704 | -44% | 0 | 0 | — |
case-16 | fail→fail | 8,035 | 4,894 | -39% | 1 | 1 | 0% | 1,270 | 808 | -36% | 0 | 0 | — |
case-17 | pass→pass | 6,443 | 2,460 | -62% | 1 | 1 | 0% | 1,043 | 874 | -16% | 0 | 0 | — |
case-18 | pass→pass | 4,455 | 1,944 | -56% | 1 | 1 | 0% | 734 | 802 | +9% | 0 | 0 | — |
case-19 | pass→fail | 7,576 | 3,440 | -55% | 1 | 1 | 0% | 1,350 | 890 | -34% | 0 | 0 | — |
case-20 | pass→fail | 2,226 | 3,718 | +67% | 1 | 1 | 0% | 368 | 766 | +108% | 0 | 0 | — |
case-21 | pass→fail | 3,609 | 6,672 | +85% | 1 | 1 | 0% | 627 | 679 | +8% | 0 | 0 | — |
case-22 | pass→fail | 2,841 | 5,016 | +77% | 1 | 1 | 0% | 480 | 691 | +44% | 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 8 counted toward the lift figure. The other 14 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 -32 percentage points is the difference between those two pass rates over the 8 comparable cases. 11 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.