Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when a tutorial module plan exists but each module still needs a verifiable teaching loop. **Trigger**: exercises, practice, verification checklist, 教程练习, 可验证作业. **Use when**: tutorial 的 C2，已有 `outline/module_plan.yml`，需要为每个模块补齐 exercise / expected output / verification steps。 **Skip if**: 还没有 module plan。 **Network**: none. **Guardrail**: 练习必须可验证，不能只给开放式思考题。
.claude/skills/willoscar-exercise-builder/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 1% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -74% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -66% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -68% | 0% |
outline/module_plan.yml，需要为每个模块补齐 exercise / expected output / verification steps。Adds the minimal deterministic exercise contract to each planned tutorial module.
outline/module_plan.ymloutline/module_plan.ymlEach module must end up with at least one exercise containing:
promptexpected_outputverification_stepsscripts/run.py should:
Keep exercise phrasing logic in shared tutorial tooling rather than hardcoding it repeatedly in the skill script.
exercisesexpected_outputverification_steps| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 2,556 | 3,561 | +39% | 1 | 1 | 0% | 266 | 372 | +40% | 0 | 0 | — |
case-02 | fail→fail | 2,939 | 4,076 | +39% | 1 | 1 | 0% | 324 | 355 | +10% | 0 | 0 | — |
case-03 | fail→fail | 5,012 | 4,253 | -15% | 1 | 1 | 0% | 218 | 343 | +57% | 0 | 0 | — |
case-04 | fail→pass | 10,849 | 7,903 | -27% | 1 | 1 | 0% | 1,854 | 1,043 | -44% | 0 | 0 | — |
case-05 | fail→pass | 18,715 | 12,774 | -32% | 1 | 1 | 0% | 2,481 | 2,500 | +1% | 0 | 0 | — |
case-06 | fail→fail | 29,171 | 4,693 | -84% | 1 | 1 | 0% | 4,322 | 452 | -90% | 0 | 0 | — |
case-07 | fail→pass | 10,782 | 1,803 | -83% | 1 | 1 | 0% | 1,758 | 449 | -74% | 0 | 0 | — |
case-08 | pass→pass | 8,407 | 1,877 | -78% | 1 | 1 | 0% | 1,236 | 473 | -62% | 0 | 0 | — |
case-09 | fail→pass | 11,086 | 2,621 | -76% | 1 | 1 | 0% | 1,661 | 558 | -66% | 0 | 0 | — |
case-10 | fail→pass | 9,058 | 2,051 | -77% | 1 | 1 | 0% | 1,585 | 514 | -68% | 0 | 0 | — |
case-11 | fail→pass | 8,920 | 1,296 | -85% | 1 | 1 | 0% | 1,169 | 341 | -71% | 0 | 0 | — |
case-12 | fail→fail | 9,908 | 1,747 | -82% | 1 | 1 | 0% | 1,392 | 425 | -69% | 0 | 0 | — |
case-13 | pass→pass | 12,298 | 4,120 | -66% | 1 | 1 | 0% | 2,033 | 814 | -60% | 0 | 0 | — |
case-14 | fail→pass | 11,054 | 1,822 | -84% | 1 | 1 | 0% | 1,583 | 467 | -70% | 0 | 0 | — |
case-15 | fail→pass | 6,559 | 2,052 | -69% | 1 | 1 | 0% | 1,049 | 447 | -57% | 0 | 0 | — |
case-16 | pass→pass | 12,067 | 7,672 | -36% | 1 | 1 | 0% | 1,954 | 1,469 | -25% | 0 | 0 | — |
case-17 | fail→fail | 10,814 | 2,087 | -81% | 1 | 1 | 0% | 1,718 | 493 | -71% | 0 | 0 | — |
case-18 | fail→pass | 6,620 | 1,593 | -76% | 1 | 1 | 0% | 980 | 408 | -58% | 0 | 0 | — |
case-19 | pass→pass | 4,326 | 2,093 | -52% | 1 | 1 | 0% | 648 | 487 | -25% | 0 | 0 | — |
case-20 | fail→pass | 12,186 | 7,116 | -42% | 1 | 1 | 0% | 1,821 | 1,221 | -33% | 0 | 0 | — |
case-21 | fail→pass | 9,140 | 1,681 | -82% | 1 | 1 | 0% | 1,427 | 450 | -68% | 0 | 0 | — |
case-22 | fail→pass | 12,871 | 2,178 | -83% | 1 | 1 | 0% | 2,128 | 487 | -77% | 0 | 0 | — |
case-23 | fail→pass | 10,626 | 2,120 | -80% | 1 | 1 | 0% | 1,712 | 529 | -69% | 0 | 0 | — |
case-24 | pass→pass | 8,613 | 1,931 | -78% | 1 | 1 | 0% | 1,314 | 492 | -63% | 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. 24 cases were attempted, and 20 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 +54 percentage points is the difference between those two pass rates over the 20 comparable cases.
The publisher has shipped newer versions since this run, so these numbers describe v1, not the version currently listed.
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.