Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when a `source-tutorial` workspace has ingested sources and needs a grounded tutorial contract before structure planning. **Trigger**: source tutorial spec, tutorial from sources, learner profile, 教程规格, 从资料生成教程. **Use when**: `source-tutorial` 的 C2，需要根据 `sources/index.jsonl` / `sources/provenance.jsonl` 锁定 audience、prerequisites、learning objectives、source scope 和 running example policy。
.claude/skills/willoscar-source-tutorial-spec/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -30% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -36% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -41% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -57% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -24% | 0% |
Builds the C2 tutorial contract from the ingested source corpus, not from a bare topic prompt.
sources/index.jsonlsources/provenance.jsonlGOAL.mdDECISIONS.mdoutput/TUTORIAL_SPEC.mdThe spec must include:
## Audience## Prerequisites## Learning objectives## Non-goals## Source scope## Running example policy## Delivery shapeIt should also embed machine-readable structured data for downstream deterministic planning.
scripts/run.py should:
Keep phrase extraction, concept selection, and source matching in shared tutorial tooling, not in the thin skill wrapper.
output/TUTORIAL_SPEC.md exists| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-10 | fail→fail | 11,354 | 1,569 | -86% | 1 | 1 | 0% | 1,642 | 489 | -70% | 0 | 0 | — |
case-01 | fail→fail | 35,450 | 4,447 | -87% | 1 | 1 | 0% | 5,916 | 467 | -92% | 0 | 0 | — |
case-02 | fail→fail | 18,190 | 4,641 | -74% | 1 | 1 | 0% | 3,006 | 439 | -85% | 0 | 0 | — |
case-03 | fail→fail | 36,514 | 4,737 | -87% | 1 | 1 | 0% | 6,155 | 443 | -93% | 0 | 0 | — |
case-04 | fail→pass | 19,195 | 2,581 | -87% | 1 | 1 | 0% | 992 | 691 | -30% | 0 | 0 | — |
case-05 | pass→pass | 13,039 | 3,868 | -70% | 1 | 1 | 0% | 1,835 | 865 | -53% | 0 | 0 | — |
case-06 | fail→pass | 16,716 | 8,137 | -51% | 1 | 1 | 0% | 2,562 | 1,645 | -36% | 0 | 0 | — |
case-07 | fail→pass | 8,102 | 2,661 | -67% | 1 | 1 | 0% | 1,101 | 645 | -41% | 0 | 0 | — |
case-08 | fail→fail | 12,332 | 2,402 | -81% | 1 | 1 | 0% | 893 | 612 | -31% | 0 | 0 | — |
case-09 | fail→fail | 12,956 | 1,629 | -87% | 1 | 1 | 0% | 1,987 | 471 | -76% | 0 | 0 | — |
case-11 | fail→pass | 7,379 | 1,656 | -78% | 1 | 1 | 0% | 1,106 | 481 | -57% | 0 | 0 | — |
case-12 | fail→pass | 4,981 | 1,837 | -63% | 1 | 1 | 0% | 642 | 490 | -24% | 0 | 0 | — |
case-13 | fail→fail | 5,507 | 1,545 | -72% | 1 | 1 | 0% | 809 | 448 | -45% | 0 | 0 | — |
case-14 | fail→fail | 9,049 | 2,872 | -68% | 1 | 1 | 0% | 1,331 | 714 | -46% | 0 | 0 | — |
case-15 | fail→pass | 4,927 | 2,696 | -45% | 1 | 1 | 0% | 814 | 646 | -21% | 0 | 0 | — |
case-16 | fail→pass | 6,916 | 1,957 | -72% | 1 | 1 | 0% | 1,024 | 528 | -48% | 0 | 0 | — |
case-17 | fail→pass | 6,031 | 1,722 | -71% | 1 | 1 | 0% | 794 | 527 | -34% | 0 | 0 | — |
case-18 | pass→pass | 13,582 | 2,807 | -79% | 1 | 1 | 0% | 2,051 | 720 | -65% | 0 | 0 | — |
case-19 | pass→pass | 18,655 | 10,605 | -43% | 1 | 1 | 0% | 2,890 | 1,921 | -34% | 0 | 0 | — |
case-20 | pass→fail | 15,216 | 4,849 | -68% | 1 | 1 | 0% | 2,485 | 517 | -79% | 0 | 0 | — |
case-21 | fail→pass | 8,380 | 18,324 | +119% | 1 | 1 | 0% | 1,326 | 3,519 | +165% | 0 | 0 | — |
case-22 | pass→pass | 26,200 | 21,991 | -16% | 1 | 1 | 0% | 3,734 | 4,063 | +9% | 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 16 counted toward the lift figure. The other 6 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 +36 percentage points is the difference between those two pass rates over the 16 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.
Other measured skills in the registry, with their headline benchmark lift.