Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when approved tutorial context packs exist and the run needs the final article-first tutorial deliverable. **Trigger**: source tutorial writer, tutorial drafting, 教程正文, 从资料写教程. **Use when**: `source-tutorial` 的 C3，`outline/tutorial_context_packs.jsonl` 已就绪，且 `DECISIONS.md` 已勾选 `Approve C2`。 **Skip if**: C2 未批准，或 context packs 还没准备好。 **Network**: none. **Guardrail**: 正文必须 reader-first，但不能写出 sources 没支持的内容。
.claude/skills/willoscar-source-tutorial-writer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-12 | ✗→✓ | ▲ Improved | -58% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -50% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -58% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -71% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -26% | 0% |
Writes the final tutorial markdown from approved module packs and the locked C2 structure.
outline/tutorial_context_packs.jsonloutline/module_plan.ymlDECISIONS.mdoutput/TUTORIAL.mdThe tutorial must:
Per module, the writer should emit:
### Why it matters### Key idea### Worked example### Check yourself### Source notesscripts/run.py should:
Approve C2output/TUTORIAL.mdKeep rendering and pack interpretation in shared tutorial tooling, not in the wrapper.
output/TUTORIAL.md exists| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | pass→pass | 12,201 | 4,645 | -62% | 1 | 1 | 0% | 2,267 | 948 | -58% | 0 | 0 | — |
case-06 | fail→fail | 15,009 | 2,026 | -87% | 1 | 1 | 0% | 761 | 483 | -37% | 0 | 0 | — |
case-01 | pass→fail | 32,268 | 4,589 | -86% | 1 | 1 | 0% | 6,187 | 444 | -93% | 0 | 0 | — |
case-02 | pass→fail | 29,168 | 4,603 | -84% | 1 | 1 | 0% | 6,183 | 506 | -92% | 0 | 0 | — |
case-03 | pass→fail | 34,068 | 4,659 | -86% | 1 | 1 | 0% | 6,180 | 397 | -94% | 0 | 0 | — |
case-04 | pass→pass | 9,746 | 2,650 | -73% | 1 | 1 | 0% | 1,445 | 645 | -55% | 0 | 0 | — |
case-12 | fail→pass | 22,189 | 1,736 | -92% | 1 | 1 | 0% | 1,243 | 518 | -58% | 0 | 0 | — |
case-07 | fail→fail | 10,121 | 2,640 | -74% | 1 | 1 | 0% | 1,522 | 817 | -46% | 0 | 0 | — |
case-08 | fail→pass | 10,831 | 3,429 | -68% | 1 | 1 | 0% | 1,493 | 752 | -50% | 0 | 0 | — |
case-09 | pass→pass | 7,411 | 4,783 | -35% | 1 | 1 | 0% | 1,172 | 997 | -15% | 0 | 0 | — |
case-10 | pass→pass | 9,641 | 3,368 | -65% | 1 | 1 | 0% | 1,503 | 737 | -51% | 0 | 0 | — |
case-11 | fail→pass | 10,045 | 2,489 | -75% | 1 | 1 | 0% | 1,640 | 684 | -58% | 0 | 0 | — |
case-13 | pass→pass | 11,029 | 5,035 | -54% | 1 | 1 | 0% | 1,698 | 1,007 | -41% | 0 | 0 | — |
case-14 | pass→fail | 9,214 | 2,137 | -77% | 1 | 1 | 0% | 1,420 | 564 | -60% | 0 | 0 | — |
case-15 | fail→pass | 9,512 | 1,756 | -82% | 1 | 1 | 0% | 1,800 | 529 | -71% | 0 | 0 | — |
case-16 | fail→pass | 11,312 | 6,955 | -39% | 1 | 1 | 0% | 2,008 | 1,483 | -26% | 0 | 0 | — |
case-17 | pass→pass | 14,545 | 4,271 | -71% | 1 | 1 | 0% | 2,190 | 840 | -62% | 0 | 0 | — |
case-18 | fail→fail | 10,186 | 2,801 | -73% | 1 | 1 | 0% | 1,962 | 647 | -67% | 0 | 0 | — |
case-19 | pass→fail | 4,402 | 2,626 | -40% | 1 | 1 | 0% | 773 | 645 | -17% | 0 | 0 | — |
case-20 | fail→pass | 14,768 | 13,824 | -6% | 1 | 1 | 0% | 2,835 | 2,534 | -11% | 0 | 0 | — |
case-21 | fail→pass | 12,396 | 4,781 | -61% | 1 | 1 | 0% | 2,858 | 938 | -67% | 0 | 0 | — |
case-22 | fail→fail | 26,692 | 11,480 | -57% | 1 | 1 | 0% | 2,440 | 2,215 | -9% | 0 | 0 | — |
case-23 | fail→pass | 10,468 | 2,611 | -75% | 1 | 1 | 0% | 1,996 | 765 | -62% | 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 18 counted toward the lift figure. The other 5 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 +13 percentage points is the difference between those two pass rates over the 18 comparable cases. 5 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.