Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when an approved tutorial spec exists and the run needs a deterministic prerequisite graph before module planning. **Trigger**: concept graph, prerequisite graph, dependency graph, 概念图, 先修关系. **Use when**: `source-tutorial` 的 C2，已有 `output/TUTORIAL_SPEC.md`，需要把教程概念转成可排序的 DAG。 **Skip if**: 还没有 tutorial spec。 **Network**: none. **Guardrail**: 只做结构，不写 reader-facing prose；图必须保持无环。
.claude/skills/willoscar-concept-graph/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -62% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -60% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -74% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -55% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -75% | 0% |
Materializes the tutorial spec's structured concept inventory into outline/concept_graph.yml.
output/TUTORIAL_SPEC.mdoutline/concept_graph.ymlThe graph must contain:
nodes: {id, title, summary, source_ids, objective_refs}edges: {from, to} meaning prerequisite orderscripts/run.py should:
Do not duplicate spec-parsing heuristics in multiple places; keep them in shared tutorial tooling.
outline/concept_graph.yml exists| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 14,211 | 3,866 | -73% | 1 | 1 | 0% | 2,409 | 345 | -86% | 0 | 0 | — |
case-02 | fail→fail | 15,735 | 4,195 | -73% | 1 | 1 | 0% | 2,868 | 332 | -88% | 0 | 0 | — |
case-03 | fail→fail | 17,933 | 4,081 | -77% | 1 | 1 | 0% | 3,348 | 345 | -90% | 0 | 0 | — |
case-04 | fail→fail | 27,212 | 22,874 | -16% | 1 | 1 | 0% | 1,093 | 4,594 | +320% | 0 | 0 | — |
case-05 | pass→fail | 28,151 | 7,828 | -72% | 1 | 1 | 0% | 5,084 | 660 | -87% | 0 | 0 | — |
case-06 | pass→fail | 36,738 | 20,686 | -44% | 1 | 1 | 0% | 6,166 | 4,324 | -30% | 0 | 0 | — |
case-07 | fail→pass | 13,944 | 4,113 | -71% | 1 | 1 | 0% | 2,204 | 842 | -62% | 0 | 0 | — |
case-08 | fail→pass | 8,916 | 2,606 | -71% | 1 | 1 | 0% | 1,486 | 598 | -60% | 0 | 0 | — |
case-09 | pass→pass | 12,563 | 4,586 | -63% | 1 | 1 | 0% | 1,852 | 922 | -50% | 0 | 0 | — |
case-10 | pass→pass | 11,392 | 2,665 | -77% | 1 | 1 | 0% | 1,596 | 564 | -65% | 0 | 0 | — |
case-11 | fail→pass | 11,724 | 2,046 | -83% | 1 | 1 | 0% | 1,880 | 490 | -74% | 0 | 0 | — |
case-12 | fail→pass | 16,354 | 1,574 | -90% | 1 | 1 | 0% | 867 | 394 | -55% | 0 | 0 | — |
case-13 | fail→pass | 9,149 | 1,391 | -85% | 1 | 1 | 0% | 1,354 | 337 | -75% | 0 | 0 | — |
case-14 | fail→pass | 14,013 | 2,835 | -80% | 1 | 1 | 0% | 2,003 | 576 | -71% | 0 | 0 | — |
case-15 | pass→pass | 11,089 | 6,668 | -40% | 1 | 1 | 0% | 1,562 | 1,147 | -27% | 0 | 0 | — |
case-16 | fail→pass | 11,469 | 2,192 | -81% | 1 | 1 | 0% | 1,882 | 541 | -71% | 0 | 0 | — |
case-17 | pass→pass | 8,734 | 4,000 | -54% | 1 | 1 | 0% | 1,399 | 747 | -47% | 0 | 0 | — |
case-18 | pass→pass | 4,224 | 1,586 | -62% | 1 | 1 | 0% | 704 | 358 | -49% | 0 | 0 | — |
case-19 | fail→pass | 12,902 | 1,428 | -89% | 1 | 1 | 0% | 1,960 | 359 | -82% | 0 | 0 | — |
case-20 | fail→pass | 12,591 | 2,979 | -76% | 1 | 1 | 0% | 1,908 | 625 | -67% | 0 | 0 | — |
case-21 | fail→pass | 12,504 | 4,550 | -64% | 1 | 1 | 0% | 2,085 | 830 | -60% | 0 | 0 | — |
case-22 | pass→pass | 13,911 | 3,332 | -76% | 1 | 1 | 0% | 1,974 | 684 | -65% | 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. 4 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.