Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when writing or editing a SKILL.md or an agent definition in this repo — required frontmatter, section ordering, and the GDScript-then-C# example convention.
.claude/skills/jame581-authoring-godot-prompter-skills/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -24% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -39% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 41% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 26% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -44% | 0% |
File format templates for this repo. See the root CLAUDE.md for the conventions and size budget that govern the content itself.
Every skill must start with YAML frontmatter:
yaml--- name: skill-name description: Use when [trigger] — [brief scope] ---
Followed by:
> **Related skills:** **skill-a** for X, **skill-b** for Y.gdscript and csharp language tags)Agent definitions in agents/<name>.md use YAML frontmatter:
yaml--- name: agent-name description: | When to use, with <example> blocks. model: inherit ---
Run node scripts/validate-skills.mjs — it enforces frontmatter, cross-references, and the 16 KB size budget.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 30,031 | 21,636 | -28% | 1 | 1 | 0% | 6,209 | 4,710 | -24% | 0 | 0 | — |
case-02 | fail→pass | 18,395 | 10,609 | -42% | 1 | 1 | 0% | 3,084 | 1,889 | -39% | 0 | 0 | — |
case-03 | fail→pass | 12,482 | 14,665 | +17% | 1 | 1 | 0% | 2,307 | 3,255 | +41% | 0 | 0 | — |
case-04 | fail→fail | 6,492 | 3,667 | -44% | 1 | 1 | 0% | 1,115 | 840 | -25% | 0 | 0 | — |
case-05 | pass→pass | 13,370 | 13,017 | -3% | 1 | 1 | 0% | 1,998 | 2,286 | +14% | 0 | 0 | — |
case-06 | fail→pass | 5,407 | 4,611 | -15% | 1 | 1 | 0% | 822 | 1,038 | +26% | 0 | 0 | — |
case-07 | fail→pass | 14,657 | 7,594 | -48% | 1 | 1 | 0% | 2,845 | 1,582 | -44% | 0 | 0 | — |
case-08 | pass→pass | 6,770 | 2,418 | -64% | 1 | 1 | 0% | 1,169 | 574 | -51% | 0 | 0 | — |
case-09 | fail→pass | 12,753 | 1,792 | -86% | 1 | 1 | 0% | 1,882 | 495 | -74% | 0 | 0 | — |
case-10 | fail→pass | 8,382 | 1,644 | -80% | 1 | 1 | 0% | 1,330 | 474 | -64% | 0 | 0 | — |
case-11 | pass→pass | 10,138 | 4,894 | -52% | 1 | 1 | 0% | 1,476 | 987 | -33% | 0 | 0 | — |
case-12 | fail→fail | 12,004 | 2,378 | -80% | 1 | 1 | 0% | 1,692 | 647 | -62% | 0 | 0 | — |
case-13 | pass→pass | 6,455 | 2,566 | -60% | 1 | 1 | 0% | 924 | 609 | -34% | 0 | 0 | — |
case-14 | fail→pass | 15,377 | 1,486 | -90% | 1 | 1 | 0% | 2,227 | 421 | -81% | 0 | 0 | — |
case-15 | fail→pass | 11,132 | 1,766 | -84% | 1 | 1 | 0% | 1,694 | 452 | -73% | 0 | 0 | — |
case-16 | fail→pass | 9,366 | 2,990 | -68% | 1 | 1 | 0% | 1,341 | 573 | -57% | 0 | 0 | — |
case-17 | pass→pass | 4,768 | 1,725 | -64% | 1 | 1 | 0% | 662 | 502 | -24% | 0 | 0 | — |
case-18 | pass→pass | 13,688 | 1,633 | -88% | 1 | 1 | 0% | 2,096 | 455 | -78% | 0 | 0 | — |
case-19 | fail→pass | 10,215 | 6,650 | -35% | 1 | 1 | 0% | 1,639 | 1,236 | -25% | 0 | 0 | — |
case-20 | pass→fail | 3,665 | 2,208 | -40% | 1 | 1 | 0% | 469 | 598 | +28% | 0 | 0 | — |
case-21 | pass→pass | 5,256 | 4,301 | -18% | 1 | 1 | 0% | 879 | 996 | +13% | 0 | 0 | — |
case-22 | pass→pass | 7,028 | 6,129 | -13% | 1 | 1 | 0% | 1,155 | 1,181 | +2% | 0 | 0 | — |
case-23 | pass→pass | 4,619 | 4,999 | +8% | 1 | 1 | 0% | 716 | 989 | +38% | 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. The headline lift of +43 percentage points is the difference between those two pass rates over the 23 comparable cases. 1 case got worse with the skill loaded, and it is 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.