Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Principles for writing skills that behave the same way every run — use when adding, editing, or reviewing a skill in this plugin
.claude/skills/hashgraph-online-skill-authoring/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | -10% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 28% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 53% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -16% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -47% | 0% |
HOTL skills are behavior-shaping code. Treat changes to skill text, agent prompts, command routers, and adapter instructions with the same care as production logic: define the failure mode, make the desired behavior explicit, verify the change, and update every index or mirror that exposes it.
Use this before changing:
skills/*/SKILL.mdagents/*.mdcommands/*.mdcline/rules/*.mdDo not use it for ordinary product docs, generated reports, or local project README edits unless they change how agents should behave.
name: kebab-case, no command-name collision.description: says when to activate, not the workflow steps.skills/using-hotl/SKILL.md.docs/skills.md, README tables, and Codex docs for user-visible skill changes.bats test/smoke.bats before claiming the skill change is ready.The frontmatter description is discovery text, not a miniature skill. If it contains a workflow summary, an agent may act on that summary without reading the full body.
Bad:
yamldescription: Use when executing plans - dispatches workers, reviews every task, then finalizes
Good:
yamldescription: Use when executing implementation workflows that can delegate contained steps to fresh subagents.
using-hotl.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 14,126 | 13,448 | -5% | 1 | 1 | 0% | 2,533 | 1,073 | -58% | 0 | 0 | — |
case-02 | fail→fail | 11,313 | 17,485 | +55% | 1 | 1 | 0% | 192 | 1,033 | +438% | 0 | 0 | — |
case-03 | fail→pass | 21,437 | 9,120 | -57% | 1 | 1 | 0% | 2,794 | 2,509 | -10% | 0 | 0 | — |
case-04 | fail→fail | 6,067 | 11,320 | +87% | 1 | 1 | 0% | 321 | 1,126 | +251% | 0 | 0 | — |
case-05 | fail→fail | 4,856 | 3,270 | -33% | 1 | 1 | 0% | 218 | 1,083 | +397% | 0 | 0 | — |
case-06 | fail→fail | 2,976 | 7,179 | +141% | 1 | 1 | 0% | 342 | 996 | +191% | 0 | 0 | — |
case-16 | pass→pass | 3,123 | 3,504 | +12% | 1 | 1 | 0% | 542 | 1,286 | +137% | 0 | 0 | — |
case-07 | pass→pass | 12,089 | 4,367 | -64% | 1 | 1 | 0% | 1,923 | 1,515 | -21% | 0 | 0 | — |
case-08 | pass→pass | 16,237 | 5,272 | -68% | 1 | 1 | 0% | 1,786 | 1,600 | -10% | 0 | 0 | — |
case-09 | fail→pass | 6,759 | 3,837 | -43% | 1 | 1 | 0% | 1,122 | 1,434 | +28% | 0 | 0 | — |
case-10 | pass→pass | 12,545 | 3,380 | -73% | 1 | 1 | 0% | 1,361 | 1,349 | -1% | 0 | 0 | — |
case-11 | pass→pass | 10,386 | 4,063 | -61% | 1 | 1 | 0% | 799 | 1,284 | +61% | 0 | 0 | — |
case-12 | fail→pass | 5,262 | 4,688 | -11% | 1 | 1 | 0% | 971 | 1,484 | +53% | 0 | 0 | — |
case-13 | pass→pass | 9,163 | 3,236 | -65% | 1 | 1 | 0% | 1,716 | 1,308 | -24% | 0 | 0 | — |
case-14 | fail→pass | 16,484 | 10,277 | -38% | 1 | 1 | 0% | 2,115 | 1,768 | -16% | 0 | 0 | — |
case-15 | fail→pass | 15,163 | 1,627 | -89% | 1 | 1 | 0% | 1,802 | 961 | -47% | 0 | 0 | — |
case-17 | pass→pass | 15,697 | 17,969 | +14% | 1 | 1 | 0% | 2,539 | 2,586 | +2% | 0 | 0 | — |
case-18 | pass→pass | 16,470 | 11,609 | -30% | 1 | 1 | 0% | 1,984 | 1,821 | -8% | 0 | 0 | — |
case-19 | fail→pass | 6,488 | 7,701 | +19% | 1 | 1 | 0% | 976 | 1,252 | +28% | 0 | 0 | — |
case-20 | pass→pass | 15,816 | 2,538 | -84% | 1 | 1 | 0% | 1,744 | 1,178 | -32% | 0 | 0 | — |
case-21 | pass→pass | 10,937 | 12,472 | +14% | 1 | 1 | 0% | 1,991 | 2,292 | +15% | 0 | 0 | — |
case-22 | fail→pass | 12,853 | 11,746 | -9% | 1 | 1 | 0% | 2,016 | 1,921 | -5% | 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 17 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 +32 percentage points is the difference between those two pass rates over the 17 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.