Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create, edit, and evaluate agent skills iteratively.
.claude/skills/hezaohezao-skill-creator/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -5% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -26% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 12% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -60% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -50% | 0% |
A skill for creating new skills and iteratively improving them.
At a high level, the process of creating a skill goes like this:
SKILL.md with YAML frontmatter + Markdown body)Your job when using this skill is to figure out where the user is in this process and then jump in and help them progress through these stages.
A Poirot skill lives in a directory containing SKILL.md:
skills/<category>/<skill-name>/SKILL.md # user skills
poirot/backend/agents/skill/builtin_skills/<category>/<skill-name>/SKILL.md # builtinFrontmatter fields:
yaml--- name: <skill-name> # unique, lowercase-hyphenated description: <≤60 char one sentence ending with a period> allowed-tools: # Poirot tools this skill may invoke - web_search - browse_page - bash - read_file - write_file - list_dir - str_replace - present_files - read_snapshot enabled: true related-skills: [<other-skill-name>] # optional cross-references license: MIT # recommended for contributed skills author: <human contributor or source attribution> ---
Body: # <Skill> title, 2-3 sentence intro, ## When to Use, ## How to Run, ## Procedure, ## Pitfalls, ## Verification.
description ≤ 60 characters, one sentence, ends with a period. Statethe capability, not the implementation. No marketing words.
MCP servers the skill explicitly expects. Do NOT name shell utilities the agent already has wrapped — grep → bash, cat/head/tail → read_file, sed/awk → str_replace, find/ls → list_dir.
scripts/, references in references/, templates intemplates/ inside the skill directory. Don't expect the model to inline-write non-trivial logic every call — ship a helper script.
SKILL.md/skill install <path> (or place under skills/)/skill list should show it activedescription for better triggering, body for clarityto non-existent tools
## When to Use section → model triggers skill on wrong tasksschema above
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 37,243 | 13,874 | -63% | 1 | 1 | 0% | 3,251 | 3,095 | -5% | 0 | 0 | — |
case-02 | fail→fail | 28,510 | 6,225 | -78% | 1 | 1 | 0% | 3,038 | 996 | -67% | 0 | 0 | — |
case-03 | fail→fail | 17,339 | 7,004 | -60% | 1 | 1 | 0% | 2,384 | 1,054 | -56% | 0 | 0 | — |
case-04 | fail→pass | 18,980 | 13,897 | -27% | 1 | 1 | 0% | 2,794 | 2,077 | -26% | 0 | 0 | — |
case-05 | fail→fail | 12,164 | 11,056 | -9% | 1 | 1 | 0% | 1,725 | 1,901 | +10% | 0 | 0 | — |
case-06 | pass→pass | 9,073 | 5,441 | -40% | 1 | 1 | 0% | 1,509 | 1,581 | +5% | 0 | 0 | — |
case-07 | pass→pass | 9,998 | 3,484 | -65% | 1 | 1 | 0% | 1,533 | 1,197 | -22% | 0 | 0 | — |
case-08 | fail→pass | 10,816 | 21,941 | +103% | 1 | 1 | 0% | 1,593 | 1,791 | +12% | 0 | 0 | — |
case-09 | fail→pass | 22,318 | 3,533 | -84% | 1 | 1 | 0% | 3,355 | 1,345 | -60% | 0 | 0 | — |
case-10 | pass→pass | 19,915 | 6,001 | -70% | 1 | 1 | 0% | 1,798 | 1,764 | -2% | 0 | 0 | — |
case-11 | pass→pass | 10,486 | 3,268 | -69% | 1 | 1 | 0% | 1,630 | 1,228 | -25% | 0 | 0 | — |
case-12 | pass→pass | 9,963 | 3,021 | -70% | 1 | 1 | 0% | 1,382 | 1,135 | -18% | 0 | 0 | — |
case-13 | pass→pass | 12,143 | 3,804 | -69% | 1 | 1 | 0% | 1,667 | 1,283 | -23% | 0 | 0 | — |
case-14 | fail→pass | 14,805 | 3,279 | -78% | 1 | 1 | 0% | 2,134 | 1,068 | -50% | 0 | 0 | — |
case-15 | fail→pass | 15,763 | 3,064 | -81% | 1 | 1 | 0% | 2,249 | 1,159 | -48% | 0 | 0 | — |
case-16 | pass→pass | 7,513 | 2,824 | -62% | 1 | 1 | 0% | 963 | 1,076 | +12% | 0 | 0 | — |
case-17 | fail→pass | 10,208 | 2,179 | -79% | 1 | 1 | 0% | 1,450 | 1,002 | -31% | 0 | 0 | — |
case-18 | pass→pass | 20,341 | 8,592 | -58% | 1 | 1 | 0% | 1,523 | 1,780 | +17% | 0 | 0 | — |
case-19 | fail→pass | 14,908 | 5,963 | -60% | 1 | 1 | 0% | 1,805 | 1,413 | -22% | 0 | 0 | — |
case-20 | pass→pass | 16,927 | 23,895 | +41% | 1 | 1 | 0% | 3,405 | 3,678 | +8% | 0 | 0 | — |
case-21 | pass→pass | 9,269 | 29,130 | +214% | 1 | 1 | 0% | 1,909 | 2,726 | +43% | 0 | 0 | — |
case-22 | pass→pass | 6,801 | 7,333 | +8% | 1 | 1 | 0% | 1,211 | 1,922 | +59% | 0 | 0 | — |
case-23 | fail→pass | 11,463 | 6,260 | -45% | 1 | 1 | 0% | 1,848 | 1,341 | -27% | 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 22 counted toward the lift figure. The other 1 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 +39 percentage points is the difference between those two pass rates over the 22 comparable cases.
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.