Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create or improve codewhale skills. Use when the user wants a new skill, wants to update an existing skill, or needs guidance on when a skill should be a skill versus MCP, hooks, tools, or a plugin scaffold.
.claude/skills/hmbown-skill-creator/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -2% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 14% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -29% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -3% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 30% | 0% |
Use this skill to create small, useful codewhale skills that match the runtime this repository actually ships.
A skill is a local folder with a SKILL.md file. Codewhale reads the skill name and description during discovery, then loads the body only when the user or task matches the skill.
Discovery paths, in precedence order:
<workspace>/.agents/skills<workspace>/skills<workspace>/.opencode/skills<workspace>/.claude/skills<workspace>/.cursor/skills<workspace>/.codewhale/skills~/.agents/skills~/.claude/skills~/.codewhale/skillsUse skills for model instructions, workflows, and lightweight conventions. Use MCP for live external APIs or durable tools. Use hooks for automatic local events. To distribute a skill through a plugin bundle, give the bundle a plugin.json whose net.codewhale extension points skills.path at a directory holding <name>/SKILL.md folders; each skill then loads namespaced as <plugin>:<skill> after review, trust, and enablement. A bare SKILL.md directory is discoverable on the skills paths above but is not a plugin bundle and cannot be installed through /plugin install.
textmy-skill/ `-- SKILL.md
markdown--- name: my-skill description: Use when Codewhale should follow this specific workflow. --- # My Skill Instructions for the agent.
Frontmatter parsing is intentionally simple. Keep name and description as plain single-line values. Use lower-case hyphen-case names.
description action-oriented and trigger-specific. It is the mainsignal Codewhale sees before loading the body.
verify the result.
material.
references/ and mention exactly when to open them.scripts/ only for deterministic helpers that are worth maintaining.assets/ only for templates, fixtures, examples, or files reused by theworkflow.
intent and trust review.
<skill-name>/SKILL.md.name and description./skills or by running the relevantskill discovery tests if editing this repository.
unless the user asked for a rewrite.
from other agents that do not apply to codewhale.
SKILL.md starts with ---.name matches the directory name unless there is a deliberate reason.description says when to use the skill, not just what it is.optional.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 22,384 | 15,426 | -31% | 1 | 1 | 0% | 2,072 | 2,417 | +17% | 0 | 0 | — |
case-02 | fail→fail | 22,544 | 18,190 | -19% | 1 | 1 | 0% | 2,847 | 2,990 | +5% | 0 | 0 | — |
case-03 | fail→fail | 18,258 | 32,339 | +77% | 1 | 1 | 0% | 144 | 1,111 | +672% | 0 | 0 | — |
case-04 | fail→pass | 20,660 | 13,815 | -33% | 1 | 1 | 0% | 2,279 | 2,228 | -2% | 0 | 0 | — |
case-05 | pass→pass | 13,432 | 14,008 | +4% | 1 | 1 | 0% | 2,018 | 1,961 | -3% | 0 | 0 | — |
case-06 | pass→pass | 13,607 | 13,543 | -0% | 1 | 1 | 0% | 1,891 | 2,345 | +24% | 0 | 0 | — |
case-07 | pass→pass | 12,033 | 8,102 | -33% | 1 | 1 | 0% | 1,820 | 1,269 | -30% | 0 | 0 | — |
case-08 | fail→pass | 15,538 | 3,831 | -75% | 1 | 1 | 0% | 1,226 | 1,393 | +14% | 0 | 0 | — |
case-09 | fail→pass | 18,464 | 9,537 | -48% | 1 | 1 | 0% | 2,219 | 1,571 | -29% | 0 | 0 | — |
case-15 | pass→pass | 18,958 | 12,381 | -35% | 1 | 1 | 0% | 1,999 | 2,104 | +5% | 0 | 0 | — |
case-10 | pass→pass | 16,795 | 3,048 | -82% | 1 | 1 | 0% | 1,880 | 1,266 | -33% | 0 | 0 | — |
case-11 | fail→fail | 13,407 | 6,567 | -51% | 1 | 1 | 0% | 1,294 | 1,964 | +52% | 0 | 0 | — |
case-12 | pass→pass | 14,799 | 4,445 | -70% | 1 | 1 | 0% | 1,381 | 1,335 | -3% | 0 | 0 | — |
case-13 | pass→pass | 10,313 | 12,244 | +19% | 1 | 1 | 0% | 1,514 | 1,929 | +27% | 0 | 0 | — |
case-14 | fail→pass | 16,287 | 6,269 | -62% | 1 | 1 | 0% | 1,612 | 1,564 | -3% | 0 | 0 | — |
case-16 | pass→pass | 22,814 | 5,849 | -74% | 1 | 1 | 0% | 2,034 | 1,613 | -21% | 0 | 0 | — |
case-17 | pass→pass | 7,267 | 5,120 | -30% | 1 | 1 | 0% | 997 | 1,540 | +54% | 0 | 0 | — |
case-18 | pass→pass | 19,543 | 10,353 | -47% | 1 | 1 | 0% | 1,916 | 2,392 | +25% | 0 | 0 | — |
case-19 | pass→pass | 13,303 | 4,917 | -63% | 1 | 1 | 0% | 1,482 | 1,547 | +4% | 0 | 0 | — |
case-20 | fail→pass | 6,765 | 8,344 | +23% | 1 | 1 | 0% | 1,031 | 1,345 | +30% | 0 | 0 | — |
case-21 | fail→pass | 17,194 | 9,098 | -47% | 1 | 1 | 0% | 1,499 | 1,397 | -7% | 0 | 0 | — |
case-22 | pass→pass | 15,863 | 5,931 | -63% | 1 | 1 | 0% | 1,512 | 1,883 | +25% | 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 21 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 +27 percentage points is the difference between those two pass rates over the 21 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/11/2026 | +36% |
| gemini-3.6-flash | verified | 8/7/2026 | +45% |
Other measured skills in the registry, with their headline benchmark lift.