| name | write-a-skill |
| description | Use when creating, revising, or evaluating an agent skill, including skill descriptions, trigger behavior, progressive disclosure, and bundled resources. |
Write a skill
Core rule
Skills should be small, triggerable, and reusable. The description decides whether the skill loads.
Workflow
- Capture intent:
- task or domain
- trigger phrases and near misses
- expected output
- scripts or references needed
- Draft
SKILL.md:
name uses lowercase words and hyphens
description starts with Use when and names trigger conditions
- body gives the smallest useful workflow
- Add resources only when they save repeated work:
references/ for heavy guidance
scripts/ for deterministic helpers
- Review against realistic prompts before treating it as done.
Quality checklist
- Description is trigger-focused, not a workflow summary.
- Body is concise and scannable.
- Skill does not duplicate an existing skill.
- References are one level deep unless there is a strong reason.
- Validation path is explicit.