SKILL.md Template
A copy-paste starting point for a well-formed agent skill: trigger-bearing description, procedure, boundaries, and verification. Copy it below, fill the bracketed fields, delete what you don't need.
---
name: your-skill-name
description: Use when [specific task shapes or phrases that should trigger this skill]. Also use when [secondary triggers]. Do NOT use for [explicit exclusions].
---
# Your Skill Name
## Purpose
One paragraph: what class of work this skill makes reliable, and what failure it prevents.
## When to use
- [Concrete situation 1]
- [Concrete situation 2]
## When NOT to use
- [Situation where this skill would misfire or add overhead]
## Instructions
1. [First step, stated as an action with an observable result]
2. [Second step]
3. [Continue: one action per step, no compound steps]
## Verification
Before declaring this skill's work done:
- [ ] [Check 1: a command to run or artifact to inspect]
- [ ] [Check 2]
## Boundaries
- Never [destructive or out-of-scope action].
- Always [non-negotiable constraint].
## Examples
### Example: [realistic task]
[Input situation, the steps as applied, the verified result.]
Field notes
| Field | Why it matters |
|---|---|
| description (frontmatter) | The single most important line: agents match it against the task to decide whether to fire. Write it as trigger conditions, not marketing. Include exclusions. |
| When NOT to use | Missing in most published skills, and the main cause of misfires. If you cannot name a non-use case, the trigger is too broad. |
| Verification | A skill without checks cannot refuse to accept bad work. Make each check runnable or inspectable, not a vibe. |
| Boundaries | The governance layer inside the skill. Never-do lists here are what make a skill safe to auto-trigger. |