Skills
What it is
A Skill is a packaged set of instructions and resources that teaches an AI coding agent how to do a specific task. Instead of re-explaining your conventions in every conversation, you write them once as a Skill — a short description plus the steps, examples, and any helper files — and the agent loads it on demand when the task comes up. Think of it as a recipe the agent reaches for when the situation matches.
Strengths
- Captures know-how once and reuses it, so you stop repeating the same setup in every prompt.
- Loaded only when relevant, so it doesn't clutter the agent's context the rest of the time.
- Plain, readable files — usually Markdown — so they're easy to write, review, and version in git.
- Shareable across a team or projects, giving everyone the same trusted procedure.
Trade-offs
- A stale or vague Skill can steer the agent wrong; like docs, it needs maintenance.
- Too many overlapping Skills can confuse which one applies.
- It only helps when the agent recognizes the task and pulls it in.
- It's guidance, not a guarantee — the agent can still go off-script.
When to use it
Reach for a Skill when you have a repeatable procedure with specific rules — a deploy sequence, a code-style checklist, a domain-specific format — that you'd otherwise paste into the chat every time.
Vibe coding fit
Skills are how a plain assistant becomes your assistant. Encode the way your project wants things done — naming, testing, the order of operations — and the agent follows it automatically. Keep each Skill narrow and name it clearly so the agent knows exactly when to invoke it.
---
name: ship-checklist
description: Steps to take before shipping any change to production.
---
1. Run the test suite and confirm it passes.
2. Bump the version in package.json.
3. Update CHANGELOG.md with a one-line summary.
4. Commit, push, then deploy.