What is a skill?

A skill is a folder with one file — SKILL.md. That file tells the AI when to use the skill and what to do.

Each SKILL.md has two parts: a YAML frontmatter block and a body.

---
name: debug          ← how the skill is stored and found
description: >-
  Systematic debugging workflow. Use when hitting an error,
  test failure, or unexpected behavior.
                     ↑ when the AI chooses this skill (the trigger)
---

## Steps            ← what the AI does
1. Reproduce with a minimal case
2. Isolate: data, logic, or environment?
3. Fix the root cause, not the symptom
4. Add a regression test

How agents find skills (and what they do not do)

Linked editors receive a catalog of skill names and descriptions each session. When your message matches a skill’s description, the assistant reads that SKILL.md and follows its playbook.

• Not automatic domain routing — saying “Cloudflare” alone does not hard-wire a skill; the description must match what you are doing (for example deploy with Wrangler, review Workers code).
• Not auto-created mid-task — skills are installed (chain add), authored (chain new), or improved via the Brain pipeline (chain capture → chain reflect). The agent does not silently invent new skill files while you work.
• Registry stays in sync — if you copy a skill folder into skills/ by hand, Chain Hub auto-registers it in skills-registry.yaml on chain setup, hub load, or chain validate --fix so the hub never shows an Unknown bucket for on-disk skills.