Skip to content

Skill Format

Eclipse Ghost edited this page Jul 7, 2026 · 1 revision

Skill Format

Every skill lives in skills/<name>/ with this structure: skills// ├── SKILL.md # main content ├── metadata.json # machine-readable metadata ├── examples/ # runnable code samples ├── prompts/ # ready-made agent prompts └── tests/ # validation tests

SKILL.md sections (in order)

  1. Overview — what the API/tool does, in 2-3 lines
  2. Installation — real install commands (npm/pip/etc.)
  3. Authentication — how auth actually works, key types, where to get them
  4. Core Concepts — key terms/objects the agent needs to know
  5. Common Workflows — step-by-step for the most frequent tasks
  6. Error Handling — real error codes/types and how to handle them
  7. Security — things that will bite you if ignored (secrets, webhook verification, etc.)
  8. Rate Limits — actual numbers, not vague statements
  9. Best Practices — patterns that avoid common mistakes
  10. Troubleshooting — real gotchas, not generic advice
  11. References — links to official docs

See skills/stripe/SKILL.md for a full example.

metadata.json

{
  "name": "stripe",
  "version": "1.0.0",
  "description": "...",
  "categories": ["Payments", "Commerce"],
  "languages": ["typescript", "python"],
  "license": "MIT",
  "links": { "API Reference": "https://..." },
  "schemaVersion": "1.0.0",
  "supportedAgents": ["cursor", "claude-code", "cline", "continue"],
  "lastVerified": "2026-07-03"
}

Quality rules (non-negotiable)

  • No placeholder text, no "TODO," no vague filler.
  • Every command/API call must be real — verified against actual docs.
  • Code in examples/ should run, not just look plausible.
  • lastVerified date must reflect an actual check, not be copy-pasted.

Full schema: SPECIFICATION.md

Clone this wiki locally