← ClaudeAtlas

writing-skillslisted

Use when creating a new skill or editing an existing one in this library. Defines the lean format — thin SKILL.md, progressive disclosure, trigger-style descriptions, complexity gates — and how to prove a skill works. This is for AUTHORING skills, not using them.
NjoyimPeguy/augments · ★ 1 · AI & Automation · score 74
Install: claude install-skill NjoyimPeguy/augments
# Writing Skills Skills are tools, not pipelines. Each one loads into context every time it fires, so every line costs tokens on every use. Write the minimum that changes behavior; push the rest to sibling files. ## When to use - Creating or editing a skill under `skills/<phase>/<name>/`. - **Skip** when you're *using* a skill — this is only for authoring. ## Skill types Match the form to the need (see `reference.md` for how much detail each needs): - **Instruction** — a prose procedure. Most skills. - **Template** — ships a `{{placeholder}}` file to copy (like `skill-template.md`). - **Script** — bundles a tested, deterministic script when prose would be error-prone. - **Reference** — a doc loaded on demand for lookup; large is fine, it isn't always-loaded. ## The format (non-negotiable) 1. **Frontmatter**: `name` (kebab-case, matches the directory) and `description` (capability + "Use when…" trigger, ≤1024 chars, third person, **never** a workflow summary). Add `disable-model-invocation: true` only for pure prompt-injection skills with no agentic steps (e.g. a one-line `zoom-out`). 2. **Body ≤ ~80 lines** (capability skills; discipline skills are the exception — see below). Intent + procedure only. Cut marketing ("why this matters") and long worked examples. 3. **Progressive disclosure.** Templates, long examples, lookup tables, scripts → sibling files. SKILL.md links to them; it never inlines them. 4. **Complexity gate up top.** State when to *skip* the skill. Cere