| name | atopile-skills |
| description | How to write and maintain `.claude/skills/*/SKILL.md` files: source-of-truth-first process, verification steps, and conventions. Use when this capability is needed. |
Skills Skill (Maintaining Skill Docs)
This skill describes the process for maintaining the skill documentation under .claude/skills/*/SKILL.md.
The goal is that future LLM edits stay accurate, actionable, and grounded in the repo (not vibes).
Quick Start
When updating any skill:
- Find the module’s “source-of-truth” docs (README/design notes).
- Verify claims directly in code (entrypoints + invariant-enforcing files).
- Fix incorrect paths/APIs/tests by searching the repo.
- Add/update a small
## Quick Start that actually runs in this repo.
- Validate frontmatter + referenced paths.
What “Good” Looks Like
A good skill doc is:
- Specific: points at exact files and the real entrypoints.
- Invariant-driven: documents the correctness rules enforced by the code (not aspirational design).
- Runnable: Quick Start snippets compile/import (or at least match the current API surface).
- Traceable: any non-obvious claim can be traced to a file path in the repo.
Standard Workflow (Source-of-Truth First)
1) Inventory the skill’s scope
- Identify the module boundary (directories, packages) and key consumers (“call sites”).
- Prefer using
rg over memory: look for imports, entrypoints, and key classes/functions.
2) Read the docs, then the code that enforces invariants
Use this hierarchy:
- A module README/design doc (if present)
