skill-authorlisted

# skill-author The meta-skill that builds other skills. It encodes this repo's skill-format contract and ships the linter that enforces it — so a new skill loads, stays small, and is honest about its status. This file is itself linted by the tool it documents (the repo dogfoods it). ## When to use - Adding a new `.claude/skills/<slug>/` skill, or editing an existing one's frontmatter. - Before committing a skill — lint it so CI does not bounce it. - Deciding whether a skill needs a `scripts/` directory at all (usually it does not yet). ## Run the linter ``` python -m harness.skills lint .claude/skills # lint every skill python -m harness.skills lint .claude/skills/eval-loop-builder # lint one python .claude/skills/skill-author/scripts/run.py lint .claude/skills ``` Exit is non-zero on any error-level issue. ## The frontmatter contract ```yaml --- name: my-skill # lowercase-hyphen, <=64 chars, no "claude"/"anthropic", == folder name description: Use this when ... # 3rd person, trigger phrases, <=1024 chars version: 1.0.0 # SemVer compat: skill-format-1.0 # the format this repo targets status: experimental # experimental | stable | deprecated --- ``` - **name** must equal the folder name (the linter checks this). - **description** is the *only* thing the model sees before loading the body — front-load trigger phrases ("Use this when ..."). - **status** starts `experimental`; promote to `stable` only after the s