← ClaudeAtlas

kilnlisted

Lint a skill - static review of an existing SKILL.md for the flaws that silently weaken it: advisory rules that won't bind, contradictions, stale refs, missing output contract, unanchored claims, dead content, premature-completion steps, frontmatter drift. Produces prioritized action items. Read-only; does NOT run the skill. Use after writing/editing a skill, or before adopting someone else's.
sgtlaugh/skills · ★ 2 · Code & Development · score 76
Install: claude install-skill sgtlaugh/skills
# /kiln - fire a skill, find the cracks (static lint) Points at an existing skill and reports structural flaws + concrete fixes. It **reads; it never edits or runs the skill.** **Scope (state this in the output): kiln checks STRUCTURE, not CORRECTNESS.** A clean pass means the skill is well-*built*, NOT that its output is *right* - only running it on real inputs proves that (firing != correct). Say so; don't let a clean lint read as "verified." **Authoring reference** (how to write skills well - NOT re-taught here): mattpocock `writing-for-agents` (https://github.com/mattpocock/skills/tree/main/skills/productivity/writing-for-agents). kiln *checks for* the failure modes that guide names; for "how should I write this," point the user there. ## Step 1 - Locate & read - **No target given -> ask which skill to lint; don't guess.** - **`--fire` flag -> after the static checks, also run the adversarial test-generation pass (`--fire` section below). Generation only: kiln NEVER executes the skill.** - Resolve target, in order: a **bare skill name** (`occam`) -> `~/.claude/skills/<name>/SKILL.md` (the installed path; resolves symlinked skills too) - this is the normal case; else a **dir** -> its `SKILL.md`; else a **`SKILL.md` path** directly. If a name doesn't resolve there, Glob `**/<name>/SKILL.md` under `~/.claude` and `~/Github`; for a plugin skill (`plugin:name`), take the plugin's `installPath` from `~/.claude/plugins/installed_plugins.json` and read `<installPath>/skills/<