Skill

---
name: skill
description: 'Use when the user wants an existing or in-progress skill checked or structured against THIS project''s house skill conventions — "what are our skill conventions", "review this skill against our standards", "does this skill follow our patterns", "structure this skill the way we do", "audit this skill''s front-matter/description for our rules". Not for: creating a skill from scratch, optimizing a skill''s description for triggering, or running skill evals (all skill-creator); planning (/aidex:plan); decisions (/aidex:decision); requests (/aidex:request); research (/aidex:research); references (/aidex:reference); ecosystem audits (/aidex:aidex).'
disable-model-invocation: false
allowed-tools: Bash Read Write
---

# Skill Conventions

Surface and apply this project's house skill conventions to an existing or
in-progress skill. This skill COMPLEMENTS `skill-creator` (it does **not**
create skills from scratch).

## Workflow

1. Read the skill conventions canon:
   `${CLAUDE_PLUGIN_ROOT}/skills/conventions/references/skill-conventions.md`
   (or the `.claude/skills/...` project-level copy if one exists).
2. Identify the target skill the user is reviewing or structuring.
3. Produce a gap report against the canon: front-matter (trigger-first
   `description`, char budget, **English-only `description`**,
   `disable-model-invocation`, `allowed-tools`), body structure, `evals/`
   presence, no per-skill `CHANGELOG.md`/`README.md`.
   - **English-only check** (canon `skill-conventions.md` rule 2): the
     `description` must contain no Spanish/other-language phrasings or accented
     tokens. The matcher bridges cross-lingually, so an English description still
     fires on non-English queries; native-language queries belong in the
     `evals/` query set, not the description.
     Flag any non-English phrasing as a gap and rewrite the description
     English-only, preserving semantic coverage (drop the foreign phrasings;
     keep their English equivalents).
4. Writes go only to the target skill's own files (`SKILL.md`, `evals/`) —
   never to the canon under `references/`, which this skill reads read-only.

## Boundaries

| The user wants to… | Route to |
|---|---|
| Create a skill from scratch / optimize a description / run skill evals | `skill-creator` |
| Plan multi-step / multi-phase implementation work | `plan` |
| Record a decision / ADR | `decision` |
| Capture a stakeholder/client request | `request` |
| Investigate / research how something works | `research` |
| Document a settled system reference | `reference` |
| Audit the Claude Code ecosystem | `aidex` |

## Related

- **skill-creator** — owns skill creation, description optimization, and
  eval running; this skill only checks/applies house conventions.
- **conventions** — owns the shared documentation canon (this skill
  reads its `references/skill-conventions.md`).