skill-quality-assurance

name	skill-quality-assurance
description	Use when reviewing or quality-checking a skill before shipping. Runs 6 checks: usefulness, authoring, structure, MCP, weight, output quality. Triggers on: 스킬 검토, skill review, validate skill, 배포 전 검토.
scenarios	["Review my skill — is it well-designed and ready to ship?","Is this SKILL.md too heavy? Should I split it?","Does my skill follow authoring best practices?","내 스킬 배포 전에 품질 검토해줘","스킬이 실제로 더 나은 출력을 만드는지 확인해줘"]
compatibility	{"recommended":["think-tool"],"optional":["mcp-reasoner"],"remote_mcp_note":"think-tool은 각 품질 차원 분석 전에 깊이 있는 판단을 위해 활용됩니다. Claude 설정 → MCP Servers에서 remote SSE 엔드포인트를 추가하세요."}

Skill Quality Assurance

Run 6 quality checks on a skill and produce an actionable improvement report.

This is the quality gate before publishing — and also useful mid-creation to catch design issues early. The "Top Improvements" section at the end maps directly to what skill-creator should fix next, making this the natural handoff between authoring and shipping.

Input

If the skill path wasn't provided, ask for it — you need the root directory containing SKILL.md.

Read all skill files before dispatching agents:

• SKILL.md (required)
• agents/*.md (if present)
• references/*.md (if present)
• scripts/* (if present)

If a directory doesn't exist, note its absence rather than skipping the check.

Execution

Run checks 1–5 in parallel. Run check 6 after — it depends on understanding the skill's promises first.

#	Check	Agent	What it examines
1	Usefulness	agents/usefulness-checker.md	Does this skill earn its place?
2	Authoring Principles	agents/authoring-checker.md	description ≤250 chars? Standing Mandates? Compaction-aware? effort field?
3	Agent Structure	agents/structure-reviewer.md	Should responsibilities split into subagents? Which steps can parallelize?
4	MCP Fit	agents/mcp-advisor.md	Which MCPs would genuinely help? (load references/mcp-catalog.md)
5	SKILL.md Weight	agents/weight-analyzer.md	Too heavy? What belongs in references/ or scripts/?
6	Output Quality	agents/eval-agent.md	Does the skill measurably improve output vs no-skill baseline?

Pass the full content of all skill files to each agent as context.

Output Format

Produce a self-contained report in this format:

## Skill QA Report: [skill-name]

### 1. Usefulness — [PASS / WARN / FAIL]
[findings]

### 2. Authoring Principles — [PASS / WARN / FAIL]
[findings]

### 3. Agent Structure — [GOOD / IMPROVABLE / MISSING]
**Persona separation:** [findings]
**Parallel opportunities:** [findings]

### 4. MCP Opportunities — [NONE / OPTIONAL / RECOMMENDED]
[findings]

### 5. SKILL.md Weight — [LIGHT / OK / HEAVY / CRITICAL]
[findings]

### 6. Output Quality — [PASS / MARGINAL / FAIL]
**With-skill pass rate:** X/N assertions
**Without-skill pass rate:** X/N assertions
**Delta:** +X
**Discriminating assertions:** [what the skill enforces that baseline misses]
**Skill gaps:** [what the skill promises but doesn't deliver]

---
## Top Improvements for [skill-name]
1. 🔴 [must fix — concrete and actionable]
2. 🟡 [recommended improvement]
3. 🟢 [optional enhancement]

Priority labels: 🔴 Must fix · 🟡 Recommended · 🟢 Optional

The "Top Improvements" section is what skill-creator reads to decide what to fix next. Make it concrete and actionable — not "improve structure" but "extract the grading logic into agents/grader.md and call it from SKILL.md with a Task()".