Skill Deep Review
Perform a comprehensive quality review of a single skill, combining static analysis with live trigger testing via /skill-creator.
When to Use
- Reviewing a single skill in depth before publishing
- Investigating why a skill isn't triggering or producing poor results
- Validating a skill after major edits
- Quality-gating a skill before it enters the ecosystem
Inputs
- Skill path — path to the skill directory (must contain
SKILL.md)
- Context (optional) — what prompted the review (e.g., "it never triggers", "outputs are wrong")
Process
Phase 1: Structural Analysis
Read the skill's SKILL.md and all files in its directory tree. Evaluate against the rubric in references/deep-review-rubric.md. Score each dimension 1–5 and note specific issues.
Dimensions:
- Frontmatter compliance — required fields present, types correct, version valid semver, description follows patterns
- Description quality — action verb, trigger contexts, keyword variants, appropriate length, "pushiness"
- Progressive disclosure — body under 500 lines, references used appropriately, clear pointers to reference files
- Instruction clarity — imperative voice, logical flow, no ambiguity, explains "why" not just "what"
- Coordination — ownership declarations, composes_with accuracy, no overlaps with existing skills
- Completeness — all referenced files exist, no dead links, validation checklists present where needed
- Anti-patterns — no hardcoded project details, no excessive MUSTs/NEVERs without rationale, no duplicate content between body and references
Phase 2: Live Trigger Testing
Use /skill-creator's eval infrastructure to test whether the skill actually works:
- Generate 3–5 realistic test prompts that should trigger this skill
- Generate 2–3 near-miss prompts that should NOT trigger it
- Run trigger evaluation using skill-creator's description optimization tooling
- Report trigger accuracy (hit rate on should-trigger, false-positive rate on should-not)
Phase 3: Output Quality Sampling
If the skill produces structured output (reports, files, configs):
- Pick 2 representative test prompts
- Run them through
/skill-creator's test infrastructure
- Evaluate output against the skill's own stated format/expectations
- Note any gaps between promised and actual output
Phase 4: Generate Review Report
Produce a structured markdown report:
# Deep Review: [skill-name]
Reviewed: [timestamp]
Skill path: [path]
## Summary
| Dimension | Score (1-5) | Issues |
|-----------|-------------|--------|
| Frontmatter compliance | X | Y |
| Description quality | X | Y |
| Progressive disclosure | X | Y |
| Instruction clarity | X | Y |
| Coordination | X | Y |
| Completeness | X | Y |
| Anti-patterns | X | Y |
**Overall score:** X.X / 5.0
**Verdict:** SHIP | NEEDS WORK | MAJOR REWORK
## Trigger Testing
- Should-trigger hit rate: X/Y (Z%)
- False-positive rate: X/Y (Z%)
- Problem triggers: [list any that failed]
## Issues
### [SEVERITY]-[N]: [Title]
- **Dimension:** [which]
- **Location:** [file:line or frontmatter field]
- **Description:** [what's wrong]
- **Suggestion:** [how to fix]
## Strengths
[What the skill does well — specific examples]
## Recommendations
[Prioritized list of improvements, ordered by impact]
Save the report to {skill-path}/deep-review-report.md (or a location the user specifies).
Output
The review report is designed to feed directly into skill-improvement-plan, which can consume it and produce an actionable edit plan. Tell the user:
"Review complete. To act on these findings, you can feed this report into /skill-improvement-plan to generate a prioritized edit plan."
Guidelines
- Be constructive — every issue should have a concrete suggestion
- Score honestly but explain your reasoning, especially for low scores
- Don't nitpick style if the skill is functionally sound
- Weight trigger testing heavily — a skill that doesn't trigger is useless regardless of how well-written it is
- If the skill has known context (user said "it never triggers"), prioritize investigating that specific complaint
Reference Files
references/deep-review-rubric.md — Detailed scoring criteria for each dimension
1---2name: skill-deep-review3version: 1.1.04description: |5 Perform a thorough, single-skill deep dive reviewing structure, description quality, instruction clarity, progressive disclosure, anti-patterns, and frontmatter compliance — then run test prompts via /skill-creator to validate triggering and output quality. Use this skill when deeply reviewing one skill, auditing a specific skill's quality, checking if a skill triggers correctly, doing a "deep dive" on a skill, or when someone says "review this skill", "is this skill good", "check skill quality", or "deep review". Not for broad multi-skill scans — use skill-audit for that.6requires_agent_teams: false7requires_claude_code: true8min_plan: starter9owns:10 directories: []11 patterns: []12 shared_read: ["skills/"]13allowed_tools: ["Read", "Grep", "Glob", "Bash", "Agent", "Skill"]14composes_with: ["skill-writer", "skill-audit", "skill-improvement-plan"]15spawned_by: []16metadata:17 source: https://skillmd.com/skills/ivy00johns/skill-deep-review18---1920# Skill Deep Review2122Perform a comprehensive quality review of a single skill, combining static analysis with live trigger testing via `/skill-creator`.2324## When to Use2526- Reviewing a single skill in depth before publishing27- Investigating why a skill isn't triggering or producing poor results28- Validating a skill after major edits29- Quality-gating a skill before it enters the ecosystem3031## Inputs3233- **Skill path** — path to the skill directory (must contain `SKILL.md`)34- **Context (optional)** — what prompted the review (e.g., "it never triggers", "outputs are wrong")3536## Process3738### Phase 1: Structural Analysis3940Read the skill's `SKILL.md` and all files in its directory tree. Evaluate against the rubric in `references/deep-review-rubric.md`. Score each dimension 1–5 and note specific issues.4142**Dimensions:**43441. **Frontmatter compliance** — required fields present, types correct, version valid semver, description follows patterns452. **Description quality** — action verb, trigger contexts, keyword variants, appropriate length, "pushiness"463. **Progressive disclosure** — body under 500 lines, references used appropriately, clear pointers to reference files474. **Instruction clarity** — imperative voice, logical flow, no ambiguity, explains "why" not just "what"485. **Coordination** — ownership declarations, composes_with accuracy, no overlaps with existing skills496. **Completeness** — all referenced files exist, no dead links, validation checklists present where needed507. **Anti-patterns** — no hardcoded project details, no excessive MUSTs/NEVERs without rationale, no duplicate content between body and references5152### Phase 2: Live Trigger Testing5354Use `/skill-creator`'s eval infrastructure to test whether the skill actually works:55561. Generate 3–5 realistic test prompts that should trigger this skill572. Generate 2–3 near-miss prompts that should NOT trigger it583. Run trigger evaluation using skill-creator's description optimization tooling594. Report trigger accuracy (hit rate on should-trigger, false-positive rate on should-not)6061### Phase 3: Output Quality Sampling6263If the skill produces structured output (reports, files, configs):64651. Pick 2 representative test prompts662. Run them through `/skill-creator`'s test infrastructure673. Evaluate output against the skill's own stated format/expectations684. Note any gaps between promised and actual output6970### Phase 4: Generate Review Report7172Produce a structured markdown report:7374```markdown75# Deep Review: [skill-name]76Reviewed: [timestamp]77Skill path: [path]7879## Summary80| Dimension | Score (1-5) | Issues |81|-----------|-------------|--------|82| Frontmatter compliance | X | Y |83| Description quality | X | Y |84| Progressive disclosure | X | Y |85| Instruction clarity | X | Y |86| Coordination | X | Y |87| Completeness | X | Y |88| Anti-patterns | X | Y |8990**Overall score:** X.X / 5.091**Verdict:** SHIP | NEEDS WORK | MAJOR REWORK9293## Trigger Testing94- Should-trigger hit rate: X/Y (Z%)95- False-positive rate: X/Y (Z%)96- Problem triggers: [list any that failed]9798## Issues99100### [SEVERITY]-[N]: [Title]101- **Dimension:** [which]102- **Location:** [file:line or frontmatter field]103- **Description:** [what's wrong]104- **Suggestion:** [how to fix]105106## Strengths107[What the skill does well — specific examples]108109## Recommendations110[Prioritized list of improvements, ordered by impact]111```112113Save the report to `{skill-path}/deep-review-report.md` (or a location the user specifies).114115## Output116117The review report is designed to feed directly into **skill-improvement-plan**, which can consume it and produce an actionable edit plan. Tell the user:118119> "Review complete. To act on these findings, you can feed this report into `/skill-improvement-plan` to generate a prioritized edit plan."120121## Guidelines122123- Be constructive — every issue should have a concrete suggestion124- Score honestly but explain your reasoning, especially for low scores125- Don't nitpick style if the skill is functionally sound126- Weight trigger testing heavily — a skill that doesn't trigger is useless regardless of how well-written it is127- If the skill has known context (user said "it never triggers"), prioritize investigating that specific complaint128129## Reference Files130131- `references/deep-review-rubric.md` — Detailed scoring criteria for each dimension