
julien-skill-reviewer
by theflysurfer
Personal marketplace for developing and sharing custom Claude skills
SKILL.md
name: julien-skill-reviewer description: "Review and improve skill quality. Use when: user asks to review skill, check skill quality, improve skill, or after creating a skill with skill-creator-pro." license: Apache-2.0 metadata: author: "Julien" version: "2.0.0" category: "development" triggers:
- "review skill"
- "check skill"
- "improve skill"
- "skill quality"
- "audit skill"
- "score skill"
- "évaluer skill"
- "vérifier skill"
- "améliorer skill"
- "qualité skill"
- "auditer skill"
- "revoir skill"
- "relire skill"
- "review my skill quality"
- "check skill score"
- "évaluer la qualité de la skill"
- "améliorer ma skill" allowed-tools:
- Read
- Edit
- Write
- Glob
- Grep
- Bash
Skill Reviewer
Systematically evaluate and improve skill quality through scoring, refactoring, and iteration.
When to Use
Observability
First: At the start of execution, display:
🔧 Skill "julien-skill-reviewer" activated
- After creating a skill with skill-creator-pro
- When SKILL.md exceeds 500 lines
- Before packaging for distribution
- When quality score is unknown or low
- When skill feels bloated or has duplication
Review Process
Step 1: Read and Parse
# Read main file
Read skills/skill-name/SKILL.md
# Discover all resources
Glob skills/skill-name/**/*
# Check line count
wc -l skills/skill-name/SKILL.md
Parse YAML frontmatter and validate all fields. See references/yaml-validation.md for complete field specifications.
Step 2: Score Against Rubric
Evaluate 11 dimensions (1-5 scale). See references/quality-rubric.md for detailed criteria.
| Dimension | Check |
|---|---|
| Clarity | Clear steps, no ambiguity |
| Completeness | All scenarios covered |
| Discoverability | Good name + description |
| Context Efficiency | < 500 lines, refs used |
| Actionability | Concrete, copy-paste ready |
| Resource Organization | Proper scripts/refs/assets |
| Examples | Concrete, realistic |
| Skill Chaining | I/O/Dependencies documented |
| Error Handling | Troubleshooting provided |
| Trigger Quality | Natural language, bilingual, 10-20 triggers |
| Observability | Announces activation, clear output markers |
Step 2.5: Auto-Fix Mode (Optional)
Before manual review, optionally run auto-fix for common issues.
Ask user:
🔧 Auto-fix mode available:
A) Run safe auto-fixes (non-destructive)
B) Interactive fix (approve each change)
C) Skip auto-fix, manual review only
Safe Auto-Fixes (automatic, no approval):
- Add missing YAML fields (
version: 1.0.0,license: Apache-2.0) - Convert Windows paths (
\→/) - Generate TOC for files > 100 lines
- Warn if triggers < 5
Interactive Auto-Fixes (require approval):
- Detect hardcoded credentials → suggest
.env - Split files > 500 lines → suggest
references/ - Check description quality → suggest "what + when" format
Execution:
# Run auto-fix script (integrated)
python skills/julien-skill-reviewer/scripts/auto-fix.py path/to/skill/ --mode safe
python skills/julien-skill-reviewer/scripts/auto-fix.py path/to/skill/ --mode interactive
python skills/julien-skill-reviewer/scripts/auto-fix.py path/to/skill/ --dry-run
Safety:
- Always creates backup:
SKILL.md.backup-YYYYMMDD-HHMMSS - Dry-run mode available for preview
- All changes logged to console
Step 3: Calculate Score
Average = Sum of 11 scores / 11
| Score | Status |
|---|---|
| 1.0-2.4 | Needs major rework |
| 2.5-3.4 | Functional, needs improvement |
| 3.5-4.4 | Good, minor refinements |
| 4.5-5.0 | Excellent, production ready |
Minimum for production: 3.5/5, no dimension < 3/5
Step 4: Identify Issues
For each dimension < 4/5, document with line references. See references/refactoring-templates.md for issue format.
Step 5: Suggest Refactoring
Create prioritized plan:
- Priority 1: Critical (must fix)
- Priority 2: High impact
- Priority 3: Polish
See references/dry-patterns.md for common patterns.
Step 6: Execute (with approval)
Ask user before changes:
A) Show detailed plan
B) Execute Priority 1 automatically
C) Walk through one-by-one
D) Focus on specific dimension
Step 7: Re-evaluate
After refactoring:
- Re-read modified files
- Re-score all 11 dimensions
- Compare before/after
- Report improvement
Step 8: Iterate if Needed
If score < 3.5/5, continue iterations. Safety limit: Max 5 iterations.
Key Validation Checks
See references/validation-checks.md for complete list.
Required Fields
| Check | Requirement |
|---|---|
name | ≤ 64 chars, lowercase + hyphens only |
description | Third person, ≤ 1024 chars, includes trigger keywords |
Structure Checks
| Check | Requirement |
|---|---|
| Line count | SKILL.md < 500 lines |
| Paths | Forward slashes only (not Windows backslashes) |
| References | One level deep from SKILL.md |
| Long files | TOC if > 100 lines |
| No duplicates | Content in one place only |
Optional Fields
| Field | Validation |
|---|---|
version | Semantic versioning (X.Y.Z) |
license | Valid SPDX identifier (Apache-2.0, MIT, etc.) |
allowed-tools | Only valid tool names (Read, Write, Edit, Bash, etc.) |
user-invocable | Boolean only (true/false) |
disable-model-invocation | Boolean only (true/false) |
mode | Boolean only (true for mode commands, NOT interactive/batch/autonomous) |
context | fork only (for isolated subagent execution) |
agent | Valid agent type (Explore, Plan, Bash) |
model | Valid model (haiku, sonnet, opus, or full model ID) |
hooks | Correct syntax: event, matcher, command, optional once |
triggers | 5-50 natural language phrases (marketplace extension) |
Reference Files
| File | Content |
|---|---|
| references/quality-rubric.md | 11-dimension scoring details |
| references/dry-patterns.md | DRY violations and fixes |
| references/refactoring-templates.md | Report and plan templates |
| references/validation-checks.md | Automated and manual checks |
| references/yaml-validation.md | YAML frontmatter field validation |
Quality Assurance Checklist
Before marking production-ready:
Score Requirements
- Average score ≥ 3.5/5
- No dimension < 3/5
Structure Requirements
- SKILL.md ≤ 500 lines
- Proper progressive disclosure
- No DRY violations
- Skill Chaining documented
- Examples are concrete
- Error handling documented
- All references linked from SKILL.md
YAML Frontmatter Validation
-
name: max 64 chars, lowercase + hyphens only -
description: non-empty, max 1024 chars, includes trigger keywords -
version: follows semantic versioning (X.Y.Z) if present -
license: valid SPDX identifier if present (Apache-2.0, MIT) -
allowed-tools: contains only valid tool names if present -
mode: is boolean (true/false) if present, NOT string -
context: isforkif present -
agent: is valid type (Explore, Plan, Bash) if present -
model: is valid (haiku, sonnet, opus, or full ID) if present -
hooks: uses correct syntax withmatcherandcommandif present - YAML syntax is valid (no parse errors)
Marketplace Requirements
- Triggers: 10-20 natural language phrases in frontmatter
- Observability: Skill announces activation at start
Skill Chaining
Skills Required Before
- julien-dev-tools-skill-creator-pro (optionnel): If reviewing newly created skill
Input Expected
- Path to skill directory with SKILL.md
- Skill must have valid YAML frontmatter
Output Produced
- Format: Review report in markdown
- Side effects:
- May create
references/directory - May create new .md files
- May modify SKILL.md (with approval)
- Creates backup:
SKILL.md.backup-YYYYMMDD-HHMMSS
- May create
- Duration: 5-10 minutes (3-5 iterations)
Compatible Skills After
Recommandés (if score ≥ 3.5/5):
- Package script from skill-creator-pro
- julien-workflow-sync-personal-skills: Deploy to ~/.claude/skills/
Called By
- julien-dev-tools-skill-creator-pro: After Step 4, before packaging
- Direct invocation: "Review my skill X"
Tools Used
Read(usage: read SKILL.md, references/)Glob(usage: discover skill files)Grep(usage: search patterns, word count)Edit(usage: refactor SKILL.md)Write(usage: create references/*.md)Bash(usage: wc -l, backup files)
Visual Workflow
User: "Review skills/my-skill"
↓
[THIS SKILL]
├─► Read SKILL.md + refs
├─► Score 9 dimensions
├─► Identify issues
└─► Generate report
↓
Score < 3.5/5?
YES → Refactor → Re-evaluate → Iterate
NO → Production ready ✅
↓
[Next steps]
├─► package_skill.py
└─► sync-personal-skills
Usage Example
Scenario: Review deployment skill
Command: "Review the skill at skills/deployment-manager/"
Result:
- Initial score: 2.8/5
- Issues: DRY violation, bloated SKILL.md (6000 words), missing chaining
- After 3 iterations: 4.3/5
- SKILL.md: 6000 → 2500 words
- Created: references/api-reference.md, references/examples.md
- Status: Production ready
Score
Total Score
Based on repository quality metrics
SKILL.mdファイルが含まれている
ライセンスが設定されている
100文字以上の説明がある
GitHub Stars 100以上
3ヶ月以内に更新がある
10回以上フォークされている
オープンIssueが50未満
プログラミング言語が設定されている
1つ以上のタグが設定されている
Reviews
Reviews coming soon