Skill Dedup Scanner

Scans installed skills for duplicates and naming conflicts. Detects similar skills that may cause model confusion. Use before publishing new skills or when t...

Security Scan

The skill's code, instructions, and requirements are coherent with its stated purpose (scanning SKILL.md files for duplicate names/descriptions); it has no network calls, no credential requests, and no install steps beyond a small Python dependency.

Purpose & Capability

Name/description (duplicate detection) match the code: the scanner reads SKILL.md frontmatter (name, description) from a skills directory, computes text similarity, and generates a report. Required artifacts (local skill files) and a lightweight YAML dependency are proportionate.

Instruction Scope

Runtime instructions tell the agent to run scripts/main.py against a skills directory; the code only reads SKILL.md files and frontmatter, computes similarities, and writes a report. Note: the scanner reads the entire SKILL.md into memory ('full_content'), so if those files contain sensitive data, that data could appear in memory and potentially in generated reports if you explicitly output them — the provided report generator only includes name/description snippets, not full content.

Install Mechanism

No install spec in registry; project is instruction+code only. The only declared dependency is PyYAML in requirements.txt, which is reasonable and expected.

Credentials

The skill requests no environment variables, no credentials, and no config paths. It does auto-detect system locale (uses Python locale) which is reasonable for its i18n feature.

Persistence & Privilege

No 'always: true' flag, no persistent system-wide changes, and no modification of other skills' configs. The tool may write an output report if given an output path, which is expected behavior.

Assessment

This package appears to do what it says: scan SKILL.md frontmatter and compare names/descriptions using SequenceMatcher. Before running: (1) inspect SKILL.md files for any sensitive data — although the report generator doesn't include full SKILL.md content, the scanner reads it into memory and you may choose to write reports to files; (2) be aware the similarity check is a simple text-based matcher (difflib.SequenceMatcher), not a semantic embedding model — it may miss conceptual similarity or over/under-report similarity; (3) the tool exits with status 1 when conflicts are found, which may cause CI or scripts to treat the run as a failure; (4) if you plan to run this in an environment that aggregates reports or shares outputs, avoid including directories that contain secrets. If you need stronger semantic similarity, consider using an embedding-based approach.

Detects duplicate or similar skills that may cause model confusion.

When to Use

• Before publishing a new skill
• When skills trigger incorrectly
• Auditing existing skill collection
• Resolving naming conflicts
• Checking for similar skill names/descriptions

Usage

Basic Scan

# Auto-detect language
python3 scripts/main.py ~/.openclaw/workspace/skills/

# Specify language
python3 scripts/main.py ~/.openclaw/workspace/skills/ --lang en
python3 scripts/main.py ~/.openclaw/workspace/skills/ --lang zh

Advanced Options

# Custom threshold (0-1)
python3 scripts/main.py ~/.openclaw/workspace/skills/ --threshold 0.8

# Output to file
python3 scripts/main.py ~/.openclaw/workspace/skills/ -o audit_report.md

# JSON output
python3 scripts/main.py ~/.openclaw/workspace/skills/ --json

# Verbose mode
python3 scripts/main.py ~/.openclaw/workspace/skills/ -v

Options

Option	Short	Description	Default
--lang	-l	Language: en, zh, auto	auto
--threshold	-t	Similarity threshold (0-1)	0.7
--output	-o	Output file path	stdout
--json	-j	Output in JSON format	false
--verbose	-v	Verbose output	false

Output

Generates a detailed report with:

• Similarity scores (name + description)
• Conflict analysis
• Actionable recommendations
• List of safe/unique skills

Similarity Thresholds

Score	Level	Action
> 0.85	🔴 High	Consider merging or renaming
0.7-0.85	🟡 Medium	Clarify descriptions
< 0.7	✅ Safe	No action needed

Language Support

• English (--lang en)
• Chinese (--lang zh)
• Auto-detect (--lang auto) - Detects system locale

Examples

Example 1: Quick Scan

$ python3 scripts/main.py ~/.openclaw/workspace/skills/
🔍 Scan Directory: /home/user/.openclaw/workspace/skills
📦 Total Skills: 15

✅ No Duplicate Skills Found

All skills have distinct names and descriptions. No optimization needed.

Example 2: Find Conflicts

$ python3 scripts/main.py ~/.openclaw/workspace/skills/ --threshold 0.6
🔍 Scan Directory: /home/user/.openclaw/workspace/skills
📦 Total Skills: 15

⚠️ Found 2 Groups of Similar Skills

### 1. 🔴 High Similarity Warning

**Skill A:** `tushare-finance`
**Skill B:** `stock-analyzer`

**Similarity Score:** 85%

**Analysis:**
- Name Similarity: 60%
- Description Similarity: 85%

**Recommendations:**
- Clarify the distinction in descriptions
- Rename to make purposes more distinct

Project Structure

skill-dedup-scanner/
├── scripts/
│   ├── main.py                 # Main entry point
│   ├── skill_scanner.py        # Skill scanner
│   ├── similarity_checker.py   # Similarity calculator
│   ├── report_generator.py     # Report generator
│   └── locale_loader.py        # Multi-language support
├── locales/
│   ├── en.json                 # English translations
│   └── zh.json                 # Chinese translations
├── references/
│   └── best_practices.md       # Skill naming best practices
└── output/                     # Generated reports

Requirements

• Python 3.7+
• PyYAML (pip install pyyaml)

License

MIT