Ultimate Skills Finder

The ultimate multi-source OpenClaw/agent skills finder. Searches across ClawHub, skills.sh, Rush registry, LobeHub, SkillsMP, llmbase.ai, skillsllm.com, and GitHub curated collections — then cross-references, deduplicates, ranks by popularity, and optionally security-scans via Gen Digital. Use when: (1) user says 'find a skill for X', 'is there a skill that can...', 'how do I do X with OpenClaw', (2) user wants to discover new agent capabilities, (3) user asks what skills exist for a specific domain or task, (4) user wants to install a skill but isn't sure which one, (5) user wants to compare skills across multiple registries.

Searches 5 sources (ClawHub, SkillsMP, awesome-list, master-skills, skills.sh) for OpenClaw/agent skills, cross-references results, and returns ranked deduplicated findings with install instructions and optional security scanning.

Quick Start

bash

python3 scripts/find_skill.py "web scraping" --scan
python3 scripts/find_skill.py "pdf editor"
python3 scripts/find_skill.py "seo tools" --popular
python3 scripts/find_skill.py "github" --scan --install

Sources Searched

#	Source	Type	Skills	CLI/Tool
1	ClawHub (clawhub.ai)	Primary registry	~20,000	clawhub search
2	skills.sh (Vercel)	Directory	~87,000	npx skills add
3	SkillsMP (skillsmp.com)	Aggregator	1,000,000+	REST API
4	GitHub: awesome-openclaw-skills	Curated	5,200+	README
5	GitHub: openclaw-master-skills	Curated	560+	README

Workflow

Step 1: Understand the Need

When the user says "find me a skill for X", identify:

• The specific task/domain (e.g., "web scraping", "PDF editing", "SEO")
• The platform (OpenClaw, Claude Code, Cursor, etc.)
• Whether they want security scanning

Step 2: Run the Finder Script

bash

python3 scripts/find_skill.py "<query>" [options]

Options:

• --scan — Security-scan results via Gen Digital
• --popular — Sort by popularity/install count
• --limit N — Max results per source (default: 5)
• --json — Output as JSON for programmatic use
• --install — Show install commands prominently
• --source <name> — Only search specific source(s)

Step 3: Interpret Results

The script outputs a ranked, deduplicated table:

text

🔍 Results for "web scraping" (6 found across 9 sources)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🥇 agent-browser-cli          ★★★★☆  12.4k ⬇  [ClawHub, SkillsMP, llmbase]
   Headless browser automation with accessible tree
   → clawhub install agent-browser-cli
   → npx skills add user/agent-browser-cli

🥇 firecrawl                   ★★★★☆  8.2k ⬇   [ClawHub, GitHub awesome]
   Web scraping with Firecrawl API
   → clawhub install firecrawl

🥈 playwright-mcp             ★★★☆☆  3.1k ⬇   [ClawHub, SkillsMP]
   Browser automation via Playwright MCP
   → clawhub install playwright-mcp

🔒 Security: 2/3 skills scanned SAFE. 1 pending review.

Step 4: Security Scan (Optional)

Run with --scan to check skills via Gen Digital's API:

bash

python3 scripts/find_skill.py "email" --scan

The scanner checks each unique skill URL and returns:

• SAFE — OK to install
• analysis_pending — Not yet reviewed, caution advised
• WARNING / DANGEROUS / MALICIOUS — Do not install

Step 5: Present & Install

Present the top 3-5 most relevant results:

1. Skill name + description
2. Sources where it was found (confirms it's well-known)
3. Install commands (primary method)
4. Security verdict if scanned

Then ask which one(s) to install and do it:

bash

# ClawHub
clawhub install <slug>

# skills.sh
npx skills add <owner/repo>

# Manual (from GitHub)
git clone <repo> ~/.openclaw/workspace/skills/<name>

Cross-Reference Logic

Results are deduplicated by fuzzy name matching across all sources. A skill found in 3+ sources gets a confidence boost. The ranking formula:

text

score = (source_count * 0.3) + (install_count_normalized * 0.4) + (source_authority * 0.3)

Where:

• source_count: How many registries list this skill (0-10)
• install_count_normalized: Popularity within its source (0-1)
• source_authority: Official ClawHub > curated lists > aggregators > individual repos

When No Results Found

If the finder returns nothing:

1. Try a broader or synonym query (e.g., "email" → "smtp" → "messaging")
2. Check if the skill might exist under a different ecosystem
3. Offer to create a custom skill using skill-creator