🔍 Skill-Seeker

Discover and Install Community AI Skills for Claude Code

Transform Claude Code with behavioral frameworks, prompt-based skills, and specialized AI personas

Quick Start • Features • Documentation • Contributing • Security

---

A Claude Code plugin that discovers, evaluates, and installs community-created skills from GitHub. Before tackling a complex task, Skill-Seeker searches for existing behavioral frameworks, prompt-based skills, and specialized personas that can be injected into your context.

Think of it as npm for AI prompts or a package manager for Claude Code behavioral frameworks. Instead of manually searching GitHub for best practices, design patterns, or coding standards, Skill-Seeker automates discovery, validates quality, scans for security issues, and installs skills with a single command.

Perfect for developers who want Claude to:

• Apply React hooks best practices
• Follow REST API design patterns
• Conduct security audits with OWASP guidelines
• Write Playwright browser tests correctly
• Use Terraform infrastructure patterns
• Apply clean code principles
• Follow specific framework conventions

⭐ Like this project? Give us a star on GitHub to help others discover Skill-Seeker!

🎯 What It Does

Skill-Seeker is a meta-skill: it teaches Claude Code to autonomously find and load other skills. Instead of searching for executable MCP tools, it focuses on behavioral frameworks and prompt-based skills—Markdown files that shape how the AI approaches a task.

✨ Features

• 🔍 Smart Discovery: Hybrid search combining curated registry + live GitHub API search
• 🛡️ Security First: Comprehensive security scanner with 9 detection categories
• ⚡ Instant Results: Curated registry provides zero-latency search for popular skills
• 🎯 Quality Scoring: Trust scores based on stars, activity, and community validation
• 📦 Easy Installation: One-command skill installation with automatic caching
• 🔄 Session Management: Load and unload skills dynamically
• 🌐 Community-Driven: Submit your own skills to the curated registry
• 🏷️ Smart Tagging: Query expansion with synonyms for better discoverability

💡 Use Cases

Skill-Seeker helps developers solve these common challenges:

Frontend Development

• React component patterns - Learn hooks, composition, and state management
• CSS architecture - BEM, CSS-in-JS, Tailwind patterns
• UI/UX design audit - Apply professional design principles
• Responsive design - Mobile-first, breakpoints, fluid layouts
• Web accessibility - WCAG compliance, ARIA attributes, screen readers
• Performance optimization - Bundle size, lazy loading, code splitting

Backend Development

• REST API design - Endpoint naming, HTTP methods, status codes
• GraphQL schema patterns - Type design, resolvers, N+1 prevention
• Database design - Schema normalization, indexing, query optimization
• Microservices architecture - Service boundaries, communication patterns
• Error handling - Centralized handlers, logging, user messages
• Authentication & authorization - JWT, OAuth, RBAC, security best practices

Testing & Quality Assurance

• Playwright browser testing - E2E tests, page objects, reliable selectors
• Unit testing patterns - Mocking, test isolation, assertion strategies
• Integration testing - API testing, database testing, test data management
• Test-driven development - Red-green-refactor, test-first approach
• Code coverage - Coverage goals, meaningful metrics

DevOps & Infrastructure

• Terraform infrastructure - Module design, state management, best practices
• Docker optimization - Multi-stage builds, layer caching, image size reduction
• Kubernetes deployment - YAML manifests, helm charts, resource management
• CI/CD pipelines - GitHub Actions, build optimization, deployment strategies
• Monitoring & logging - Observability, alerting, debugging production issues

Security

• OWASP Top 10 - SQL injection, XSS, CSRF, security misconfiguration
• Secure coding practices - Input validation, output encoding, parameterized queries
• Authentication security - Password hashing, session management, token security
• API security - Rate limiting, API keys, OAuth flows
• Dependency scanning - Vulnerability detection, update strategies

Code Quality

• Clean code principles - SOLID, DRY, KISS, naming conventions
• Refactoring patterns - Extract method, simplify conditionals, remove duplication
• Code review checklist - What to look for, constructive feedback
• Documentation standards - JSDoc, docstrings, README templates
• Git workflow - Commit messages, branching strategies, PR conventions

See TROUBLESHOOTING.md for detailed usage tips and common questions.

Example Workflow

You: "Help me redesign this landing page"

Claude (with Skill-Seeker loaded):
  1. Recognizes this is a design-taste task
  2. Searches GitHub for frontend design skills
  3. Finds pbakaus/impeccable (strict design audit framework)
  4. Shows you the results with trust scores
  5. You pick which skill to install
  6. Runs security scan, shows preview
  7. You approve → skill is loaded into context
  8. Claude now applies the design framework to your task

📚 Quick Start

Installation

From the Claude Code plugin marketplace (recommended):

Inside Claude Code, run:

/plugin marketplace add mmmantasrrr/skill-seeker
/plugin install skill-seeker@skill-seeker

One-line install from terminal:

curl -fsSL https://raw.githubusercontent.com/mmmantasrrr/skill-seeker/main/install.sh | bash

Test locally without installing:

git clone https://github.com/mmmantasrrr/skill-seeker.git
claude --plugin-dir ./skill-seeker

After installing, restart Claude Code to load the plugin. Then try /skill-seeker:seek react hooks.

Dependencies: curl, jq, and python3. Set GITHUB_TOKEN for higher API rate limits (5,000 vs 60 requests/hour).

Updating

Plugin updates are handled automatically by Claude Code. To manually refresh:

/plugin marketplace update skill-seeker

New Features (v2.0)

⚡ Reduced Context Overhead: Pre-flight skill slimmed from ~961 to ~200 tokens (80% reduction)

🔍 Combined Search: Single script merges registry + GitHub results with automatic deduplication

🚀 Registry Quick-Install: Install verified skills by ID: /skill-seeker:install metaskills-skill-builder

📊 Status Command: See loaded skills and context budget at a glance

💾 Cache Reload: Instantly reload previously installed skills without re-fetching

📁 Project Profiles: Define project-specific skill preferences in .skill-seeker-profile.json

🎯 Auto-Browse: HIGH trust verified skills auto-browse to collapse seek+browse into one step

See PROJECT-PROFILES.md for detailed information on project profiles.

Commands

Command	Purpose
/skill-seeker:seek <query>	Search GitHub for skills matching a query
/skill-seeker:browse <owner/repo>	List all skills in a specific repository
/skill-seeker:install <target>	Fetch, scan, and install a skill (supports registry IDs)
/skill-seeker:status	Show loaded skills and context budget
/skill-seeker:reload [target]	Quickly reload a skill from cache
/skill-seeker:unload	Clean up temporary skill files

Prerequisites

• curl and jq for GitHub API calls
• python3 for security scanning
• Optional: GITHUB_TOKEN environment variable for higher API rate limits (5,000 vs 60 requests/hour)

Architecture

See ARCHITECTURE.md for a detailed technical breakdown covering:

• Lead 1: Discovery & search mechanics (GitHub API strategies)
• Lead 2: Ingestion & parsing (file identification and fetching)
• Lead 3: Context injection & lifecycle (loading, temp storage, unloading)
• Lead 4: Trigger mechanism & security (invocation patterns, prompt injection defense)

🤝 Contributing

skill-seeker/
├── .claude-plugin/
│   ├── plugin.json          # Plugin metadata
│   └── marketplace.json     # Marketplace catalog for plugin discovery
├── commands/
│   ├── seek.md              # Search GitHub for skills
│   ├── browse.md            # Browse skills in a repo
│   ├── install.md           # Fetch, scan, and install a skill (supports registry IDs)
│   ├── status.md            # Show loaded skills and context budget
│   ├── reload.md            # Quickly reload skills from cache
│   └── unload.md            # Clean up temporary files
├── scripts/
│   ├── search-combined.sh   # Combined registry + GitHub search
│   ├── search-registry.sh   # Curated registry search
│   ├── search-github.sh     # GitHub API search & browse
│   ├── resolve-skill-id.sh  # Resolve registry IDs to repo/path
│   ├── fetch-skill.sh       # Raw markdown fetching with caching
│   └── scan-skill.py        # Security scanner (9 detection categories)
├── skills/
│   └── seeking-skills/
│       └── SKILL.md         # Meta-skill: when/how to seek skills
├── tests/
│   └── test_scan_skill.py   # Security scanner tests (34 tests)
├── registry.json            # Curated skill registry
├── .skill-seeker-profile.example.json  # Example project profile
├── ARCHITECTURE.md          # Technical deep-dive
├── PROJECT-PROFILES.md      # Project skill profiles guide
└── README.md                # This file

🔐 Security Model

All community content passes through the security scanner before injection. See SECURITY.md for details.

The scanner checks for:

Category	Severity	What It Detects
Instruction overrides	CRITICAL	"ignore previous instructions", mode switching
Shell command injection	CRITICAL	Tool abuse, piped curl commands
Data exfiltration	CRITICAL	Attempts to read SSH keys, credentials, env vars
Hidden HTML content	HIGH	Invisible text, CSS tricks, script tags
Social engineering	HIGH	"don't tell the user", secret-keeping instructions
Encoded payloads	HIGH	Base64-encoded malicious instructions
Conversation injection	MEDIUM	Fake turn markers, prompt format exploitation
Unicode obfuscation	MEDIUM	Zero-width chars, RTL overrides, homoglyphs
Frontmatter abuse	MEDIUM	Unexpected YAML keys that could alter behavior

Risk levels: CLEAN (0) → LOW (1–14) → MEDIUM (15–29) → HIGH (30–49) → CRITICAL (50+)

Skills flagged HIGH or CRITICAL are blocked by default. The user must explicitly acknowledge findings before proceeding.

How It Finds Skills

The search uses a hybrid approach combining curated registry + live GitHub search:

Strategy 0: Curated Registry (NEW)

• Instant search of verified, high-quality skills
• Rich metadata with tags, aliases, and synonyms
• Query expansion: "frontend" automatically searches "ui", "web-design", etc.
• No rate limits: Always available, no API calls needed
• Community-curated: Submit your skills via PR (see CONTRIBUTING-REGISTRY.md)

Strategies 1-3: Live GitHub API Search

1. Topic search: Repos tagged with claude-skills or claude-code-skills + query terms
2. Broad topic search: Repos tagged with claude-skills + query terms (wider net)
3. Description search: Repos with "claude" + "skill" in description/name + query terms

Results are scored on: stars (log scale), recency, forks, proper tagging, license presence, and archive status. Only repos with 10+ stars are shown by default.

Why the registry? Ensures you never encounter "skill not found" for popular use cases. The registry guarantees that well-known skills (Playwright testing, Terraform patterns, security checklists, etc.) are always discoverable, even if they're not properly tagged on GitHub.

📖 Documentation

• ARCHITECTURE.md - Technical deep-dive into system design
• CONTRIBUTING-REGISTRY.md - How to add skills to the registry
• TROUBLESHOOTING.md - Solutions to common problems and FAQ
• PROJECT-PROFILES.md - Project skill profiles guide
• SECURITY.md - Security policy and vulnerability reporting
• CHANGELOG.md - Version history and updates
• CODE_OF_CONDUCT.md - Community guidelines

🌟 Showcase

Want to see your skill featured here? Submit it to our registry!

Featured Skills:

• Impeccable Design Audit - Strict frontend design framework
• Agency Agents - Specialized AI personas

🎯 Roadmap

• Curated registry with verified skills
• Hybrid search (registry + GitHub API)
• Security scanning with 9 detection categories
• Native Claude Code plugin marketplace integration
• Multi-source search (GitLab, Bitbucket)
• Local caching and offline mode
• Semantic search with embeddings
• Usage analytics and recommendations
• Browser-based skill marketplace

💬 Community & Support

• Issues: Report bugs or request features via GitHub Issues
• Discussions: Join conversations about skills and best practices
• Pull Requests: Contribute code or skills following our guidelines

📊 Comparison

Feature	Skill-Seeker	Manual Search	MCP Tools
Discovery	Automated with trust scores	Manual GitHub browsing	Requires knowledge of tool names
Security	Built-in scanner	User responsibility	Varies by tool
Type	Behavioral frameworks	N/A	Executable tools
Installation	One command	Manual copy-paste	Installation required
Session Management	Dynamic load/unload	N/A	Always available

🙏 Inspiration

• pbakaus/impeccable — Strict frontend design taste framework
• msitarzewski/agency-agents — Dozens of specialized AI personas
• GitHub topics: claude-skills, claude-code-skills

📝 License

MIT License - see LICENSE for details.

---

Made with ❤️ for the Claude Code community

If you find Skill-Seeker useful, please consider starring the repository to help others discover it!