--- name: tool-discovery-agent description: Automatically searches for and installs helpful Claude Code tools/plugins from code.claude.com based on current task context. Use when starting any complex task to ensure optimal tooling is available. --- # Tool Discovery Agent Skill This skill autonomously discovers and installs Claude Code plugins/tools to optimize task completion. ## When to Use **Trigger this skill when:** - Starting a complex multi-step task - User asks "what tools can help with this?" - Encountering limitations in current capabilities - Task could benefit from specialized plugins - Need for automation, testing, or advanced features ## Discovery Process ### Step 1: Analyze Task Context Identify task requirements: ```bash # What type of task? - Development/coding → Check for language-specific tools - Design/UI → Search for design plugins - Testing → Find testing/automation tools - Documentation → Look for doc generators - Deployment → Search for DevOps tools - Data processing → Find data analysis plugins ``` ### Step 2: Search Claude Plugin Registry Use web-search or fetch Claude's plugin documentation: ``` https://code.claude.com/docs/en/discover-plugins https://github.com/anthropics/claude-plugins-official ``` **Search Query Pattern:** ``` "Claude Code plugin for [task_type]" "[language/framework] Claude Code tool" "Claude Code [feature] automation" ``` ### Step 3: Evaluate Plugin Options For each discovered plugin, assess: - **Relevance**: Does it solve current need? (0-10 score) - **Popularity**: Stars, downloads, community usage - **Maintenance**: Last updated, active development - **Compatibility**: Works with current Claude Code version - **Dependencies**: Required tools/languages installed? **Priority Matrix:** ``` HIGH PRIORITY (install immediately): - Score 8-10 relevance - Official Anthropic plugins - >1000 stars/active maintenance MEDIUM PRIORITY (ask user): - Score 5-7 relevance - Community plugins with good reviews - Specific use case matches LOW PRIORITY (note for future): - Score <5 relevance - Experimental/alpha plugins - Heavy dependencies ``` ### Step 4: Install Plugin **Installation Methods:** 1. **Marketplace Plugin:** ```bash /plugin marketplace add [repo] /plugin install [plugin-name] ``` 2. **Git Clone:** ```bash cd ~/.claude/plugins/ git clone [repo-url] ``` 3. **Skill Installation:** ```bash mkdir -p ~/.claude/skills/[skill-name] # Copy skill files ``` 4. **NPM Package:** ```bash npm install -g [package-name] # Link to Claude Code ``` ### Step 5: Verify Installation ```bash # Check plugin loaded ls -la ~/.claude/plugins/ | grep [plugin-name] # Test plugin functionality # Run simple command to verify # Update auto-trigger config if needed ``` ### Step 6: Configure Auto-Triggers Add to `~/.claude/hooks/auto-trigger-integration.json`: ```json { "triggers": { "[task_category]": { "keywords": ["keyword1", "keyword2"], "plugins": ["plugin-name"], "skills": ["skill-name"], "priority": "high" } } } ``` ## Tool Categories ### Development Tools - **Languages**: Python, JavaScript, TypeScript, Go, Rust - **Frameworks**: React, Next.js, Vue, Django, Rails - **Databases**: PostgreSQL, MongoDB, Redis - **APIs**: REST, GraphQL, gRPC ### Design Tools - **UI/UX**: Design systems, component libraries - **Prototyping**: Wireframing, mockups - **Assets**: Image optimization, icons ### Testing Tools - **Unit**: Jest, PyTest, Go tests - **E2E**: Playwright, Cypress, Puppeteer - **Performance**: Lighthouse, benchmarks ### DevOps Tools - **CI/CD**: GitHub Actions, GitLab CI - **Deployment**: Docker, Kubernetes, Vercel - **Monitoring**: Logging, metrics, alerts ### Documentation Tools - **Generators**: JSDoc, Sphinx, Docusaurus - **API Docs**: OpenAPI, Swagger - **Wikis**: Notion, Confluence integrations ### Automation Tools - **Workflows**: n8n, Zapier - **Scripts**: Shell, Python automation - **Tasks**: Task runners, Make ## Example Workflow **User Request:** "I need to test this React app" **Tool Discovery Process:** 1. **Analyze**: React app + testing = Playwright/Cypress 2. **Search**: "Claude Code plugin playwright testing" 3. **Find**: playwright-skill (https://github.com/lackeyjb/playwright-skill) 4. **Evaluate**: 10/10 relevance, official plugin 5. **Install**: ```bash cd /tmp && git clone https://github.com/lackeyjb/playwright-skill.git mkdir -p ~/.claude/skills/playwright-skill cp -r /tmp/playwright-skill/* ~/.claude/skills/playwright-skill/ ``` 6. **Configure**: Add browser_automation trigger 7. **Verify**: Test basic Playwright command 8. **Report**: "✅ Installed playwright-skill for React testing" ## Safety Checks **Before Installing:** - ✅ Verify plugin source (official GitHub/org) - ✅ Check for malicious code (review files) - ✅ Confirm no conflicts with existing plugins - ✅ Ensure system dependencies met - ✅ Ask user for permission on high-impact installs **Red Flags:** - ❌ Unofficial sources - ❌ Excessive permissions - ❌ Suspicious file operations - ❌ Outdated dependencies - ❌ Poor maintenance history ## Output Format **After discovery, report:** ``` === TOOL DISCOVERY RESULTS === Task: [user_task_description] 🔍 Found [N] relevant tools: [1] TOOL_NAME (HIGH PRIORITY) Description: [what it does] Repository: [url] Relevance: [X]/10 Status: ✅ INSTALLED [2] TOOL_NAME (MEDIUM PRIORITY) Description: [what it does] Repository: [url] Relevance: [X]/10 Status: ⏸️ AWAITING USER APPROVAL Configuration: • Auto-triggers updated for: [categories] • Plugin directory: [path] • Next steps: [recommendations] === ``` ## Best Practices 1. **Be Proactive**: Don't wait for user to ask 2. **Be Selective**: Only install high-value tools 3. **Be Transparent**: Always explain what you're installing 4. **Be Safe**: Verify sources before installation 5. **Be Efficient**: Batch related installations 6. **Document**: Keep record of installed tools ## Common Plugins to Install **Essential:** - claude-hud (monitoring) - claude-code-safety-net (safety) - playwright-skill (testing) - dev-browser (web automation) **Development:** - Language-specific plugins (Python, JS, Go, etc.) - Framework tools (React, Next.js, Django) - Database tools (PostgreSQL, MongoDB) **Productivity:** - planning-with-files (project planning) - repomix (code context packaging) - claude-mem (session memory)