UNPKG

claude-loop

Version:

Pure Claude CLI autonomy - iteratively fixes your repo until everything works. No hardcoded patterns, just raw AI power.

231 lines (189 loc) 6.5 kB
# Claude Loop Debug Prompts ## Main Loop Prompt (Based on Your Style) ``` systematically debug and fix this repository so all functionality works as expected without adding complexity and leveraging existing logic Update Todos: ⎿ ☐ Analyze current application structure and identify all components ☐ Identify and fix all broken button handlers and workflows ☐ Audit all links and routes for missing or broken connections ☐ Test all UI components for visual and functional issues ☐ Ensure all database operations work correctly ☐ Fix any permission or authentication issues ☐ Verify all sections display correctly with proper styling ``` ## Specialized Agent Prompts ### 1. Button & Workflow Debug Agent ``` You are debugging button functionality and workflows. Focus on: - Finding all onClick handlers that don't work - Tracing button actions to their implementations - Identifying missing or broken workflows - Fixing handleCreateProduction-style function issues - Ensuring all user interactions have proper responses Common issues to look for: - Missing event handlers - Incorrect function bindings - Broken promise chains - Missing route definitions - Database constraint violations ``` ### 2. Route & Navigation Debug Agent ``` You are auditing links and navigation. Tasks: - Systematically check all Link components and href attributes - Identify routes that lead to 404s - Find buttons linking to non-existent pages - Verify all navigation flows work end-to-end - Check for missing route definitions Pattern to follow: 1. List all routes defined in the application 2. List all links/buttons that navigate 3. Cross-reference to find mismatches 4. Fix broken connections ``` ### 3. UI Component Debug Agent ``` You are testing visual and functional UI issues. Check each component for: - Visual rendering problems - Language switching functionality - Responsive design issues - Component state management - Data display correctness Sections to test systematically: - banco-vocabulario functionality - listas-vocabulario display - produccion workflows - All form submissions - Modal and dialog behaviors ``` ### 4. Database & Backend Debug Agent ``` You are fixing backend and database issues. Focus areas: - Database constraint violations - Missing migrations or schema issues - API endpoint failures - Data validation problems - Permission and authentication errors Approach: 1. Test each database operation 2. Verify constraints are satisfied 3. Check API responses 4. Ensure proper error handling ``` ## Your Specific Debug Patterns Based on your history, here are the patterns claude-loop follows: ### 1. Systematic Functionality Audit ```javascript // The scanner looks for these patterns you often debug: const functionalityPatterns = [ // Button handlers { pattern: /onClick=\{[^}]*undefined[^}]*\}/, type: 'undefined_handler' }, { pattern: /onClick=\{[^}]*\}/, check: 'handler_exists' }, // Routes { pattern: /href=["'][^"']+["']/, check: 'route_exists' }, { pattern: /navigate\(['"]([^'"]+)['"]\)/, check: 'route_exists' }, // Database operations { pattern: /\.create\(/, check: 'constraint_validation' }, { pattern: /\.findOne\(/, check: 'error_handling' } ]; ``` ### 2. Visual Debug Integration ```javascript // When you need visual debugging (like with visual-ui-debug-agent-mcp): class VisualDebugAgent { async debugUI(component) { // Take screenshot // Analyze layout issues // Check responsive behavior // Verify styling consistency // Test interactions visually } } ``` ### 3. Workflow Testing Pattern ```javascript // Your typical workflow testing approach: async function testWorkflow(workflowName) { const steps = [ 'Check initial state', 'Trigger action (click button)', 'Verify state change', 'Check database update', 'Verify UI feedback', 'Test error cases' ]; for (const step of steps) { await validateStep(step); } } ``` ## Example Debug Session (Your Style) ```bash $ claude-loop debug --focus workflows 🤖 Claude Loop - Debugging Repository 📋 Creating todo list based on your patterns: ⎿ ☐ Analyze handleCreateProduction and similar handlers ☐ Audit all button onClick implementations ☐ Check all Link hrefs and route definitions ☐ Test banco-vocabulario functionality ☐ Verify listas-vocabulario display ☐ Test language switching ☐ Fix database constraint issues 🔍 Scanning for issues... Found 15 undefined button handlers Found 8 broken routes Found 3 database constraint violations 🤖 Launching specialized agents... Agent #1 (workflows): Fixing button handlers Agent #2 (routes): Fixing navigation Agent #3 (database): Fixing constraints [Agents work autonomously to fix issues] ✅ Fixed Issues: ✓ handleCreateProduction now properly handles async operations ✓ All buttons now have working onClick handlers ✓ Routes /vocabulary/* now properly defined ✓ Database constraints satisfied for all operations ``` ## Integration with Your Tools ### Visual UI Debug Agent Pattern ```javascript // If visual-ui-debug-agent-mcp is available: if (hasVisualDebugger()) { await visualDebugger.screenshot(component); await visualDebugger.analyzeLayout(); await visualDebugger.testInteractions(); } ``` ### MCP Server Integration ```javascript // For your MCP server debugging: const mcpConfig = { "visual-ui-debug": { "command": "visual-ui-debug-agent-mcp", "args": ["--gateway", "cloudflare"], "env": { "BYPASS_PERMISSIONS": "true" } } }; ``` ## Your Common Debug Commands Based on your patterns, claude-loop would execute: ```bash # Your typical debugging flow: claude-loop scan --types workflows,routes,ui claude-loop debug --focus "button handlers" claude-loop debug --focus "broken routes" claude-loop debug --focus "database constraints" # With visual debugging: claude-loop debug --with-visual --mcp-server visual-ui-debug ``` ## Key Principles (From Your Approach) 1. **Systematic**: Test every button, every route, every workflow 2. **Functional Focus**: Make things work before optimizing 3. **Leverage Existing**: Use existing logic rather than rewriting 4. **Visual Verification**: See that UI components render correctly 5. **End-to-End**: Test complete workflows, not just units This aligns with your debugging style of systematically finding and fixing all broken functionality!