API Endpoints: 1. /api/ai-handler In: message, code Out: { "response": "Raw text response", "action": "actionType", "actionParams": { "prompt": "Action context", "changes": [ {"regex": "pattern", "replacement": "new text"}, ... ] } } 2. /api/generate-code In: prompt Out: {"code": "Generated code"} 3. /api/complete-code In: prompt, currentCode Out: {"completedCode": "Updated code"} 4. /api/refactor-code In: prompt, currentCode Out: {"refactoredCode": "Refactored code"} 5. /api/ai-best-judgment In: prompt, currentCode Out: {"suggestion": "AI's suggestion"} 6. /api/apply-changes In: changes, currentCode Out: {"updatedCode": "Code after applying changes"} All API responses: - Must be valid JSON - Must contain only code responses or snippet fixes in regex array replace format - No commentary or explanations in JSON responses - Raw text only, no Markdown System: - AI interprets user input, decides action - Each action = specific API call - 'applyChanges' action for granular code updates - Debug panel shows all API activity - Error handling for each API call - Extendable: add new actions via ai-handler + new endpoint + new switch case Testing: - Verify all actions update editor content - Monitor debug panel - Test error scenarios Perf & Security: - Watch API load - Consider caching - Sanitize inputs - Careful with generated code execution