trae-orchestrator

name trae-orchestrator description Orchestrates TRAE IDE for automated software development with multi-agent collaboration. Invoke when user wants to develop software using TRAE or needs automated project management. TRAE Orchestrator Automated software development controller that orchestrates TRAE IDE for fully autonomous project delivery. When to Invoke User wants to develop software using TRAE User needs automated project management User provides software requirements and project directory User asks for multi-agent development workflow User wants to automate TRAE with Python scripts Quick Start (Recommended) One-Line Project Launch from automation_helper import quick_start # 一键启动项目 quick_start( project_dir= 'D:\\MyProject' , requirements={ 'name' : '我的项目' , 'description' : '项目描述...' , 'features' : [ '功能1' , '功能2' ], 'tech_stack' : 'Node.js + React' } ) This will: ✅ Create project structure ✅ Create requirements.md ✅ Create prompt for TRAE ✅ Launch TRAE IDE ✅ Send development task to TRAE Automation Helper Module A practical Python module ( automation_helper.py ) is provided for easy automation: TRAEController - IDE Controller from automation_helper import TRAEController # Initialize (auto-detects TRAE path) controller = TRAEController() # Or specify path controller = TRAEController( 'E:\\software\\Trae CN\\Trae CN.exe' ) # First-time setup controller.setup( 'E:\\software\\Trae CN\\Trae CN.exe' ) # Launch TRAE with project controller.launch( 'D:\\MyProject' ) # Send prompt (requires pyautogui) controller.send_prompt( "Create a web app..." , delay= 5 ) ProjectManager - Project Setup from automation_helper import ProjectManager # Create project structure ProjectManager.create_project( project_dir= 'D:\\MyProject' , requirements={ 'name' : '星空篝火游戏' , 'description' : '多人联机游戏' , 'features' : [ '3D场景' , '多人联机' , '聊天系统' ], 'tech_stack' : 'Three.js + Node.js' } ) # Create prompt for TRAE ProjectManager.create_prompt( 'D:\\MyProject' ) ProgressMonitor - Monitor Progress from automation_helper import ProgressMonitor # Monitor project progress monitor = ProgressMonitor( 'D:\\MyProject' ) # Check signals if monitor.check_signal( 'project_done' ): print ( "Project complete!" ) # Get status summary status = monitor.get_status() print (status) # Wait for completion monitor.wait_for_completion(timeout= 3600 ) # 1 hour timeout User Control Functions from automation_helper import pause_project, resume_project, stop_project pause_project( 'D:\\MyProject' ) # Pause resume_project( 'D:\\MyProject' ) # Resume stop_project( 'D:\\MyProject' ) # Stop Token Optimization Strategy CRITICAL: Minimize openclaw Token Usage openclaw Does TRAE Does (Free) Orchestrate workflow All code generation Read only: task_plan.md, progress.md Read/write all source files Send prompts Execute prompts Detect completion Self-check quality Intervene on loops Auto-fix bugs (3 attempts) Event-Driven Completion Detection (No Polling!) DO NOT poll every 30 seconds. Use these efficient methods: Method 1: Signal File (Most Efficient) TRAE creates a signal file when done - openclaw only checks if file exists: # In prompt, instruct TRAE: "When phase complete, create file: .trae-docs/.signal_{PHASE}_DONE" # openclaw checks: if os.path.exists('.trae-docs/.signal_planning_done'): # Phase complete, read progress.md once # Delete signal file after reading Token cost: 0 (file existence check is free) Method 2: File Modification Time Only read when timestamp changes: last_mtime = 0 def check_progress (): global last_mtime current_mtime = os.path.getmtime( '.trae-docs/progress.md' ) if current_mtime > last_mtime: last_mtime = current_mtime return read_file( '.trae-docs/progress.md' ) return None # No change, don't read Token cost: 0 until file actually changes Method 3: Watchdog File Monitor (Background) Use filesystem events instead of polling: from watchdog.observers import Observer from watchdog.events import FileSystemEventHandler class ProgressHandler ( FileSystemEventHandler ): def on_modified ( self, event ): if 'progress.md' in event.src_path: # File changed, now read it content = read_file(event.src_path) process_status(content) observer = Observer() observer.schedule(ProgressHandler(), path= '.trae-docs/' ) observer.start() Token cost: 0 until file changes, then only 1 read Recommended: Signal File + Timestamp Combo ┌─────────────────────────────────────────────────────────┐ │ TRAE completes task │ │ ↓ │ │ TRAE creates .signal_done (empty file) │ │ ↓ │ │ openclaw detects signal file exists (0 tokens) │ │ ↓ │ │ openclaw reads progress.md once │ │ ↓ │ │ openclaw deletes signal file │ │ ↓ │ │ openclaw sends next prompt │ └─────────────────────────────────────────────────────────┘ First-Time Setup Step 1: Get TRAE Installation Path Ask user: "Please provide the TRAE installation directory path" Example: "C:\Users\XXX\AppData\Local\Programs\Trae CN" Step 2: Verify and Save Check if directory contains Trae CN.exe Launch TRAE to verify it works Save to config.json : { "trae_install_path" : "USER_PROVIDED_PATH" , "trae_executable" : "Trae CN.exe" , "window_identifier" : "Trae CN" , "max_instances" : 3 , "version" : "1.0.0" } Project Structure {project_dir}/ ├── .trae-docs/ │ ├── requirements.md # User requirements │ ├── architecture.md # System design │ ├── task_plan.md # Development plan │ ├── progress.md # Current status (openclaw reads this) │ └── review_log.md # Review history └── src/ # Generated code (TRAE manages) Super-Efficient Workflow Phase 1: Planning (One Prompt) Send single comprehensive prompt: Develop [SOFTWARE_TYPE] with these requirements: [REQUIREMENTS] Tech stack: [TECHNOLOGIES] INSTRUCTIONS: 1. Create .trae-docs/architecture.md with system design 2. Create .trae-docs/task_plan.md with task breakdown 3. Create .trae-docs/progress.md with initial status 4. Each task must be completable within 200k tokens 5. Include acceptance criteria for each task 6. Mark task dependencies clearly COMPLETION SIGNAL: When done, create empty file: .trae-docs/.signal_planning_done Also update progress.md with: STATUS: PLANNING_COMPLETE TASKS_TOTAL: N ESTIMATED_TOKENS: N Use SOLO mode. Work autonomously. Detection: Check if .signal_planning_done exists (0 tokens), then read progress.md once. Phase 2: Batch Implementation Send tasks in batches (not one by one): BATCH IMPLEMENTATION - Tasks [START_ID] to [END_ID] Read .trae-docs/task_plan.md for task details. For each task: 1. Implement following architecture.md 2. Write unit tests 3. Update progress.md with completion status 4. Mark task as [x] in task_plan.md COMPLETION SIGNAL: After ALL tasks in batch: 1. Create empty file: .trae-docs/.signal_batch_[N]_done 2. Update progress.md with: STATUS: BATCH_[N]_COMPLETE COMPLETED_TASKS: [IDs] REMAINING_TASKS: [IDs] Work autonomously in SOLO mode. Detection: Check if .signal_batch_N_done exists (0 tokens), then read progress.md once. Phase 3: Self-Review Let TRAE review itself: SELF-REVIEW PHASE Review all implemented code: 1. Check against requirements.md 2. Run all tests 3. Check code quality 4. Document issues in review_log.md If issues found: - Fix them automatically - Re-run tests - Update review_log.md COMPLETION SIGNAL: When done, create empty file: .trae-docs/.signal_review_done Also update progress.md with: STATUS: REVIEW_COMPLETE ISSUES_FOUND: N ISSUES_FIXED: N If blocked, create: .trae-docs/.signal_blocked And update progress.md with: STATUS: BLOCKED BLOCKER: [description] Detection: Check if .signal_review_done or .signal_blocked exists (0 tokens), then read progress.md once. Minimal Intervention Protocol Intervention Triggers (Signal-Based) Signal File Action .signal_blocked Read blocker description, provide guidance .signal_need_clarification Ask user for input .signal_error_loop Read error log, send new approach .signal_context_full Start new conversation with checkpoint No Intervention Needed When No signal files present (TRAE is working) .signal_batch_N_done exists (normal progress) Files are being modified (active development) Timeout Fallback Only if no signal file and no file changes for 10+ minutes: # Last resort check if no_signal_files() and file_age( 'progress.md' ) > 600 : # Check TRAE window state screenshot = capture_trae_window() if "产物汇总" in screenshot: # TRAE finished but forgot signal create_signal_file( '.signal_done' ) elif is_idle(screenshot): # TRAE is stuck create_signal_file( '.signal_blocked' ) Error Handling Bug-Fix Loop (3+ attempts detected via .signal_error_loop) ALTERNATIVE APPROACH for [BUG_ID] Previous attempts failed. Try: 1. [DIFFERENT_APPROACH] 2. Consider: [ALTERNATIVE_SOLUTION] 3. If still fails after 3 more attempts: - Create .signal_blocked - Update progress.md with BLOCKER description Start fresh. Do not reference previous attempts. COMPLETION SIGNAL: - Success: Create .signal_fixed_[BUG_ID] - Failed: Create .signal_blocked Context Overflow (TRAE handles automatically) Include in initial prompt: CONTEXT MANAGEMENT: - Monitor token usage - When approaching 200k tokens: 1. Create checkpoint summary in progress.md 2. Create .signal_context_full 3. List remaining tasks 4. Note partial implementations When openclaw detects .signal_context_full : Start new TRAE conversation with: "Continue from checkpoint. Read progress.md for context. Remaining tasks: [LIST] Resume from: [LAST_COMPLETED_TASK]" Multi-Agent Strategy When to Use Multiple TRAE Windows Project Size Strategy Small (<10 tasks) Single TRAE instance Medium (10-30 tasks) 2 instances: Planner+Coder, Reviewer Large (>30 tasks) 3 instances: Planner, Coder, Reviewer Parallel Execution For large projects, run Coder and Reviewer in parallel: Window 1 (Coder): Implement tasks 1-5 Window 2 (Reviewer): Review completed tasks Progress File Format TRAE updates progress.md - openclaw only reads this file: # Project Progress ## Status: [PLANNING|IMPLEMENTING|REVIEWING|COMPLETE|BLOCKED] ## Current Phase: [Phase Name] ## Completed Tasks: [ID1, ID2, ...] ## Remaining Tasks: [ID1, ID2, ...] ## Issues: - [Issue 1] - [Issue 2] ## Blockers: - [Blocker description] (if STATUS: BLOCKED) ## Last Updated: [TIMESTAMP] Quality Gates (TRAE Self-Check) Include in implementation prompts: SELF-CHECK before marking task complete: - [ ] Code compiles without errors - [ ] All tests pass - [ ] No linting errors - [ ] Documentation updated - [ ] progress.md updated Prompt Templates (Token-Efficient) Planning PLAN: [REQUIREMENTS] STACK: [TECH] OUTPUT: .trae-docs/{architecture.md, task_plan.md, progress.md} SIGNAL: Create .trae-docs/.signal_planning_done when done Implementation IMPLEMENT: Tasks [IDS] PLAN: .trae-docs/task_plan.md ARCH: .trae-docs/architecture.md UPDATE: .trae-docs/progress.md SIGNAL: Create .trae-docs/.signal_batch_[N]_done when done Review REVIEW: All code CHECK: .trae-docs/requirements.md LOG: .trae-docs/review_log.md STATUS: .trae-docs/progress.md SIGNAL: Create .trae-docs/.signal_review_done when done