An LLM-driven agent orchestrator backend built with Node.js, TypeScript, and Fastify.

• src/api/routes/: API endpoint definitions (Fastify plugins).
• src/core/: Core logic including the Agent (orchestrator) and TaskEngine (scheduler/executor).
• src/plugins/: Modular tool implementations (e.g., GitHub, Shell, etc.).
• src/types/: Shared TypeScript interfaces and types.
• src/utils/: Utility modules like llmService (OpenAI integration).
• src/index.ts: Application entry point.

• POST /task/create: Create a task from a natural language prompt.
• GET /task/list: List all tasks and their status.
• GET /task/status/:id: Get detailed status and logs for a specific task.
• POST /task/approve/:id: Approve a task for execution.

1. Prompt: User sends a natural language prompt to /task/create.
2. Spec Generation: LLM (OpenAI) generates a structured TaskSpec (steps, tools, args).
3. Pending Approval: Task is created with status pending_approval.
4. Approval: Human calls /task/approve/:id.
5. Execution: TaskEngine executes steps sequentially using registered plugins.
6. Logging: Real-time logs are appended to the task object.

Tools are defined by the ITool interface and registered in the TaskEngine. Default tools:

• shell: Executes local commands via child_process.