Coach AI

Instructions

Implementation Reference

• src/coach_ai/server.py:501-544 (handler)

  The `break_down_task` function, decorated with `@mcp.tool()`, implements the tool handler. It fetches the task from the database and uses `break_down_task_with_claude` to generate a prompt for task decomposition.

  async def break_down_task(todo_id: int, subtask_count: int = 5) -> str:
      """Use LLM to decompose large task into smaller subtasks.
  
      This returns a prompt for Claude to generate subtasks. After Claude responds,
      call this tool again with the response to create the subtasks in the database.
  
      Workflow:
      1. Call break_down_task(todo_id=X) -> Returns prompt
      2. Send prompt to Claude
      3. Claude returns JSON with subtasks
      4. Subtasks are automatically created in database
  
      Args:
          todo_id: ID of the task to break down
          subtask_count: Target number of subtasks (default: 5)
  
      Returns:
          Prompt for Claude to generate subtasks, or confirmation if creating them
      """
      db = await storage.get_db()
  
      # Get todo
      cursor = await db.execute(
          """
          SELECT id, title, priority, notes, timeframe, theme_tag, time_estimate
          FROM todos WHERE id = ?
          """,
          (todo_id,),
      )
      row = await cursor.fetchone()
      if not row:
          return f"Error: Todo #{todo_id} not found"
  
      todo = dict(row)
  
      # Generate prompt for Claude
      breakdown = break_down_task_with_claude(todo, subtask_count)
  
      response = f"**Breaking down task #{todo_id}: {todo['title']}**\n\n"
      response += "I'll help break this down into smaller steps. Here's what I recommend:\n\n"
      response += breakdown['prompt']
      response += "\n\n*Note: This is a prompt for planning. The subtasks will be created automatically based on the breakdown.*"
  
      return response

• src/coach_ai/task_breakdown.py:14-30 (helper)

  The `break_down_task_with_claude` helper function generates the prompt for an LLM to decompose a task into subtasks.

  async def break_down_task_with_claude(
      todo: dict[str, Any],
      subtask_count: int = 5,
  ) -> dict[str, Any]:
      """
      Use Claude (via prompt) to break down a large task into smaller subtasks
  
      This function returns a prompt that should be sent to Claude.
      The caller (MCP tool) will handle the actual LLM call.
  
      Args:
          todo: The todo dict to break down
          subtask_count: Target number of subtasks
  
      Returns:
          Dict with prompt for LLM
      """