Coach AI

Instructions

Implementation Reference

• src/coach_ai/recommendations.py:6-106 (handler)

  The core logic for generating recommendation context based on database records.

  async def get_recommendation() -> str:
      """Get a personalized recommendation for what to do next.
  
      Analyzes the user's current todos, goals, recent activity, and known facts
      to suggest the most appropriate next action.
  
      Returns:
          Context string for AI to generate recommendation from
      """
      db = await get_db()
  
      # Get active todos
      todos_cursor = await db.execute(
          """
          SELECT id, title, priority, notes
          FROM todos
          WHERE status = 'active'
          ORDER BY
              CASE priority
                  WHEN 'high' THEN 1
                  WHEN 'medium' THEN 2
                  WHEN 'low' THEN 3
              END,
              created_at ASC
          """
      )
      todos = await todos_cursor.fetchall()
  
      # Get goals
      goals_cursor = await db.execute(
          "SELECT goal, timeframe, category FROM goals WHERE status = 'active' ORDER BY created_at DESC"
      )
      goals = await goals_cursor.fetchall()
  
      # Get user facts
      facts_cursor = await db.execute(
          "SELECT fact, category FROM user_facts ORDER BY created_at DESC LIMIT 10"
      )
      facts = await facts_cursor.fetchall()
  
      # Get recent accomplishments
      accomplishments_cursor = await db.execute(
          "SELECT description, created_at FROM accomplishments ORDER BY created_at DESC LIMIT 5"
      )
      accomplishments = await accomplishments_cursor.fetchall()
  
      # Build comprehensive context
      context = "=== CURRENT STATE FOR RECOMMENDATION ===\n\n"
  
      # Todos section
      if todos:
          context += "ACTIVE TODOS:\n"
          for i, todo in enumerate(todos[:10], 1):  # Limit to top 10
              context += f"  {i}. [{todo['id']}] {todo['title']} (priority: {todo['priority']})\n"
              if todo["notes"]:
                  context += f"      Notes: {todo['notes']}\n"
      else:
          context += "ACTIVE TODOS: None\n"
  
      context += "\n"
  
      # Goals section
      if goals:
          context += "ACTIVE GOALS:\n"
          for goal in goals:
              context += f"  - {goal['goal']} ({goal['timeframe']}, {goal['category']})\n"
      else:
          context += "ACTIVE GOALS: None set yet\n"
  
      context += "\n"
  
      # User context section
      if facts:
          context += "KNOWN ABOUT USER:\n"
          for fact in facts:
              context += f"  - {fact['fact']} ({fact['category']})\n"
      else:
          context += "KNOWN ABOUT USER: Learning about you as we go\n"
  
      context += "\n"
  
      # Recent wins section
      if accomplishments:
          context += "RECENT ACCOMPLISHMENTS:\n"
          for acc in accomplishments:
              context += f"  - {acc['description']}\n"
  
      context += "\n"
      context += "=== RECOMMENDATION REQUEST ===\n\n"
      context += (
          "Based on the above context, provide a specific, actionable recommendation for what "
          "the user should focus on RIGHT NOW. Consider:\n"
          "- Their priorities and goals\n"
          "- Known patterns and preferences\n"
          "- ADHD considerations (decision paralysis, activation energy, time blindness)\n"
          "- The need for clear, concrete next steps\n"
          "- Breaking down overwhelming tasks\n\n"
          "If there are no active todos, encourage the user to add some or reflect on their goals."
      )
  
      return context

• src/coach_ai/server.py:619-627 (registration)

  MCP tool registration for get_recommendation, which calls the recommendation engine implementation.

  @mcp.tool()
  async def get_recommendation() -> str:
      """Get a personalized recommendation for what to do next.
  
      This analyzes the user's current todos, goals, recent activity, and known facts
      to suggest the most appropriate next action. This is the core "What should I do now?"
      feature designed to help with decision paralysis.
      """
      return await recommendations.get_recommendation()