🎓 Agent Architecture Learning Checklist

Use this list to track your progress in understanding the codebase.

🟢 Module 1: The Foundation (Philosophy)

• [ ] Concept: Read "The Tool-Agnostic Paradigm" in docs/dev/agent_curriculum.md.
• [ ] Concept: Understand why we use MCP (Model Context Protocol).
• [ ] Self-Check: Explain to yourself (or a rubber duck) why we simply don't hardcode tool API calls in the prompt.

🧠 Module 2: Orchestration (The Brain)

• [ ] Read: "The Nervous System" in the curriculum.
• [ ] Code Action: Open backend/app/services/agent/agent_orchestrator.py.
• [ ] Code Action: Locate the StateGraph definition (~line 300).
• [ ] Code Action: Find the route_tools function and see how it checks for requires_approval.
• [ ] Self-Check: Which node runs first? agent or tools?

🛠️ Module 3: Dynamic Tooling (The Body)

• [ ] Read: "The Body" (Dynamic Assembly) in the curriculum.
• [ ] Code Action: Open backend/app/services/agent/agent_factory.py.
• [ ] Code Action: Find create_final_agent_pipeline.
• [ ] Code Action: Open backend/app/services/agent/tool_registry.py and see the BM25Okapi usage.
• [ ] Self-Check: How does the agent "know" about a new tool you just added to an MCP server?

🛡️ Module 4: Safety & Persistence (The Conscience)

• [ ] Read: "The Conscience" in the curriculum.
• [ ] Code Action: Look at human_review_node in agent_orchestrator.py.
• [ ] Code Action: Find where PendingApproval calls happen.
• [ ] Self-Check: If the server crashes while waiting for user approval, where is the state saved? (Hint: Redis Checkpointer).

🏁 Final Verification

• [ ] Interactive: Run the agent in the UI.
• [ ] Trace: Follow the logs in your terminal as the agent thinks and acts.
• [ ] Complete: Mark this checklist as done!