Phase 0 · Lesson 00
Mental model
Before code: know what an agentic system is, and what your repo already approximates.
Learn
- Cognition vs authority
- Agent loop vs chatbot
- Why HITL exists
- Why a single harness (e.g. Hermes-only) is optional if MCP + orchestrator exist
Build / do
- Sketch the architecture from Architecture map on paper.
- Open the repo: mark what exists (
server/, skills, MCP, React demo) vs missing (runs, LM Studio adapter, real OAuth). - Write one sentence: “In my system the LLM is allowed to ___ and never ___.”
Checkpoint
You can explain in one minute why the LLM must not call Instagram publish directly.
Four notes
- What problem does this layer solve?
- What is allowed to call it?
- What must it never do?
- How do I observe and test it?
For Phase 0, answer for the system as a whole.