Reference
Full syllabus
Eighteen lessons. One concept per page. Build the checkpoint before advancing.
Week plan
| Week | Lessons | Focus |
|---|---|---|
| 0–1 | 00–04 | Brain, skill, loop, runs |
| 2 | 05–08 | Orchestrator, policy, HITL, API |
| 3 | 09–11 | React truth, monitor, MCP |
| 4 | 12–15 | Memory, media, OAuth, publish |
| 5 | 16–17 | Evals + hardening |
Lesson list
- Mental model — cognition vs authority
- Model adapter — LM Studio / OpenAI-compatible
- Structured skill — schema contracts
- Agent loop — ReAct lite + budgets
- Durable runs — observability foundation
- Orchestrator — domain status machine
- Policy gate — safety core
- HITL approvals — approve / revise
- REST interface — stable HTTP contracts
- React truth UI — no demo data
- Live monitoring — SSE / poll runs
- MCP harness — any IDE client
- Memory layers — brand + learn
- Skills & media — free stock pipeline
- Instagram OAuth — compliant connect
- Gated publish — idempotent Graph publish
- Evals suite — golden + policy tests
- Hardening — threat model + ops
Definition of done
- Budgeted agent loop
- LM Studio via OpenAI-compatible adapter
- Schema’d skills
- Orchestrator + status machine
- Policy gate on side effects
- HITL approve / revise
- MCP + React on the same truth
- Live run monitoring
- Brand memory injection
- Real media + gated Instagram publish
- Evals + redacted traces
- Documented threat / compliance notes