Agent Engineering Capstone

1. Pick a project

1. Week 1Agent-ready repo

   Starter kit →
   • AGENTS.md under 80 lines with real commands, plus CLAUDE.md or tool pointers
   • One scoped rule and one spec in docs/specs/
   • Before/after note: the same task attempted with and without AGENTS.md

2. Week 2Guardrails and review

   Test-gate hook lab →
   • A hook that blocks one real bad action (commits with stale tests, secret reads)
   • A read-only reviewer subagent and one skill for a procedure you repeat
   • A PR template with an agent disclosure section

3. Week 3Build the agent

   Build your own agents →
   • Your capstone agent running end to end on one real input
   • Least-privilege tools, a turn limit, and a cost cap
   • Structured output (JSON) that a script or CI step consumes

4. Week 4Evals and CI

   Skill eval lab →
   • 10–20 eval cases with expected outcomes, run in CI
   • A tracked metric: pass rate, cost per run, and median latency
   • One failure you found with evals and how you fixed it

5. Week 5Ship and tell the story

   Shipping chapter →
   • README with an architecture diagram, threat model, and cost numbers
   • A 3-minute demo video or GIF
   • A written retrospective: what you would change, and what the agent still gets wrong