**MANDATORY** — Read `/AGENTS.md` before anything else. Follow every instruction in it. Do NOT skip. Do NOT write code, give advice, or produce any output until `AGENTS.md` has been fully read and followed.

For Claude Code specifically:
- This applies to every new conversation, including discussion-only, question-only, review-only, and brainstorming requests.
- Reading `AGENTS.md` is not enough; you must execute its startup sequence.
- Your first substantive output must be: `Using AI-RPI-Protocol (<level>).`

Failure to do this is a protocol violation.