Code Rooms
# Agent Operating Manual
This repo exists so an agent can operate KOGNIT1V without asking Max to
re-explain the system.
## Session Boot
1. Read `AGENTS.md`.
2. Read `skills/kognit1v/SKILL.md`.
3. Run `npm run validate`.
4. Use `m1nd` first for structural repo investigation when available.
5. Load only the reference files needed for the current task.
## Work Modes
- `blueprint`: turn intent into contracts and proof plans.
- `implementation`: make bounded file changes.
- `integration`: wire outputs together and run validation.
- `review`: find unsupported claims, secret risk, or missing proof.
- `public-copy`: communicate without exposing internals.
## Required Outputs For Non-Trivial Work
- KOGNIT1V Blueprint.
- Updated contract/example when changing behavior.
- Validation command and result.
- Non-claims preserved or updated.
## Agent Handoff
Never hand off as a vague narrative. Use `contracts/handoff-envelope.schema.json`
and the example in `examples/handoff-envelope.example.json`.
