Code Rooms
# KOGNIT1V Agent Contract
This repo is agent-first. Do not treat it like a human notes folder or a product
monorepo.
## Required First Steps
Before acting on any request inside this repo:
1. Read `docs/AGENT_OPERATING_MANUAL.md`.
2. Read `skills/kognit1v/SKILL.md`.
3. Run `npm run validate` unless the request is a trivial text question.
4. If m1nd MCP is available, use it before broad file search for structural
investigation.
## Global Rules
- No secrets. Use env names only, never values.
- No claim without proof. A generated artifact is not a proven capability.
- No external mutation without an apply gate, idempotency key, and explicit
human authorization.
- No public copy may expose internals such as proofs, gates, queues, dry-runs,
checkpoints, schedulers, runtime mechanics, or credentials.
- Handoffs must be structured artifacts, not chat-memory summaries.
- If capability is missing, name it as a gap. Do not fake it.
- Do not add private product/project doctrine here. This repo carries cognitive
skills only.
## Operating Loop
```text
intent
-> KOGNIT1V Blueprint
-> capability gaps
-> contracts / SSOT
-> proof plan
-> dry-run / local proof
-> handoff envelope
-> apply gate if mutation is needed
-> checkpoint only after evidence
```
## Finish Line
Work is complete only when:
- changed files are explicit;
- contracts/examples still validate;
- no secret-like values are present;
- non-claims are preserved;
- the next agent can resume from files, not from memory.
