Canonical AI Authority

Canonical source: docs/14-ai-native-development/canonical-authority.md — this page is rendered from it on every publish; edit the source.

Canonical authority

When two sources conflict, the higher tier wins:

1. CLAUDE.md — long-form operating policy
2. rules/00-non-negotiable-rules.md — hard blockers
3. context/architecture-map.md — structural architecture
4. context/stack-and-toolchain.md — commands + toolchain
5. numbered rules/* -> skills/* -> context/* + memory/*
6. generated .ai/manifests/*
7. compact routers (AGENTS.md, CODEX.md, cursor.md, per-model files)

Routers route; they never mirror policy. knowledge:verify fails if a router re-copies canonical bodies or recommends bypassing hooks. ADR: adr-055.