llm-wiki-governance

Governance for AI-written project docs. llm-wiki-governance is a zero-dependency CLI that keeps an AI coding agent's project knowledge (docs/llm-wiki/) trustworthy and current: it ties every claim to real code, flags docs when that code moves on, keeps AI-written content behind human review, and enforces all of it in CI. Works on any stack, with any agent, and is OKF-compatible.

Why a governance layer?

An LLM wiki — a distilled, interlinked knowledge base an agent reads instead of re-deriving your codebase on every task — is a proven pattern (popularized in 2026, and formalized by Google's Open Knowledge Format). The hard part was never making one. It's keeping it honest.

What "governed" means here:

• Trust states — AI-written docs stay needs_review; only a human promotes to verified. The CLI can never self-approve.
• Evidence + drift — each claim links to a real file/line/symbol; when that source changes, evidence.stale / drift flags the doc for re-review.
• CI-enforced — validate runs in pre-commit / GitHub Actions, so an unreviewed or drifted wiki fails the build instead of rotting quietly.
• Agent-queryable — a read-only MCP server lets agents ask the wiki instead of re-scanning the code.
• Safe by construction — preview-first writes, append-only change log, sensitive-value redaction, and zero runtime dependencies.