The AI governance plugin, governed

Anthropic's ai-governance-legal plugin, part of the May 2026 Claude for Legal launch, industrialises the production of governance artefacts: usage policies, EU AI Act inventories, use-case triage, impact assessments, vendor AI reviews and drift reports. This guide treats it as what it is, a serious authoring surface with a clearly declared boundary, and shows how to get real governance value out of it. It covers the nine skills and what each produces, the five setup moves that decide output quality (the cold-start interview and a live research connector chief among them), an honest two-band map of where the plugin performs and where it stops, and which of its eleven sibling plugins to run alongside it, including the one whose install gate is worth studying rather than merely using. The closing section is the reason the guide exists: a wiring table that takes each of the six outputs from Level 4, Complete, where the artefact exists and is maintained, to Level 5, Substantive, where something outside the authoring process refuses to proceed without it. Seven pages, A4, sourced from the public repository as read on 17 August 2026.