research(policy): governance policies for AI coding agents touching infra #360

Tracked on the public feedback board (idea: govern third-party AI coding agents touching infra).
Governing what third-party AI agents may do to infrastructure is a policy question over documents
Tirith can already read through the json provider — agent/MCP configs, permissions manifests,
audit logs — so the research is the input shape and a starter pack, not an engine change.