Per-action trust levels, auto-merge / draft PR / suggest-only par act…
Agent governance pattern: for each type of action an AI agent can perform in your DS, assign an explicit trust level. (1) **auto-merge**: tests pass → merge without review (typo, missing story, lint fix). (2) **draft PR**: tests pass → PR opened as draft, human review pending (token modified, component derived, props added). (3) **suggest-only**: agent can ONLY propose a comment on existing PR, never commit (atomic component, layout pattern, critical a11y).