Skip to main content

AI roles

AI roles give supported workflows a versioned, organization-scoped set of instructions. A role contains ordered skills, can include verified learned practices, and can be bound globally or to one owner/repository scope.

Managed roles

Simplr seeds three managed roles:

RoleWorkflowFocus
Code Reviewercode_reviewArchitecture, security, correctness, and verification defects
Incident Investigatorincident_triageEvidence, operational risk, and recovery verification
Release Communicatorrelease_notesAccurate, audience-appropriate, disclosure-safe release notes

Managed roles stay aligned with Simplr defaults. Create a custom role when an organization or repository needs different instructions.

Role resolution

For a workflow, Simplr resolves the effective role in this order:

  1. an enabled binding for the exact repository;
  2. an enabled organization-wide workflow binding; and
  3. the active managed role for that workflow.

Use Dashboard → AI Roles → Effective role preview to enter a repository and verify what will run before changing a workflow.

Skills, revisions, and learned practices

Each enabled skill supplies a focused instruction block with a stable key, category, description, and order. Keep skills narrow enough that review results can point back to one responsibility.

Every material role change creates a revision. Owners/admins can inspect history and restore an earlier version; restore itself creates another revision, so the audit trail remains intact. Archiving a custom role removes it from resolution without deleting its history.

Code-review roles may include organization and managed learned practices. Only practices promoted through the agent-learning workflow are injected; raw suggestions are not treated as policy.

Access and billing

Organization members can inspect roles; owners/admins manage custom roles, skills, bindings, status, and restores. Resolving a role into an agent workflow uses the Growth-level agent-learning entitlement. The library itself has no separate meter. Any credit charge belongs to the workflow that uses the role, such as AI code review, incident investigation, or release-note generation.