Skip to content

Design: agent role authority model (recommend vs execute, guardrails) #3

Description

@gidutz

Each agent needs a declared authority scope — what it can recommend vs. execute, and under what guardrails. Real money is moving, so this is load-bearing.

Key questions:

  • Schema for authority: per-action, per-budget-band, per-entity-type?
  • Default policies per role (media buyer, campaign-setup, AdOps)
  • How are authorities composed when one agent calls another?
  • Override / approval flow when an agent wants to act outside scope
  • How is authority versioned and audited?

Deliverable: authority schema plus a worked example for each of the three roles.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    pillar:rolesAgent authority scope and guardrailstype:designDesign discussion

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions