AGT-08 · Agent governance & permissioning¶
Agents often run under a shared service principal with broad roles, so nobody can say what the agent may do until after it has done something wrong. Permission changes happen by ticket without review, and there is no tested kill switch when an agent acts outside its scope.
- Who reads it: Agent leads, Responsible AI officers, security engineers and DevOps engineers who own agent identities
- Outcome: Every agent has a written, enforced and tested permission envelope
- Pack owner: Prasanna (catalog owner role: Responsible AI Officer)
Master catalog entry¶
- Catalog name: Agent Governance & Permissioning Framework (P1)
- Why it matters: Answers the board-level question of what the agent is allowed to do
- What it removes: Blast-radius controls defined before build, not after an incident
- Stack: Entra ID, managed identity, policy engine, action allow-lists
- Maturity target: Level 2 — scope, spend, action and data-access boundaries
Playbooks¶
| ID | Story | Title | Kind | Complexity | Urgency | Status | Owner | Planned publish |
|---|---|---|---|---|---|---|---|---|
| PB-AGT-08-01 | AGT-08-S1 | Define an agent's permission envelope: scope, spend, actions and data | Decision guide | Medium | 3 | draft | Prasanna | 2026-11-27 |
| PB-AGT-08-02 | AGT-08-S2 | Enforce action allow-lists and approval gates in agent code | How-to | Medium | 4 | draft | Devadharshini Raja | 2026-11-25 |
| PB-AGT-08-03 | AGT-08-S3 | Change an agent's permissions safely | Runbook | Medium | 4 | draft | Prasanna | 2026-11-13 |
| PB-AGT-08-04 | AGT-08-S4 | Respond to an agent acting outside its permission envelope | Runbook | High | 5 | draft | Prasanna | 2026-11-11 |
| PB-AGT-08-05 | AGT-08-S5 | Agent governance go-live review checklist | Checklist | Low | 3 | draft | Devadharshini Raja | 2026-12-02 |