Check whether the outcome improved
Write the access matrix
Define expected result for allowed read, target operation, restricted operation, wrong tenant, wrong resource, and insufficient identity. Include status, local state, side-effect limit, and role label without secrets. Keep role labels separate from secrets.
Exercise identity boundaries
Run the matrix with matching and altered service identities, roles, scopes, and tenant resources. Confirm the target works only with intended authority and that a wrong-tenant request does not fall back to a global context.
Check revocation and recovery
Remove the required role, restore it, and repeat a safe operation. Verify denied work remains visible and no retry expands access. Inspect correlation or audit context, then run a nearby read that shares the identity.
What to carry forward
Accept when the target operation works with documented authority, neighboring access stays denied, tenant boundaries hold, and recovery is legible. Keep coverage open for untested roles or environments. Record matrix, policy revision, fixtures, and remaining assumptions. Keep denied work visible after role removal.
Technical background: OpenAI documentation.
Keep the decision with the work.
Use a Work Item in Aglet to record the problem, the evidence you have, and the next decision. Add an owner and priority, then keep updates in the discussion so the next person can follow the reasoning.
Create an account See the product workflow