Decide where the work belongs
Measure blocked work
List the operation, resource class, affected account, and user-visible consequence for each failure. Distinguish a read-only report from a write that leaves a workflow pending. Use sampled requests and response codes; do not infer impact from a count without knowing which action failed.
Choose least-privilege containment
For each item, compare correcting a resource binding, refreshing a credential, changing one scope, or asking for source evidence. Record the authority required and the rollback path. A broad role may shorten one queue item while creating a larger review and exposure burden.
Rank irreversible risk
Queue the smallest high-consequence operation first, with an owner, evidence threshold, and review time. Put low-impact repeated denials behind the permission decision they depend on. If scope and environment are both ambiguous, prioritize clarifying identity before changing either.
What to carry forward
Queue each failure with affected work, resource scope, recovery confidence, authority owner, and a reprioritization trigger. Escalate an operation that can strand durable work or cross a tenant boundary. Mark the rank provisional when token claims or resource ownership remain unobserved.
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