what's next

Nothing listed in this lane yet.

1. Per-policy fail-open / fail-closed configuration

   There is no per-policy choice today. The only fail-open control is one process-wide environment variable, and it does not change approval evaluation at all.

2. Maintenance-window policy primitive

   A deploy window should be a policy that closes itself when the window shuts, rather than one that relies on someone remembering.

3. TS / JS / Go runtime SDKs

   The wire contract is already published for other languages. What is missing is the runtime around it: reservation, tracking, retries.

4. Agents keep approvals through crashes and restarts

   A restart should not erase an approval. A granted action should outlive the process that granted it.

5. Accurate cost when the same model has different prices per provider

   The same model can cost different amounts at different providers. The bill should follow the provider, not the model name.

6. 30-second rollback on platform incidents

   When a policy change misfires, the operator needs an undo that is faster than the deploy that caused it.

7. Operator can cancel an approval they've already given

   An approver who changes their mind should still be able to say so before the action runs.

8. See whether your agent is still running

   What did not happen and what is happening now are two different questions. On a list, a workflow that stopped an hour ago looks the same as one that is still burning tokens.

9. Per-region pricing support

   One price per model does not survive a deployment that spans more than one region.