Agent, crew, and tool problems

1. Confirm the active agent or crew and its published configuration.
2. Check that the required model tier, skill, tool, plugin, or connection is available to the account.
3. Review the task’s autonomy mode and any waiting approval card.
4. Run one harmless tool action outside the full workflow.
5. Read the first failed tool card, including target, status, and error, before retrying.

A model response is not proof that device tools are connected, and a healthy tool catalog is not proof that the selected agent may use it. For a crew, test the failing member directly before changing routing for every member.

Expected result: the isolated action succeeds or identifies the exact entitlement, permission, configuration, or tool boundary.