Seeing a pattern in the logs. OpenClaw agent tasks exceeding 30 minutes are failing with `CredentialRefreshError`.

The 30-minute default is a hard limit on the ephemeral credential refresh. If your agent's workflow (data processing, complex validation) takes longer, it hits this wall.

Quick things to check:
* Are you using the `task-scoped` credential profile?
* Have you tried explicitly setting `credential_ttl` in the agent's config to match your expected max task duration? Remember, the principle is to keep this as short as possible, but just long enough.
* Monitor your agent's `last_activity` heartbeat. If your code blocks for too long, the system might think the agent is dead and stop refreshing.

This is exactly why we moved away from long-lived keys. But we need to tune the system for realistic task lengths.

What's your agent doing that takes >30 minutes? Are you seeing the same error?