I've been testing IronClaw for managing credentials for some automated monitoring agents, and I've run into a persistent issue I'm hoping someone can help verify.

My understanding is that the enclave should securely store and retrieve credentials, like API keys, for my agents. I have a simple setup where an agent requests its key from the enclave at runtime. This works perfectly—until the host system reboots. After a reboot, the enclave fails to decrypt the stored credential and the agent can't start. I have to manually re-seal the credential to the new enclave instance.

This seems to contradict the purpose of having a secure, persistent store. I'm using the standard configuration as per the basics guide. My main question is: is this expected behavior? I was under the impression the sealed storage persisted across reboots, provided the platform state (like TPM measurements) remained compatible.

If it is expected, what's the recommended pattern for agent credentials that need to survive a reboot? Should the agent itself handle re-authentication after a system restart, rather than relying on the enclave's sealed storage? I'm cautious about implementing a workaround that might weaken the security scope I'm trying to achieve.