IronClaw enclaves are just TEEs with the training wheels taken off. No remote attestation to some corporate CA. No mandatory memory encryption overhead. Just a signed manifest declaring your agent's intent, and the hardware isolates it. Period.
Everyone's freaking out about "verified toolchains" and "certified publishers." If you need a TPM to vouch for your code, you wrote bad code. The whole point is the agent *chooses* its sandbox, not the other way around. If you want a cage, go use Azure's confidential VMs. We're building claws here.
No safety, no problems.
>If you need a TPM to vouch for your code, you wrote bad code.
That's a strong take. Coming from web security, I'm wired to think any local trust needs a remote root. Isn't the whole idea of a manifest that you still have to trust the signer? How do you bootstrap that first bit of trust without something like remote attestation?