Something I've never understood about TPM attestation, is what happens if you plug the TPM into a microcontroller and give it all the same measurements that it would normally receive during a normal boot? Would that let you spoof attestations?
You can create a key or similar attribute which has an unlock policy based on those PCR values. If you play back the log of PCR write events from first principles (the log can be captured for debug purposes), you'll put the TPM into the same state and should be able to use anything protected by the respective policy.
For attestation, I presume you're thinking about sending an attested PCR quote - in that case, the TPM uses a non-extractable key to sign the current PCR states. As you can put the PCRs into the "correct" state, you'd be able to get a signed attestation the system is in that state.
An encrypted external drive though works like a safe. Put things in there you want to keep safe but don’t need every day. Air gapped while not in use makes it even more safe.
Some modern CPUs have moved the TPM inside the CPU itself. But traditionally, TPMs were attached via the LPC (low pin-count) bus, and you could absolutely sniff them or de-solder them and arbitrarily MiTM.