Exactly! Coming from a system architecture background (I built Linux deepin), I’ve always viewed 'context sharing' as a challenge similar to shared memory in multi-threading. As you know in the AI security space, without strict boundaries, 'context bleeding' isn't just a bug—it’s a security risk. That’s why I emphasized the plumbing; if the filesystem and state isolation aren't deterministic, the 'side effects' become unpredictable. Thanks for the insightful comment！