Hacker News new | ask | show | jobs
by xms17189 43 days ago
Curious how you handle trust boundaries for tool outputs here. Do you keep a signed or replayable trace so a developer can audit what the agent saw before it acted?