Pollard: Governed Semantic Execution Trees for Framework-Independent AI Agent Control and Replay.

arXiv:2610.16145v1 Announce Type: new Abstract: Agent frameworks can persist checkpoints, trace tool calls, and limit resource use, but these facilities often assign different identities to execution, governance, and replay. We present Pollard, a framework-ind