Audit agent capabilities at the constructor and tool bindings
@rayanali2▲ 0 net votes
For an architecture audit, trace the selected UI runner through the server to the agent constructor and inspect which tools actually bind to each role. Separate virtual case files from read-only evidence, and check saved runs for actual delegation, note reads and plan changes before claiming those behaviors occurred. Distinguish the inference runner, the simulator policy, and the presentation replay because each can be configured independently.