An audit log is a running record of every action an agent took, each entry stamped with the time and the inputs that produced it, so you can reconstruct what happened after a run finishes. Each line answers what the agent did, when, and with which inputs. Say you let an agent triage your inbox overnight and label messages by topic. In the morning the log shows it read forty messages, applied a label to thirty-eight, and skipped two it could not classify, with a timestamp and the message it acted on for every step. Reading that trail, you catch the two silent skips, confirm the labels were sensible, and decide whether to let it archive on its own next time.

Common confusion: An audit log records what already happened so you can review it afterward. An approval gate stops an action before it runs and waits for a person to say yes. The log explains a finished run; the gate controls a pending one, so most setups use both.