An agent audit trail is the ordered record of what an AI agent did, end to end: the instruction it was given, the action it took, the systems it touched, the answers that came back and what authorized each one.
A log line is whatever the application chose to write. An audit trail is ordered, attributed and made of what actually passed: the request that went out, the answer that came back, who sent which. If the entries can be rewritten in place, it is a diary, not a trail.
"Who did this" cannot live in the message, because anyone can write a field. FlurryPORT records which key signed each capture, so attribution is a property of the record rather than a claim inside it.
A trail is worth what it outlives. Captures age out with your plan’s window; a question you will ask in three weeks needs a window longer than three weeks. Pick the window before you need the history, not after.
Put one capture URL in front of the systems an agent acts on. Every post is signature-checked and stored with the label of the key that signed it; auto-forward delivers it to each destination and records what each one answered; a watch labels the actions that matter. Afterwards the record answers what was sent, when, where, and what came back, without asking the agent. The walkthrough below does it in seven steps.
The glossary is the canonical index of every FlurryPORT term these pages use.