Reviewing the agent activity log

The activity log shows every autonomous action the agent took, the plain-language reason behind it, whether it can be undone, and the exact policy it was operating under.

The agent activity log is where you see everything the agent has done on its own, each entry with a plain-language "why" — the reasoning it used, not just the result. So an automatic action is never a black box: you can read after the fact why a reminder went out, why a repair was dispatched, or why a fee was or was not applied.

It reads as one chronological feed you can scan or filter, sitting alongside your approval queue. The queue is what needs you now; the activity log is the history of what has already happened, handled and autonomous.

Each action carries a reversibility tier, so you can tell at a glance whether it can be walked back. There are two ways to do that. Cancel stops a pending action that is still inside its soft-commit safety window, before it ever executes. Undo reverses an action that has already committed, for the ones that are reversible.

Both live right on the activity feed next to the action they apply to, so walking something back is a direct step rather than a support request. Actions that genuinely cannot be reversed are labeled as such, so you are never misled about what an undo will and will not do.

The log records the policy that was in force at the moment of each action, so you can see the exact thresholds the agent was operating under when it made a given call — not just the settings you happen to have today. That is what lets a past decision be judged on its own terms.

For anything formal, the activity feed is backed by a tamper-evident, hash-chained audit log of every agent action and every policy change. On the web app you can filter it by actor and date, export it to CSV, and produce a Decision Integrity Certificate attesting the record has not been altered — useful if a tenant disputes a charge or a regulator or your attorney asks how an automated decision was made.