Observability & audit

See what an agent did through ExecWarden.

Useful agent work should leave enough evidence to review the caller, capability, approval, provider result, transcript, and output.

Recent activity

See what ran, what failed, and what paused for approval.

Recent activity showing successful, failed, and approval-paused GitHub calls from the Maintenance agent
The identity view shows a small operational history: what succeeded, what failed, and which sensitive action paused for approval.

What you can inspect

How this boundary works in practice.

The activity log covers capability calls mediated by ExecWarden. Pair it with transcripts, approval records, run history, provider results, and outputs to review the wider run.

Activity log

Record the governed action boundary.

Each record can show the caller, capability, grant evidence, provider or package action, status, timestamps, and configured payload capture.

Transcript

Keep the run narrative visible.

Prompts, tool calls, command output, approvals, and artifacts help reviewers understand how the result happened.

Approval history

Tie approvals to exact actions.

Approval records should show the payload that paused, who decided, and what happened after the decision.

Examples

Example tasks and outcomes.

1Review a PR run with branch output and its activity records.
2Inspect why a package-backed write was approved or rejected.
3Revoke or expire an identity after reviewing its recent calls.

Try it with your own run

Make one governed call, then inspect its activity record.

Open ExecWarden to create an Agent identity and run a real task, or review the evidence model in the docs first.

Open ExecWardenRead the evidence model