Record every agent run.
Know what happened.

Capture the process, terminal, models, tools, approvals, and multi-agent handoffs in one portable trace. Investigate failures wherever agents run, then share the evidence—not a guess.