Runs as a batch job over historical logs — no live interception, no change to any agent.
No stated priorities are set, so this produces a behavior report: what your agents optimize for, described but not judged. Confirm or correct those findings — or state priorities up front — to get a gap report.