There is no queryable record of what a Bot actually did

SymptomAfter an incident, the only evidence is per-Bot chat transcripts, not queryable across a team, and only the 20 most recent run records per routine.

CauseAn audit view is described in the docs as coming. Per-run usage metering, success-rate benchmarks and export paths between Bots are all unpublished.

DetectionDiscovered at the moment an incident needs reconstructing, which is the worst time.

FixExport run-history evidence into source systems as the routine runs, rather than relying on in-product retention. Require every routine to write a reviewable artifact with source links.