Skills
Auto-generate your regression suite from production failures via trace classification
2026 continuous-evaluation practice: classify production traces automatically so a scored failure can trigger alerts, review queues, and new regression tests without manual trace-by-trace scoring — the highest-value habit being that every production regression that reaches a user becomes a permanent test case. This builds a golden dataset grounded in how your agent actually fails rather than in synthetic cases. Builder move: wire a classifier over your trace logs that promotes any failed-or-flagged trajectory into your replay suite, so your eval coverage compounds automatically with real usage.
Source
↳ Follow the thread