JevMade field notes / Architecture and workflow guide
skillranker
A rigorous operational guide to discovering, ranking, and evaluating agent skills with dry runs, shadow mode, privacy boundaries, replay, and promotion gates.
Original by DicklesworthstoneEvaluationRepository README and documentationSource reviewed
Before you dive in
What you’ll find in the original
Preview redacted wide-pass requests before allowing network or persistence effects.
Separate shadow observations from advisory behavior, and count fallbacks and unavailable outcomes in the denominator.
Use labeled fixtures and replayable provider records before promoting a ranking configuration.
Worth knowing
Behavior and measurements are reported by the repository authors; source and critical implementation were inspected but not executed.