Local/offline audit for AI security-agent traces: repeated tool-output work, stale replay risk, and token-savings reports.