Use SDK logging and the metrics collectors to observe polling, task execution,
workflow latency, retries, and failures. Configure log level with
CONDUCTOR_LOG_LEVEL; use METRICS.md for metric names and Prometheus setup.
For agent executions, inspect the shared workflow record for inputs, outputs, tool calls, retries, and status. Avoid logging credentials or unredacted sensitive data.