An incident review finds secret tokens inside agent traces. Which improvement is appropriate?

1. Publish all traces so users can debug themselves
2. Put the same secret in the system prompt for consistency
3. Redact sensitive fields and restrict trace access while retaining useful operational evidence
4. Disable all monitoring permanently