Harness logs are per-step only today. There is no cross-step search, no correlation with infrastructure events, and no way to download a diagnostic bundle for escalation. Developers navigate: execution view → step logs → delegate page → manual correlation across separate screens. Slack data: cryptic errors with zero context requiring multi-hop investigation to diagnose.

We need:

— Structured JSON log output with severity classification

— Full-text search across all steps in an execution

— Log correlation with system events (delegate restart, pod eviction, OOM kill)

— Export to S3 / GCS / ELK / Splunk for teams with existing log infrastructure

— One-click diagnostic bundle download: all step logs, infra events, delegate data packaged for escalation

This pairs with FR-9 (semantic log level classification) and FR-1 (resource metrics).

Business value

Eliminates multi-hop diagnosis. Enables self-service root cause analysis without opening a support ticket.