AgentProbe/docs/references/observability.md

Observability Reference

Required primitives

• Structured logs
• Metrics with stable names and labels
• Traces/spans around critical workflows
• Correlation identifiers for runs and scenarios

Critical paths to instrument

• CLI startup
• Config and YAML parsing
• Endpoint/session lifecycle calls
• Judge/model calls
• Persistence writes and reads
• Report rendering

Why this matters

Observability is not just for production incidents. It is part of the local agent feedback loop that makes latency requirements and debugging enforceable.