Agent Observability

Use this skill when agent behavior needs to be traceable, debuggable, and measurable beyond simple pass/fail checks.

When to Use

• deploying agents to production
• diagnosing unexpected or wrong agent outputs
• attributing token/cost across tenants
• building evaluation datasets from real failures

Core Rules