agent-observabilitylisted

# Agent Observability ## Overview Agents are non-deterministic — they make different decisions on the same input, they hallucinate, they exceed rate limits, they cost money. Without observability, you're operating blind. This skill covers logging, metrics, alerting, and cost tracking for GTM agents. ## Frameworks Referenced This skill is grounded in public frameworks and source material relevant to the task: - **Agent Observability Framework.** Use the relevant method or published guidance where it improves the requested deliverable; do not cite it as decoration. - **LLM Ops Best Practices.** Use the relevant method or published guidance where it improves the requested deliverable; do not cite it as decoration. - **Anthropic — Agent Skills Progressive Disclosure.** Use the relevant method or published guidance where it improves the requested deliverable; do not cite it as decoration. ## When to Use - "Monitor my agents" - "Set up agent observability" - "Track agent performance" - "Agent cost analysis" - "Debug agent behavior" ## Step-by-Step Process ### Phase 1: What to Log Every agent action must be logged: **Input layer:** - Task description / user prompt - Session context (account, opportunity, campaign) - Timestamp **Decision layer:** - Which tool was called and why - Tool inputs and outputs (sanitized — no PII in logs) - Latency: time from tool call to response **Output layer:** - Agent's final response - Decision path: what tools were called, in what order, wi