Sf Ai Agentforce Observability

When your Agentforce agent misbehaves in production, this pulls the telemetry to figure out why. It extracts Session Tracing Data Model records from Data 360, dumps them to Parquet files, and gives you Polars-based analysis patterns to reconstruct what happened during actual user conversations. You get session timelines, topic routing paths, action failures, and latency breakdowns. It's strictly observability, not testing or building the agent itself. The skill assumes you already have Data 360 and session tracing turned on, plus JWT auth wired up. If you find a root cause here, you hand off to the authoring or testing skills to actually fix it. Think of it as the black box recorder for agent conversations.

Install to Claude Code

npx -y skills add jaganpro/sf-skills --skill sf-ai-agentforce-observability --agent claude-code