Observing Agentforce

This is the monitoring and debugging counterpart to agent development. It queries Salesforce Data Cloud's Session Trace Data Model to pull real production conversation logs, then lets you reproduce issues locally with `sf agent preview` and fix them by editing the .agent file directly. The workflow is observe, reproduce, improve. If your org doesn't have STDM enabled, it falls back to running test suites and analyzing local traces instead. The skill resolves agent names against the org metadata, handles the double-nesting bug in `sf project retrieve`, and walks you through the three-phase loop. Use this when production agents are misbehaving and you need actual session data to figure out why, not when you're writing new agents from scratch.

Install to Claude Code

npx -y skills add forcedotcom/sf-skills --skill observing-agentforce --agent claude-code