observing-agentforce

To debug Agentforce production failures, analyze STDM session trace data from Data Cloud to surface action errors, subagent misroutes, and variable capture bugs. This approach queries session traces to pinpoint production regressions and performance issues.

Low instruction adherence in Agentforce sessions is diagnosed by classifying STDM session traces and mapping symptoms back to the .agent authoring bundle structure. It often indicates missing actions, incorrect inputs, or publish drift requiring bundle edits.

You can reproduce Agentforce production issues locally using sf agent preview with generated scenarios when Data Cloud STDM is unavailable. This uses fallback local preview traces to confirm whether failures are consistent or intermittent.

Diagnose abandoned Agentforce sessions and subagent misroutes by extracting and classifying session traces from Data Cloud. The process resolves the agent identity, discovers the Data Space, and optionally deploys an STDM helper Apex class to query the traces.

An STDM helper Apex class is required to analyze Agentforce session traces only when needed by your Data Cloud setup. The workflow discovers your Data Space and deploys the helper class to extract STDM session trace data for diagnosis.

The best way to fix Agentforce publish drift is to validate fixes by editing and publishing the .agent authoring bundle. This workflow maps detected symptoms from session traces back to the bundle structure to apply a safe deployment and verification loop.