What problem does it solve? Agentforce conversations are opaque once deployed: teams cannot see why an agent routed to the wrong topic, which action failed, or where latency and hallucinations occur. This Skill extracts Session Tracing Data Model (STDM) records from Salesforce Data Cloud and turns them into analyzable Parquet datasets for trace-driven debugging and quality monitoring. ## Core Features & Use Cases - High-Volume Extraction: Pull 1-10M records per day from 24 verified STDM and GenAI Trust Layer DMOs via the Data Cloud Query API with JWT Bearer authentication, storing results in compact Parquet files. - Polars-Based Analysis: Compute session summaries, topic routing distributions, action failure rates, latency hotspots, and hallucination or toxicity detection using lazy evaluation over 100M+ rows. - Session Debugging: Reconstruct complete session timelines (requests, responses, LLM steps, action steps) for a single conversation to diagnose routing or action failures. - Use Case: After users report an agent giving wrong answers, extract the last 7 days of telemetry, find sessions flagged UNGROUNDED by the Trust Layer, and reconstruct the failing session timeline to identify the misconfigured topic. ## Quick Start Ask the agent to extract the last 7 days of Agentforce session tracing data from your org and generate a summary analysis of topics, completion rates, and failed sessions.
