What problem does it solve? Running Spark code in Microsoft Fabric normally requires creating and managing notebook artifacts, which adds overhead for quick transformations, ad-hoc queries, or agent-driven automation. This Skill lets you submit arbitrary Python or PySpark code directly to Fabric Spark compute through the Livy API with no persisted notebook. ## Core Features & Use Cases - Ephemeral Livy Sessions: Create, poll, submit statements to, and delete Spark sessions against a Fabric lakehouse, with full read/write access to Delta tables via Spark SQL. - Correct Authentication Flow: Uses an Azure CLI token scoped to the Fabric API, avoiding the common failure of using fab auth tokens that lack OneLake storage access. - Session Lifecycle Safety: Enforces cleanup of sessions in a finally block so idle sessions do not keep consuming Fabric capacity units. - Use Case: An agent needs to aggregate a lakehouse table and write a summary Delta table as part of an ETL pipeline. It creates a PySpark session, submits the transformation, reads the result, and deletes the session, all without creating any notebook artifact. ## Quick Start Ask the agent to run a PySpark query against your Fabric lakehouse using a Livy session and write the results back as a Delta table.
