skills-for-fabric: Cursor rule for Cursor

Microsoft Fabric Development Rules

Updates: These instructions do not run a session-start update check. Refresh them through your host's plugin or marketplace update flow, or run git pull if you use a manual clone.

Context

This project works with Microsoft Fabric for data engineering, warehousing, and analytics. Primary API reference: https://learn.microsoft.com/en-us/rest/api/fabric/articles/

Architecture Mode

• Follow repository layering: Agents -> Skills -> Common.
• For medallion and other cross-endpoint workflows, start with agents/FabricDataEngineer.agent.md.
• Delegate endpoint-specific tasks to corresponding skills in skills/.

Authentication

All operations require Azure AD tokens via az login:

az login
az account get-access-token --resource https://api.fabric.microsoft.com

Developer Pattern: REST + Protocol

Developers use REST APIs to manage artifacts, then protocol-specific connections to access data:

• REST → ODBC for Warehouse
• REST → Spark for Lakehouse
• REST → XMLA/DAX for Semantic Models
• REST → KQL for Real-Time Intelligence

Code Generation Rules

Lakehouse & Spark

• Always use Delta Lake format for tables
• Use mssparkutils for Fabric-specific operations (fs, credentials, notebook)
• Include partition columns for large tables
• Run OPTIMIZE and VACUUM for table maintenance
• Use skills/project-osmos/SKILL.md when a local agent should carry a complex Lakehouse, OneLake, notebook, or Spark outcome end to end; Fabric portal Copilot returns availability guidance without Project Osmos API calls.
• Project Osmos owns target selection, MWC authentication and continuation in one skill. Use the selected Lakehouse URL or saved routing with its bundled resolver.
• Preserve the selected target, task IDs and saved /aichat or /tasks routes. Validate credential destinations, let the service enforce access, and never substitute production. Recheck assignments before execution; the production SKU policy applies only to production.
• Immediately after a successful Osmos POST /{taskId}/run response, show the run card and complete controls/feedback notice together in chat before further task operations or acting on follow-ups. Read task status/messages, report progress, then return control to the user; preserve cancel/delete confirmations and production resume capacity preflight.
• Load project-osmos for task controls and offline plans; select the existing task from conversation/recovery context and confirm ambiguity. Explicit Resume of a Completed task uses the same task's empty-body /run after execution checks, not a follow-up message. Offline mode persists across turns; Resume alone is not live execution consent. Return the complete requested offline plan each turn. Reject invalid portal URLs before calls; may request a valid URL for the intended environment without automatically switching targets.
• Explanation-only requests do not authorize sign-in, API calls or saved-state changes.
• Use skills/spark-cli/SKILL.md for notebook authoring and runs, Livy analysis, Spark diagnostics, and the full Materialized Lake View lifecycle.
• MLV authoring resources: skills/spark-cli/references/authoring/resources/materialized-lake-view-patterns.md and skills/spark-cli/references/authoring/resources/mlv-incremental-refresh-patterns.md.