The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill is designed to ingest and process semantic model metadata, such as table descriptions, measure names, and DAX code, which are external to the agent's instructions. This presents an inherent surface for indirect prompt injection, as malicious instructions could be embedded within these model properties.
Ingestion points: Fabric semantic model metadata (tables, measures, descriptions) and lakehouse schema definitions accessed via az rest and MCP tools.
Capability inventory: The skill has the capability to write local files (TMDL scaffolding), perform network operations (via az rest), and modify live models in the Fabric service.
Boundary markers: No specific delimiters or "ignore embedded instructions" markers are defined for the metadata content being processed.
Sanitization: The instructions do not specify sanitization or validation routines for external metadata strings before they are interpolated into the agent's context.
[External Dependency Management]: The skill depends on the @microsoft/powerbi-modeling-mcp package, which is fetched and executed via npx at runtime. This package originates from a trusted organization (Microsoft) and is a standard component for the skill's modeling functionality.
[Command Execution via CLI]: The skill frequently uses az rest to interact with Microsoft Fabric and Power BI APIs. These commands are used within the intended scope of managing workspace items, model definitions, and connection bindings.

semantic-model-authoring