The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is instructed to strictly follow 'CustomInstructions' and 'VerifiedAnswers' retrieved from Power BI semantic models via the GetSemanticModelSchema tool. Because this content originates from external data sources (the semantic model metadata), it represents a surface where third-party instructions could influence agent behavior. The skill includes mandatory steps to read and apply this metadata, which is essential for functionality but requires the agent to handle potentially untrusted input.
External Data Ingestion via URLs: The ResolveReportIdFromUrl tool accepts user-provided Power BI or Fabric URLs. Processing these external inputs is a necessary part of the artifact discovery process, but it introduces a point where the agent interacts with user-controlled strings to resolve resource identifiers.
Dynamic Query Construction: The agent dynamically constructs DAX queries to be executed via the ExecuteQuery tool based on user questions and the discovered schema. To manage the risks associated with dynamic query generation, the skill provides a detailed set of 'DAX Rules' and 'Must/Prefer/Avoid' guardrails designed to ensure query integrity and restrict the scope of data access to the intended artifacts.

fabriciq