The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill manages the attack surface of processing report data through the dataset variable by providing specific defensive recommendations.
Ingestion points: The dataset R data.frame is automatically injected into the visual's execution context at runtime, containing data bound to the visual in Power BI from the underlying report model.
Boundary markers: The skill relies on the pbir CLI tool's internal string escaping and handling when injecting generated R scripts into PBIR metadata, though no explicit prompt delimiters are used in the generated code itself.
Capability inventory: The skill uses the pbir CLI to manage report visuals and injects R scripts that execute statistical visualizations. In Power BI Desktop environments, the R runtime has unrestricted networking capabilities, while in the Power BI Service, networking is blocked.
Sanitization: The instructions explicitly teach defensive programming, such as using index-based access (dataset[,1]) to avoid name escaping issues and including input validation guards (if (nrow(dataset) == 0)) to handle empty data gracefully.
[EXTERNAL_DOWNLOADS]: The skill references documentation and community examples from established and well-known sources, including Microsoft Learn, the official ggplot2 website, and the R Graph Gallery. It also provides snippets for the installation of official CRAN packages, which are standard for the R development workflow and do not constitute a security risk in this context.

r-visuals