The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides an extensive index of community-maintained visual templates hosted on various third-party GitHub repositories.
Evidence: The file references/community-examples.md contains dozens of download links to raw.githubusercontent.com for repositories such as avatorl/Deneb-Vega-Templates, shadfrigui/vega-lite, and clemviz/Deneb-Templates.
Risk: These templates are fetched and injected into the user's Power BI report environment. Automated file reputation scanners have flagged these community resources, indicating potential risks associated with unverified external configuration data.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data directly from the Power BI report environment, which acts as an untrusted input source.
Ingestion points: The skill reads field names, measure names, and PBIR metadata from the report files (e.g., report.json, visual.json) using the pbir CLI tool.
Capability inventory: The skill writes JSON specification files and executes CLI commands (pbir set, pbir visuals bind, pbir add visual) to modify the report definition.
Boundary markers: No explicit boundary markers or "ignore instructions" warnings are utilized when interpolating report data into specifications.
Sanitization: While technical escaping for Power BI syntax is present, there is no security-focused sanitization to prevent malicious metadata from influencing the agent's behavior or the generated CLI commands.
[COMMAND_EXECUTION]: The workflow relies on the pbir CLI tool for all Power BI report mutations, executing commands based on user-supplied or report-derived inputs.
Evidence: SKILL.md and references/pbir-structure.md define a workflow centered on pbir command execution.
Risk: Maliciously crafted visual specifications or report metadata could lead to command argument manipulation, potentially impacting the integrity of the Power BI report definition.

deneb-visuals