The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill manages Power BI Project (PBIP) structures which include metadata files like Copilot/Instructions/instructions.md. This file contains unstructured AI instructions that are read into the agent's context. A malicious project could contain adversarial instructions in these files or in project object names (tables, columns, measures) to manipulate agent behavior.
Ingestion points: scripts/validate_pbip.py and the agent's logic read project files including .pbip, .pbir, .tmdl, and markdown files in the Copilot/ folder.
Boundary markers: Absent. The skill does not instruct the agent to use delimiters or ignore instructions found within the project data.
Capability inventory: The skill leverages subprocess.run in its validation script and shell-based grep for project-wide searches.
Sanitization: Absent. File content is processed as raw text or JSON without sanitization.
[COMMAND_EXECUTION]: The validation script scripts/validate_pbip.py executes an external command-line tool pbir using the subprocess module. While the command is targeted at a specific tool, it operates on user-supplied project paths.
Evidence: subprocess.run(["pbir", "validate", str(report_dir), "--quiet"], ...) in scripts/validate_pbip.py (line 349).
[EXTERNAL_DOWNLOADS]: The validation process may fetch updated JSON schemas from Microsoft's official schema repository when local versions are outdated.
Evidence: Use of the --allow-download-schemas flag with the pbir tool as documented in SKILL.md.

pbip