power-query

Warn

Audited by Gen Agent Trust Hub on Apr 25, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses subprocess.run in examples/execute_m.py to call az account get-access-token for credential acquisition and in examples/preview_partition.py to call the fab CLI for model definition retrieval.
  • [COMMAND_EXECUTION]: examples/preview_partition.py performs dynamic module loading using __import__ to resolve the path of a sibling module at runtime.
  • [DATA_EXFILTRATION]: The skill retrieves environment-specific access tokens and transmits Power Query mashup documents—which may contain sensitive logic or connectivity metadata—to the Microsoft Fabric API at api.fabric.microsoft.com.
  • [PROMPT_INJECTION]: The skill manages an indirect prompt injection surface by generating and executing code based on natural language inputs.
  • Ingestion points: User prompts mentioning 'Power Query', 'M code', or 'partition expression' in SKILL.md.
  • Boundary markers: No specific delimiters or warnings for the agent to ignore embedded instructions are identified.
  • Capability inventory: Execution of Power Query M code via external APIs, shell command execution via subprocess, and dynamic Python module loading.
  • Sanitization: No sanitization or validation logic for the generated M code or CLI parameters is specified.
  • [EXTERNAL_DOWNLOADS]: The skill requires external dependencies including the pyarrow Python package and the az and fab CLI tools.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 25, 2026, 01:34 AM