The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads and processes external Word documents (.docx) to extract criteria, weights, and dates. While the driver script uses structured parsing logic, the ingestion of untrusted external documents constitutes a surface for indirect prompt injection.\n
Ingestion points: The read_source function in scripts/selection_toolkit.py extracts text and tables from external .docx files.\n
Boundary markers: No explicit boundary markers or instructions to ignore embedded content are present in the prompt context.\n
Capability inventory: The skill possesses capabilities to write Excel files and execute PowerShell snippets via Excel COM (through the ane_package dependency).\n
Sanitization: The script uses regex-based validation for weights (NUMBER) and dates (DATE_PATTERNS) which limits the impact of arbitrary text.\n- [DYNAMIC_EXECUTION]: The scripts/selection_toolkit.py script uses dynamic path manipulation (sys.path.insert) to bootstrap the ane_package library (a vendor resource) from the user's environment. It also executes a hardcoded PowerShell script (VERIFY_BODY) via an Excel COM bridge to automate workbook recalculation and verification, which is a functional requirement for the toolkit's operation.\n- [COMMAND_EXECUTION]: The included test suite (scripts/test_selection_toolkit.py) uses subprocess.run to execute the driver script during integration testing. This is a standard and safe practice for testing command-line interfaces.

selection-toolkit