The Agent Skills Directory

[COMMAND_EXECUTION]: The test suite in scripts/test_selection_toolkit.py utilizes subprocess.run to execute the main driver script, verifying the command-line interface and handling of accented characters across different console encodings.
[COMMAND_EXECUTION]: The verification module in scripts/selection_toolkit.py uses Excel COM automation via a PowerShell script payload (VERIFY_BODY) to perform full workbook recalculations and assert formula accuracy. This allows the tool to catch calculation errors by executing the generated workbooks in a live Excel environment.
[EXTERNAL_DOWNLOADS]: The script implements a custom bootstrap function, _bootstrap_ane_package, to locate and load the ane_package dependency from the local filesystem or predefined paths like WORK_FOLDER_ROOT and OneDrive. This is a mechanism for managing internal library dependencies.
[PROMPT_INJECTION]: The skill processes untrusted external Word documents (.docx) to identify and extract scoring criteria, weights, and thresholds. While the extraction logic is deterministic (based on table structures and regex), the ingestion of external data represents an indirect injection surface if the extracted text is subsequently processed by an LLM without strict boundary enforcement.
Ingestion points: The read_source function in scripts/selection_toolkit.py ingests text and table data from user-provided Word documents.
Boundary markers: The skill does not employ specific LLM boundary markers but instead uses regex and table-matching logic to isolate data.
Capability inventory: The skill has the capability to write Excel files (.xlsx) and execute PowerShell code via Excel's COM interface.
Sanitization: Content is normalized using Unicode NFKD folding to remove accents and parsed with regular expressions to extract numerical values.

selection-toolkit