The Agent Skills Directory

[DYNAMIC_EXECUTION]: The Python scripts use __import__("pathlib") within the preamble to resolve the local directory and append it to sys.path. This is a common pattern for local module resolution to allow scripts to import the bundled _common and _catalog modules without requiring external package installation. It does not involve executing untrusted strings or loading remote code.
[DATA_EXPOSURE]: The skill is designed to process user-provided data files (CSV, TSV, or JSON). The read_input function in _common.py implements security best practices by enforcing a maximum input size of 5MB and limiting the number of processed rows to 20,000 to prevent denial-of-service via resource exhaustion. Access is limited to the files explicitly provided to the tool arguments.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data files but uses strict numeric validation (to_float) and structured parsing. The output consists of statistical summaries and findings written to stdout and stderr. The skill does not generate executable code or instructions for the agent based on the content of the data files, effectively neutralizing the risk of indirect injection.

analytical-method-validation