The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external CSV files which presents a surface for indirect prompt injection if the data contains malicious instructions.
Ingestion points: Untrusted data enters the agent context through read_csv in scripts/_common.py, utilized by scripts like scripts/validate_survival_csv.py.
Boundary markers: The skill does not use specific delimiters to isolate external data from instructions, relying instead on standard numeric and tabular parsing.
Capability inventory: The skill includes capabilities to read and write local files via scripts/_common.py but does not include network operations or arbitrary shell command execution.
Sanitization: Extensive validation is implemented in scripts/_common.py, including checked_input_file which rejects URLs and symlinks, and strict type and size validation for all inputs.
[DYNAMIC_EXECUTION]: The script scripts/validate_survival_csv.py uses __import__("numpy") to lazily load the NumPy library for median calculations. This is a standard dynamic loading pattern for a well-known scientific dependency and does not incorporate untrusted input into the module name.
[SAFE]: The skill implements security best practices by disabling the unsafe pickle protocol in NumPy operations and explicitly rejecting network-based file paths and symbolic links for all file operations.

scikit-survival