The Agent Skills Directory

[DATA_EXFILTRATION]: The skill implements strong preventative measures against unauthorized data access. The shared utility library (scripts/_common.py) strictly validates all file paths, rejecting those that include parent directory traversal ('..'), symbolic links, or URI schemes, ensuring that the agent only interacts with approved local data.\n- [INDIRECT_PROMPT_INJECTION]: The skill identifies and secures attack surfaces related to external genomic data processing. Included audit and planning scripts enforce hard bounds on input sizes, record counts, and line lengths, and utilize standard-library-only parsing to safely handle external manifests and BED files.\n- [DYNAMIC_EXECUTION]: The skill includes documentation and scripts designed to safely inspect machine learning artifacts. The model_artifact_inspector.py script inventories and hashes model bundles without deserializing untrusted formats like pickle or PyTorch checkpoints, mitigating the risk of arbitrary code execution from malicious models.

geniml