The Agent Skills Directory

[DYNAMIC_EXECUTION]: The utility script scripts/normalize_depth_paths.py uses pickle.load() to deserialize annotation files.
The pickle module is notoriously insecure as it can be used to execute arbitrary code during the unpickling process.
The script accepts file paths as command-line arguments and proceeds to load them using pickle, creating a direct path for code execution if an agent is tricked into running the script against a malicious file.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an ingestion surface for potentially untrusted data that could influence agent behavior.
Ingestion points: The script scripts/normalize_depth_paths.py reads data from .pkl annotation files provided via command-line arguments.
Boundary markers: The script does not utilize delimiters or specific instructions to the AI to ignore embedded content within these data files.
Capability inventory: The skill has the capability to read and write files and execute local Python scripts. The deserialization process itself provides a mechanism for arbitrary code execution.
Sanitization: There is no evidence of validation or sanitization of the data retrieved from pickle.load() before it is processed or written back to disk.

tao-train-sparse4d