The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted simulation output files (JSON/CSV) which could potentially be used as a vector for indirect prompt injection if an attacker controls the simulation results. \n
Ingestion points: Multiple scripts, including field_extractor.py and statistical_analyzer.py, read and process external simulation data files. \n
Boundary markers: No explicit boundary markers or instructions to ignore embedded content are present in the prompt interpolation. \n
Capability inventory: The agent is granted Read, Write, Grep, and Glob tools, and the Python scripts perform numerical analysis and data extraction. \n
Sanitization: The skill implements regex-based validation for field names (FIELD_NAME_PATTERN) and a secure manual parser for --region conditions in statistical_analyzer.py that avoids the use of eval() or exec(). \n- [SAFE]: The skill implements several best-practice security measures: \n
Input validation: User-supplied parameters such as field names, coordinates, and region conditions are validated against strict patterns. \n
Resource limits: A 500 MB file size limit is checked before parsing data files to protect against resource exhaustion attacks. \n
Directory protection: The report generator limits directory traversal to 10,000 entries. \n
Secure parsing: The skill avoids dynamic code execution and subprocess shell invocation, preferring manual parsing for expressions.

post-processing