The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface\n
Ingestion points: The scripts data_wrangler.py and excel_toolkit.py read data from potentially untrusted external sources, including CSV, Excel, JSON, and Parquet files.\n
Boundary markers: The skill does not employ specific delimiters or instruction-isolation techniques when processing data from these files.\n
Capability inventory: The skill possesses capabilities for file system writes, directory creation, and dynamic string evaluation, which could be exploited if malicious data influences the agent's logic.\n
Sanitization: While filename sanitization is present in the split operation, there is no comprehensive sanitization of data content before it is used in expression evaluation.\n- [COMMAND_EXECUTION]: Dynamic Evaluation of Expressions\n
The data_wrangler.py script utilizes pandas.DataFrame.eval() and pandas.DataFrame.query() for core features like filtering, computed columns, and row dropping. These methods evaluate string-based expressions at runtime.\n
The query operation in data_wrangler.py supports SQL execution via the pandasql library, which involves dynamic translation and execution of SQL against local DataFrames.

data-wrangler