The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill is designed to ingest and process external data from various file formats (CSV, HDF5, Parquet, Arrow). This creates a surface for indirect prompt injection if the processed data contains malicious natural language instructions intended to influence the agent's logic.
Ingestion points: vaex.open(), vaex.from_csv(), and vaex.from_pandas() in references/core_dataframes.md.
Boundary markers: Absent. The skill does not provide specific delimiters or 'ignore' instructions for the agent when reading external content.
Capability inventory: File system read/write (export_hdf5), data manipulation, and statistical computation.
Sanitization: Absent. Data is loaded and processed directly using standard library functions without content validation.
[Dependencies] (SAFE): The skill references standard, well-maintained data science libraries including vaex, pandas, numpy, and pyarrow. These are used for their intended purposes.
[Data Exposure & Exfiltration] (SAFE): While the skill accesses the local file system for data processing, there are no patterns suggesting unauthorized exfiltration or access to sensitive system paths (e.g., .ssh, .aws).

vaex