The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from various external sources which could be attacker-controlled.
Ingestion points: The skill uses vaex.open(), vaex.from_csv(), vaex.from_arrow_table(), and pd.read_sql() to load data from local files, remote WebSocket servers, and cloud storage providers (S3, GCS, Azure) in SKILL.md and references/io_performance.md.
Boundary markers: The instructions do not specify any boundary markers, delimiters, or instructions for the agent to ignore potentially malicious content within the data being processed.
Capability inventory: The skill includes capabilities to write files (export_hdf5, export_parquet, state_write), perform network operations (cloud storage access and WebSocket connectivity), and visualize data, which could be leveraged if malicious instructions were successfully injected.
Sanitization: There is no evidence of data sanitization, schema validation, or content filtering before the data is used in computations or visualizations.
[EXTERNAL_DOWNLOADS]: The skill includes instructions to install external Python dependencies using pip.
Evidence: The Prerequisites section in SKILL.md lists vaex, vaex-hdf5, vaex-arrow, vaex-ml, vaex-viz, vaex-jupyter, s3fs, gcsfs, and adlfs. Other modules such as xgboost, lightgbm, catboost, and pyarrow are referenced in the documentation. These are standard packages within the data science ecosystem.
[DYNAMIC_EXECUTION]: The skill demonstrates the use of just-in-time (JIT) compilation for performance optimization.
Evidence: The references/io_performance.md file contains examples using numba.jit to compile Python functions for faster execution on DataFrame values. This is a standard practice for high-performance data processing tasks.

vaex-dataframes