The Agent Skills Directory

[PROMPT_INJECTION]: The skill describes processes for ingesting large datasets from various file formats and remote sources, creating a potential indirect prompt injection surface.
Ingestion points: Documented in references/core_dataframes.md and references/io_operations.md using methods like vaex.open() and vaex.from_csv() for local and cloud storage (S3, GCS, Azure).
Boundary markers: The instructions do not specify explicit delimiters or warnings for the agent to ignore instructions embedded within the processed data.
Capability inventory: The skill documents extensive capabilities including file system writes (export_hdf5), network operations for cloud storage, and parallel data processing across multiple cores.
Sanitization: There is no mention of sanitization or validation of the data content before processing, which is standard for data analysis tools but constitutes an attack surface.
[REMOTE_CODE_EXECUTION]: The skill documentation includes patterns for dynamic code execution and model evaluation for performance and analytics purposes.
Evidence: references/performance.md demonstrates the use of numba.jit for Just-In-Time compilation of custom mathematical operations and df.apply() for row-wise transformations.
Evidence: references/machine_learning.md provides patterns for integrating and executing models from third-party libraries such as scikit-learn, XGBoost, and TensorFlow.
[COMMAND_EXECUTION]: The skill includes instructions for starting a remote data server using the vaex-server command-line utility in references/io_operations.md, which is a standard feature for distributed data access.

vaex