The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill documentation includes implementation patterns for model persistence using pickle.load() and joblib. While these methods are common in data science, the documentation explicitly identifies that loading untrusted files is a security risk and recommends migration to ONNX or TorchScript for production use cases.
[COMMAND_EXECUTION]: The skill references administrative workflows involving the exaslct and exapump CLI tools. These are standard platform utilities for managing script language containers and file synchronization with BucketFS.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from database tables and storage artifacts for processing by machine learning models.
Ingestion points: Python UDFs ingest data via ctx.get_dataframe() from database tables (e.g., ml.training_data) and load artifacts from the /buckets/ mount points.
Boundary markers: The skill relies on standard Python and SQL syntax boundaries; it uses ctx.reset() to manage data cursors for multi-pass algorithms.
Capability inventory: UDF scripts have access to the BucketFS file system and can perform network uploads to the BucketFS API using credentials stored in Exasol CONNECTION objects. Lua scripts orchestrate SQL execution via the query API.
Sanitization: The skill processes numerical and categorical data for statistical modeling; no explicit natural language sanitization is applied to input data, which is standard for internal database ML pipelines.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions for configuring environment containers with standard machine learning libraries (e.g., scikit-learn, PyTorch, TensorFlow) and vendor-maintained packages (e.g., pyexasol, exasol-bucketfs) sourced from official package registries.

exasol-distributed-ml