The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill facilitates the generation of Python scripts using the ibm-watsonx-data-integration SDK to define DataStage ETL flows.
Evidence: The references/sdk-conventions.md file provides extensive examples and rules for constructing these scripts, which are eventually passed to the create_datastage_flow or update_datastage_flow tools.
Mitigation: The skill explicitly prohibits complex Python features, such as loops, conditionals, functions, classes, and arbitrary imports, effectively restricting the generated scripts to a flat configuration-like format.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a transformer for user requirements and existing flow code into executable SDK scripts, creating a surface for potential instruction injection via data ingestion.
Ingestion points: User input utterances and existing flow code retrieved from the retrieve_datastage_flow_code tool in SKILL.md.
Boundary markers: No specific delimiters or "ignore" instructions are defined for the interpolation of untrusted data into the generated SDK code.
Capability inventory: The generated code is processed by tools that can interact with project data assets, database connections, and create/run jobs on the watsonx platform.
Sanitization: The skill relies on strict syntax validation and the requirement that all stage/property names must be verified against official lookup tools like datastage_property_lookup.

di-agent-flow-datastage