The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface area because it processes user-provided natural language and metadata from external data assets (like table and column names) to generate code.
Ingestion points: Natural language queries from the user and asset metadata retrieved via tools like inspect_project_asset.
Boundary markers: There are no explicit instructions for using delimiters when interpolating user input or metadata into the generation prompts.
Capability inventory: The skill generates executable SQL and uses tools to browse and inspect data assets and connections.
Sanitization: The skill mandates strict error redaction to prevent the exposure of secrets, though it does not explicitly specify sanitization for user-provided input strings before generation.
[COMMAND_EXECUTION]: The skill dynamically generates Substrait DSL and SQL queries at runtime. This behavior is the primary intended purpose of the skill. The workflow includes deterministic compilation steps and error correction logic to ensure the generated code is valid.
[EXTERNAL_DOWNLOADS]: The skill references the /data_intg_ai/v1/runtime/sql API endpoint via the substrait_to_sql tool. This is a vendor-owned resource associated with IBM's data integration infrastructure and is used for the legitimate purpose of SQL conversion.

di-agent-query-sql