The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to process external event data through transformations, which creates a potential surface for indirect prompt injection.
Ingestion points: Test event fixtures located in tests/input/ and production event data processed by the JavaScript or Python transformation code.
Boundary markers: The skill utilizes standard JSON structure for input data but does not define specific natural language delimiters or 'ignore' instructions for the agent context.
Capability inventory: The skill utilizes the rudder-cli tool (via Bash) and file system operations (Read, Write, Edit) to test and deploy transformations.
Sanitization: The instructions include a dedicated 'Handling External Content' section recommending event validation, string sanitization, and avoiding dangerous functions like eval().
[DYNAMIC_EXECUTION]: The skill facilitates the creation, local testing, and deployment of JavaScript and Python code for real-time data manipulation.
Execution pattern: Transformation code is executed locally during the rudder-cli transformations test phase against JSON fixtures.
Mitigation: The documentation explicitly advises against using dynamic execution constructors like Function() or eval() on event data.

rudder-transformations