The Agent Skills Directory

[DYNAMIC_EXECUTION]: The documentation describes the 'Execute custom script' action, which allows embedding Python source code directly within a pipeline JSON definition to be executed at runtime within the platform's infrastructure.
[INDIRECT_PROMPT_INJECTION]: The skill defines a system that processes external inputs (such as project IDs, asset IDs, and user-provided parameters) and uses them to drive pipeline execution logic.
Ingestion points: inputParameters defined in the JSON schema and data retrieved from the Asset Manager via UI picker metadata.
Boundary markers: Instructions specify that template expressions ({{ }}) are resolved in the parameters field but remain literal text inside the script field, which acts as a design-level boundary to prevent unintended code injection during script creation.
Capability inventory: Actions include reading from and writing to /workspace/ directories, interacting with the Unity Asset Manager API via REST endpoints, and executing Python code using the pxz library.
Sanitization: The reference focuses on schema correctness and does not provide specific instructions for sanitizing inputs before they are processed by the internal Python scripts.

pipeline-authoring