The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and processing of untrusted data from multiple sources, creating a surface for potential injection attacks.
Ingestion points: Data enters the script environment through the args global (user-supplied parameters), the fetch() function (external API responses), and the query() function (database results).
Boundary markers: The documentation and examples do not define explicit delimiters or instructions to ignore embedded commands within these data sources.
Capability inventory: The environment provides the fetch() tool for outbound network requests and the query() tool for SQL data access.
Sanitization: No specific sanitization or escaping mechanisms for external content are mentioned in the provided documentation.
[DYNAMIC_EXECUTION]: The skill describes a system that executes Python code provided by the agent. This execution is performed using RustPython compiled to WebAssembly (WASM), which provides a sandbox environment. The documentation explicitly notes that the environment has no file system access and is subject to memory and time constraints.
[DATA_EXFILTRATION]: The skill documents the simultaneous availability of data reading tools (query) and network transmission tools (fetch). While this enables potential data exfiltration, the capability is presented as a primary feature of the Datadata platform for legitimate data transformation and integration workflows.

datadata-python