The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill introduces jythonEval() and jythonEvalFile(), which are designed for the runtime execution of Python code. These functions act as an 'eval' sink, allowing strings to be interpreted as executable logic, which is a high-risk pattern if an AI agent constructs the code string using unsanitized external data.
[COMMAND_EXECUTION]: Scripts executed via this skill run within the Jython environment on the JVM. This gives the executed code the ability to interact with the host system, access the file system, and utilize Java libraries to perform system-level commands, potentially leading to full host compromise.
[INDIRECT_PROMPT_INJECTION]: The skill implements automatic variable binding where all values in the BoxLang variables scope are mapped directly into the Python engine. This creates a large attack surface for indirect injection:
Ingestion points: Data enters through the BoxLang variables scope as described in SKILL.md.
Boundary markers: There are no documented boundary markers or instructions to isolate the execution environment from these variables.
Capability inventory: The skill allows full Python 2.7 execution and JVM access via jythonEval and jythonEvalFile.
Sanitization: The documentation does not specify any sanitization or validation requirements for the data being passed from BoxLang to Python, allowing untrusted data to potentially influence the control flow of the Python scripts.

bx-jython