The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents the use of $.child_process.exec() to execute system commands, providing examples such as $.child_process.exec("dir"). This capability allows scripts to perform arbitrary command execution on the host system.
[DYNAMIC_EXECUTION]: The documentation covers several dynamic execution features of the GraalJS environment, including load() for executing JavaScript files, Polyglot.eval() for other languages, and Java.extend() for creating Java proxies. These allow for runtime code generation and execution.
[EXTERNAL_DOWNLOADS]: The platform documentation notes that the load() function can accept a URL as a source, which enables the execution of remote scripts within the environment.
[INDIRECT_PROMPT_INJECTION]: The skill describes patterns for ingesting untrusted data from ArchiMate models and using it in sensitive sinks.
Ingestion points: Model content is accessed via $(selection), $("element"), and resolveSelection.selectedConcepts() as defined in the development guides.
Boundary markers: No delimiters or specific instructions for isolating untrusted model data are recommended.
Capability inventory: The environment supports powerful functions like $.child_process.exec(), load(), and java.io.File, which serve as high-privilege sinks for injected content.
Sanitization: The guide lacks instructions for escaping or sanitizing model data before passing it to system commands or dynamic execution functions.

jarchi-scripting