The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides extensive guidance on using MoonBit's extern "js" and inline JavaScript literals (#|) to execute JavaScript code directly. This includes accessing browser globals like window and document, as well as Node.js built-ins.
[DATA_EXFILTRATION]: Documentation includes examples of using the JavaScript fetch API via FFI declarations to perform network requests. While presented as a developer tutorial, this represents a potential path for data exfiltration if used by an agent to send sensitive data to an external server.
[REMOTE_CODE_EXECUTION]: The references/error-handling.md file contains a test example that demonstrates calling JavaScript's eval function via the call_method FFI primitive to trigger and catch exceptions.
[PROMPT_INJECTION]: The skill establishes a functional bridge for processing external data (e.g., via json_parse and generic JsValue wrappers). This creates an attack surface for indirect prompt injection (Category 8) where untrusted data could influence agent behavior through the execution of generated FFI logic.
Ingestion points: json_parse in ffi.mbt processes external strings into JS objects.
Boundary markers: Absent in the provided examples.
Capability inventory: Subprocess-like execution via extern "js", network access via fetch, and dynamic execution via eval wrappers are present across several files (e.g., ffi.mbt, references/error-handling.md).
Sanitization: No sanitization or validation of external content is demonstrated in the examples.

moonbit-js-binding