The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/JsEnv_Dev.js implements a WebSocket-based RPC client that includes an _execjs action handler. This handler uses eval() to execute arbitrary JavaScript code received from the connected WebSocket server in the browser context.
[DYNAMIC_EXECUTION]: The skill contains several Python generators, specifically scripts/emit_flask_proxy.py and scripts/emit_jsrpc_stub.py, which create new executable Python and JavaScript files at runtime based on the analysis of external websites.
[COMMAND_EXECUTION]: The automation workflow relies on the chrome-devtools-mcp tool to programmatically control a web browser, allowing the agent to perform complex interactions with any user-provided URL.
[OBFUSCATION]: The references/antidebug/debugger-loop.md documentation includes JavaScript snippets designed to overwrite Function.prototype.toString. This is a deceptive technique intended to hide the presence of security hooks and instrumentation from detection scripts running on the target website.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external websites which then influences the generation of local scripts and documentation.
Ingestion points: Browser DOM content and network traffic (XHR/Fetch) from user-supplied target URLs are analyzed.
Boundary markers: No specific delimiters or "ignore instructions" warnings were found in the prompt interpolation logic for analyzed data.
Capability inventory: The skill can control a browser, write files to the local filesystem, and execute generated code via a local Flask proxy.
Sanitization: While the skill performs input normalization on parameters, it lacks robust sanitization for the arbitrary JavaScript content it captures and processes during reverse engineering phases.

js-reverse-automation