The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The template script scripts/probe-template.mjs uses createRequire to dynamically load the playwright module from the current working directory (process.cwd()). This pattern allows for the loading and execution of arbitrary local code if the script is run in an untrusted directory.
[REMOTE_CODE_EXECUTION]: The skill utilizes page.evaluate to execute arbitrary JavaScript strings within a headless browser context. While intended for game state manipulation, this constitutes dynamic execution of potentially untrusted or agent-generated code.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface. It ingests content from external web pages via page.goto and monitors console and pageerror events (Ingestion points). It lacks boundary markers and sanitization for this untrusted data while possessing powerful capabilities (Capability inventory) like page.evaluate and page.keyboard controls.
[COMMAND_EXECUTION]: The skill instructions direct the agent to copy and adapt a shell-executable Node.js script to interact with the file system and launch browser processes.

probing-web-game-mechanics