The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute arbitrary JavaScript within a browser context using the exec command. This capability is exposed through scripts/browser.py and the TMWebDriver class, allowing any user with access to the CLI to perform unrestricted actions in the browser.
[DATA_EXFILTRATION]: A major security vulnerability exists in assets/extension/content.js. The script uses a MutationObserver to watch for a specific DOM element ID (ljq_cmd). Any website visited by the user can inject this element with a JSON payload to command the extension to perform actions such as stealing cookies, querying other tabs, or executing Chrome DevTools Protocol (CDP) commands. The results are then written back to the DOM element, allowing the malicious site to exfiltrate the data.
[CREDENTIALS_UNSAFE]: The Chrome extension provides specific commands (cmd: 'cookies') to extract all cookies from the current or specified URLs. Furthermore, assets/extension/popup.js is designed to automatically copy all extracted cookies to the system clipboard, which could lead to accidental exposure of session tokens.
[REMOTE_CODE_EXECUTION]: The bridge server implemented in scripts/tmwd_bridge/TMWebDriver.py starts an unauthenticated HTTP and WebSocket server on 127.0.0.1. The /link endpoint allows for the remote execution of JavaScript code in the browser instance. If the port is exposed or accessible by other local processes, it provides a direct path for remote command execution within the user's browser.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of several external Python libraries (bs4, simple-websocket-server, bottle, requests) as documented in SKILL.md. While these are common packages, they are installed at runtime and provide the infrastructure for the unauthenticated command bridge.

browser-bridge