The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The tool browser_evaluate allows the agent to execute arbitrary JavaScript code within the context of the loaded web page. Additionally, browser_add_script_tag can be used to inject and execute remote scripts from external URLs.
[DATA_EXFILTRATION]: The skill includes tools to extract sensitive session information, such as browser_get_cookies, browser_get_local_storage, and browser_storage_state. If an agent is compromised or follows a malicious instruction, this data could be sent to an external server.
[COMMAND_EXECUTION]: The browser_upload_file tool allows the agent to select and upload files from the local filesystem to a website. This presents a high risk of sensitive file exposure if the agent is manipulated into uploading files like SSH keys or configuration files (e.g., .env).
[COMMAND_EXECUTION]: Tools such as browser_screenshot, browser_pdf, and browser_storage_state allow the agent to write data to specified local file paths, which could be used to overwrite system files or create persistence if not properly sandboxed.
[PROMPT_INJECTION]: The skill is highly susceptible to indirect prompt injection. Since it navigates to external websites via browser_goto and extracts text content via browser_get_text, a malicious website could host instructions that the agent might interpret as authoritative commands.
[DATA_EXFILTRATION]: The browser_grant_permissions tool allows the agent to bypass standard browser security prompts for geolocation, microphone, and camera access, potentially leading to unauthorized surveillance or tracking.

playwright-browser