The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of browser-act-cli==1.1.0 using the uv tool. The documentation explicitly states that the PyPI release consists of platform-specific binary wheels without a source distribution, which prevents independent security inspection of the underlying code.
[COMMAND_EXECUTION]: The agent is instructed to run the browser-act CLI to perform browser tasks. The skill relies on the local --help output of this external tool to determine available command syntax and capabilities.
[DATA_EXFILTRATION]: The CLI enables analytics and exception reporting by default, transmitting data to the vendor's infrastructure. Furthermore, features such as solve-captcha and remote-assist transmit challenge materials and browser session control to the vendor.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests untrusted data from the web.
Ingestion points: Navigation to external URLs, DOM extraction, and network capture.
Boundary markers: The skill warns the agent not to follow provider-served runtime guides but lacks specific delimiters or markers to separate web content from agent instructions.
Capability inventory: The tool can perform clicks, form inputs, file uploads, and state-changing actions on remote websites.
Sanitization: No sanitization or filtering of extracted content is specified before the agent processes it.
[REMOTE_CODE_EXECUTION]: The skill automates the execution of JavaScript within a browser environment to render pages and interact with web elements, which represents a controlled form of code execution on remote content.

browser-act