The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the bash tool to execute a wide variety of browse CLI commands. These commands control browser instances, interact with page elements, and manage cloud-based browser sessions.\n- [EXTERNAL_DOWNLOADS]: The skill instructions include commands to install the browse CLI globally via npm (npm install -g browse). Additionally, it features a plugin system where site-specific automation scripts ("skills") are discovered and downloaded from the browse.sh catalog (browse skills add <domain>/<task>). These resources are owned and operated by the vendor (Browserbase).\n- [PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it retrieves content from untrusted external websites and provides it to the agent for analysis (e.g., via browse snapshot, browse get markdown, or browse cloud search).\n
Ingestion points: Commands such as browse open, browse snapshot, browse get markdown, browse cloud fetch, and browse cloud search bring external web content into the agent's context.\n
Boundary markers: The instructions do not define clear delimiters or warnings to the agent regarding the untrusted nature of the retrieved web content.\n
Capability inventory: The agent has access to bash, which allows for file manipulation, network requests, and further skill installation, as well as the ability to execute arbitrary JavaScript in the browser via browse eval.\n
Sanitization: No explicit sanitization or filtering of the retrieved web content is performed before it is presented to the agent.

browse