The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to browse and extract data from the web, creating an attack surface where malicious website content could attempt to influence agent behavior.
Ingestion points: Untrusted data enters the agent context through browse open, browse snapshot, browse get html, browse get markdown, and browse cloud fetch (SKILL.md).
Boundary markers: No specific delimiters or instructions are provided to help the agent distinguish between tool output and instructions embedded in web content.
Capability inventory: The skill uses the Bash tool and can execute shell commands, write files to the local system (browse screenshot), and execute JavaScript within the browser environment (browse eval).
Sanitization: There is no mention of sanitizing, filtering, or escaping the HTML/Markdown content fetched from external URLs before it is processed by the agent.
[DYNAMIC_EXECUTION]: The browse eval command executes arbitrary JavaScript within the browser context. This is a core feature for browser automation but presents a vector for dynamic code execution if the input is influenced by untrusted data.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the browse CLI package from NPM (npm install -g browse) and supports downloading site-specific automation scripts from the browse.sh catalog. These resources originate from the vendor's infrastructure.
[COMMAND_EXECUTION]: The skill relies on the Bash tool to execute browse CLI commands, which involve network operations, local browser management, and file system interactions.

browse