The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill implements an eval command that allows the execution of arbitrary JavaScript within the browser instance. Documentation in references/javascript-evaluation.md details how to pass complex scripts via standard input or Base64 encoding, which provides a powerful but high-risk capability for running dynamic code.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to scrape and interact with external websites using commands like snapshot and get text. Because it processes untrusted HTML and text content from these sites without explicit boundary markers or sanitization logic in the provided templates, it is vulnerable to indirect prompt injection where malicious website content could attempt to influence the agent's behavior.
[OBFUSCATION]: The eval functionality supports a -b flag for Base64-encoded payloads. Although the documentation suggests this to prevent shell character corruption, it effectively obscures the JavaScript logic being executed from simple text-based monitoring and security audits.
[COMMAND_EXECUTION]: The skill operates through a custom command-line interface agent-browser and includes several shell templates that perform file operations, environment variable access, and process management to automate browser sessions and state persistence.

agent-browser