The Agent Skills Directory

Arbitrary JavaScript Execution: The eval command enables the execution of JavaScript within the browser context. This is a powerful feature intended for complex DOM manipulation and data extraction, though it necessitates that the input to the command be handled by a trusted agent.
Sensitive Data Management: The skill manages authentication state, including cookies and session tokens, which are stored in local files. The documentation provides clear guidance on securing these files using environment-based encryption (AGENT_BROWSER_ENCRYPTION_KEY) and standard file-exclusion practices.
Local File System Access: The skill supports the --allow-file-access flag, permitting the browser to open local documents via file:// URLs. This capability is useful for automated document processing but grants the browser access to the local filesystem.
Indirect Prompt Injection Surface: Since the tool processes content from external websites, there is a possibility that web pages could contain instructions intended to influence the agent's behavior. The skill includes an optional --content-boundaries feature that uses nonces to help the agent distinguish between system output and untrusted web content.
Clipboard and Network Inspection: The inclusion of clipboard read/write and network requests commands allows for comprehensive automation workflows, such as handling 2FA or debugging API calls, while requiring the agent to have access to system-level data.
Cloud Provider Integration: The skill facilitates connections to various third-party cloud browser services. These integrations are managed through standard configuration patterns and are intended for scaling automation tasks.

agent-browser