The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the agent-browser CLI to perform browser actions, navigate URLs, and manage session states. These operations are core functionalities and are documented with clear usage instructions.
[REMOTE_CODE_EXECUTION]: The eval command allows for the execution of JavaScript within the browser context. This is used for advanced page manipulation and data extraction. The documentation recommends using base64 encoding or stdin to ensure reliable execution and avoid shell escaping issues.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing dependencies from the Vercel-maintained agent-browser package via NPM.
[DATA_EXFILTRATION]: The skill manages sensitive data such as cookies and authentication tokens through session state persistence. It explicitly recommends best practices such as encrypting state files and adding them to .gitignore to prevent accidental exposure.
[PROMPT_INJECTION]: The skill interacts with external web content, creating an attack surface for indirect prompt injection.
Ingestion points: Web content is ingested into the agent context through the snapshot and get text commands in SKILL.md.
Boundary markers: Absent; there are no specific delimiters used when presenting web content to the agent.
Capability inventory: The skill can execute CLI commands, run JavaScript in the browser, and write files to the local system (SKILL.md, references/commands.md).
Sanitization: Absent; the skill does not specify any sanitization of the retrieved web content.

browser-qa