The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is entirely based on executing the agent-browser CLI tool to perform actions like opening URLs, interacting with DOM elements, and managing browser sessions.
[DYNAMIC_EXECUTION]: The agent-browser eval command allows for the execution of arbitrary JavaScript code within the browser context, which is a powerful capability that could be misused to manipulate page state or extract data.
[DATA_EXFILTRATION]: Several commands provide mechanisms for data extraction:
agent-browser upload @e1 file.pdf could be used to upload local files to a web form.
agent-browser screenshot and agent-browser pdf can save browser views to local files.
agent-browser cookies and agent-browser storage local allow access to potentially sensitive session data and authentication tokens.
agent-browser state save auth.json exports browser state (cookies, storage) to a local file.
[INDIRECT_PROMPT_INJECTION]: The skill's core workflow involves ingesting and acting upon content from external websites:
Ingestion points: agent-browser open <url> and agent-browser snapshot bring external DOM content into the agent's context.
Boundary markers: The instructions do not specify any delimiters or safety warnings for the agent when processing this untrusted data.
Capability inventory: The agent has access to a wide range of capabilities including network operations (open, network route), file system writes (screenshot, pdf, trace, state save), and code execution (eval).
Sanitization: There is no evidence of sanitization or filtering of the content retrieved from web pages before it is processed by the agent.

agent-browser