The Agent Skills Directory

[DATA_EXFILTRATION]: Access and exfiltration of sensitive data through browser controls.\n
agent-browser upload allows the agent to select and send local files to remote websites.\n
agent-browser cookies and agent-browser storage local provide tools to retrieve session tokens, credentials, and personally identifiable information (PII) from the browser context.\n
agent-browser state save enables the agent to write session state, including authentication cookies, to a local file for later persistence.\n- [COMMAND_EXECUTION]: Arbitrary code execution within the browser environment.\n
agent-browser eval allows the execution of arbitrary JavaScript strings, which can be used to bypass UI constraints or manipulate page logic in ways not intended by the user.\n
agent-browser wait --fn evaluates JavaScript conditions to gate execution flows.\n- [PROMPT_INJECTION]: High risk of indirect prompt injection from processing external web content.\n
Ingestion points: The agent retrieves untrusted content from the web using agent-browser snapshot, agent-browser get text, and agent-browser get html in SKILL.md.\n
Boundary markers: The skill does not provide any delimiters or instructions to help the agent distinguish between trusted instructions and untrusted data fetched from the DOM.\n
Capability inventory: The agent has extensive capabilities including file system write (state save), local file read/upload (upload), and script execution (eval).\n
Sanitization: No sanitization, validation, or filtering of the extracted web content is implemented before it is presented to the agent's context.

agent-browser