The Agent Skills Directory

[Data Exposure & Exfiltration] (MEDIUM): The skill provides tools to extract sensitive session data.
Evidence: Commands agent-browser cookies, agent-browser storage local, and agent-browser state save allow for the extraction and local persistence of authentication tokens and session secrets.
Evidence: The agent-browser upload command allows the agent to read files from the local filesystem and transmit them to external web servers.
[Dynamic Execution] (MEDIUM): The skill permits arbitrary code execution within the browser context.
Evidence: The agent-browser eval command enables the execution of arbitrary JavaScript, which can be used to manipulate page logic or exfiltrate data not accessible via the standard DOM.
[Indirect Prompt Injection] (LOW): High vulnerability to instructions embedded in web content.
Ingestion points: agent-browser snapshot, agent-browser get text, and agent-browser console (File: SKILL.md).
Boundary markers: Absent. The skill does not provide delimiters or instructions to treat scraped content as untrusted data.
Capability inventory: High risk combination of file system access (upload, state save), credential access (cookies), and dynamic execution (eval).
Sanitization: Absent. External web content is passed directly back to the agent's context.

agent-browser