browser-act
Pass
Audited by Gen Agent Trust Hub on Aug 2, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill installs the
browser-act-clipackage from PyPI, a well-known package registry, using theuvtool manager. - [COMMAND_EXECUTION]: The skill requires permission to execute
browser-actcommands via the Bash tool to perform its primary automation functions. - [PROMPT_INJECTION]: The skill processes untrusted external data from the web (URLs, rendered JavaScript content, and form fields), creating a surface for indirect prompt injection.
- Ingestion points: Rendered content from URLs, JavaScript execution results, and interactive page elements.
- Boundary markers: None explicitly defined in the provided file to separate external data from agent instructions.
- Capability inventory: Network access, local filesystem read/write for profiles/logs, and full browser automation (clicks, typing, uploads).
- Sanitization: The file does not specify how external content is sanitized, but it mentions a 'Confirmation Gate protocol' requiring user approval for sensitive actions.
- [REMOTE_CODE_EXECUTION]: The skill instructs the agent to run
browser-act get-skills coreto fetch its primary workflow instructions dynamically from the installed CLI tool.
Audit Metadata