browserbase-sdk
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates browser automation and AI-powered web scraping, which creates a surface for indirect prompt injection where untrusted content from the web could influence the agent's behavior.
- Ingestion points: Untrusted web content is ingested through commands like
page.gotoand processed using Stagehand primitives such asactandextract, as detailed inSKILL.mdandreferences/stagehand.md. - Boundary markers: The skill does not explicitly provide or require the use of delimiters or 'ignore' instructions for the data being scraped.
- Capability inventory: The agent has the ability to navigate the web, interact with page elements, take screenshots, download files, and access environment variables for API keys (e.g.,
BROWSERBASE_API_KEY). - Sanitization: The documentation encourages the use of Zod schemas for structured data extraction in
references/stagehand.md, which helps validate and sanitize the information retrieved from external pages.
Audit Metadata