agent-browser

Pass

Audited by Gen Agent Trust Hub on Jul 22, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches the agent-browser package and Chrome for Testing binary from Vercel Labs infrastructure. This is standard behavior for this browser automation tool.
  • [COMMAND_EXECUTION]: Utilizes shell-based commands to control browser actions, including navigating to URLs, capturing snapshots of the accessibility tree, and interacting with UI elements via specific references.
  • [PROMPT_INJECTION]: As a tool that processes external website content, the skill is subject to indirect prompt injection risks. However, the documentation explicitly instructs the agent to treat all browser output (HTML, console logs, network data) as untrusted data and not as instructions, which is a key security mitigation.
  • [CREDENTIALS_UNSAFE]: The skill provides an 'auth vault' (auth save / auth login) for managing secrets, specifically advising against echoing sensitive credentials directly into commands, which aligns with security best practices.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 22, 2026, 10:16 PM
Security Audit — agent-trust-hub — agent-browser