agent-browser

Warn

Audited by Gen Agent Trust Hub on Apr 18, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill requires installing the agent-browser CLI tool globally and downloading the Chromium browser binary. These actions involve fetching code and executable binaries from external sources that are not listed as trusted vendors.
  • [COMMAND_EXECUTION]: The skill's core functionality relies on executing the agent-browser CLI, which allows for complex interactions with the web, management of browser sessions, and the creation of local files such as screenshots and PDF documents.
  • [PROMPT_INJECTION]: The skill exposes the agent to indirect prompt injection through its web browsing capabilities.
  • Ingestion points: Untrusted web content is ingested via the open, snapshot, and get text commands as described in SKILL.md.
  • Boundary markers: There are no instructions or delimiters implemented to distinguish external web data from valid agent instructions.
  • Capability inventory: The skill provides the agent with the ability to navigate URLs, interact with form elements, capture screenshots, and manipulate network traffic.
  • Sanitization: No mechanisms for sanitizing or validating the content retrieved from external websites are documented.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 18, 2026, 03:40 AM
Security Audit — agent-trust-hub — agent-browser