agent-browser
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the user to install an external global NPM package
agent-browserand run an installation script (agent-browser install) which typically downloads browser binaries (Chromium). - [DYNAMIC_EXECUTION]: The skill functions as a 'discovery stub' that does not contain the actual operational instructions. It directs the agent to dynamically load workflows, command references, and templates at runtime from a remote source using the
agent-browser skills getcommand. - [INDIRECT_PROMPT_INJECTION]: As a browser automation tool, the agent processes accessibility-tree snapshots and page content from arbitrary, untrusted websites. This content can contain adversarial instructions intended to hijack the agent's session.
- Ingestion points: Accessibility-tree snapshots and web page data in
SKILL.md(via CLI output). - Boundary markers: None provided in this discovery stub.
- Capability inventory: Subprocess execution via
Bash, session management, and access to an 'authentication vault'. - Sanitization: Not explicitly defined in the instruction stub.
- [COMMAND_EXECUTION]: The skill relies on shell command execution to interact with the browser CLI, manage sessions, and fetch remote skill content.
Audit Metadata