agent-browser

Warn

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • EXTERNAL_DOWNLOADS (MEDIUM): The skill instructions require the global installation of agent-browser via NPM (npm install -g agent-browser). This package is not hosted in a trusted repository or organization as defined in the security policy, making it an unverifiable dependency that could execute arbitrary code during installation or runtime.
  • PROMPT_INJECTION (LOW): The skill is vulnerable to Indirect Prompt Injection (Category 8) because it ingests untrusted data from external websites.
  • Ingestion points: The agent-browser snapshot command reads the DOM and accessibility tree of arbitrary websites and provides it to the agent context.
  • Boundary markers: There are no instructions or delimiters defined to separate web content from system instructions or to warn the agent to ignore instructions embedded in the page.
  • Capability inventory: The skill possesses significant capabilities, including form filling (fill), interaction (click, check), and authentication state management (state save/load).
  • Sanitization: No sanitization or filtering of the ingested web content is performed before processing.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 17, 2026, 06:31 PM