browser-automation-skill

Fail

Audited by Gen Agent Trust Hub on May 11, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The README.md file directs users to download and install the 'OpenClaw CLI' from https://openclawcli.vercel.app/, a domain that has been explicitly flagged as a phishing risk by security scanners.
  • [REMOTE_CODE_EXECUTION]: The SKILL.md instructions require the installation of an external NPM package 'agent-browser' and the execution of initialization scripts ('agent-browser install --with-deps'). This package is not linked to the skill author's verified infrastructure and originates from a project associated with a flagged malicious domain.
  • [COMMAND_EXECUTION]: The skill promotes the execution of high-privilege system commands including global npm installations and environment setup scripts that can execute arbitrary code on the host machine without integrity verification.
  • [PROMPT_INJECTION]: The skill provides a significant surface for indirect prompt injection as it is designed to ingest and interact with untrusted third-party web content (HTML, interactive elements) without using boundary markers or content sanitization.
  • Ingestion points: 'agent-browser open' and 'agent-browser snapshot' commands (SKILL.md)
  • Boundary markers: Absent; instructions do not include delimiters or warnings for the agent to ignore instructions embedded in page content.
  • Capability inventory: High-risk capabilities including 'agent-browser click', 'agent-browser fill', 'agent-browser state save', and 'agent-browser upload' (SKILL.md)
  • Sanitization: Absent; the skill lacks any mechanisms to escape or validate data retrieved from external URLs before processing.
Recommendations
  • AI detected serious security threats
  • Contains 2 malicious URL(s) - DO NOT USE
Audit Metadata
Risk Level
HIGH
Analyzed
May 11, 2026, 10:49 AM
Security Audit — agent-trust-hub — browser-automation-skill