vibium-browser-automation

Pass

Audited by Gen Agent Trust Hub on Mar 16, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to the way it processes external web content.
  • Ingestion points: Untrusted data from web pages is ingested via the browser_navigate tool and converted to text through screenshots processed by easyocr, as detailed in references/easyocr-workflow.md and references/limitations.md.
  • Boundary markers: The skill's instructions and examples do not include delimiters or specific instructions (e.g., "ignore embedded instructions") to prevent the agent from obeying malicious commands found within the extracted text.
  • Capability inventory: The skill provides high-privilege interaction tools such as browser_click, browser_type, and browser_navigate (defined in SKILL.md), which could be leveraged by an attacker who successfully injects instructions into the processed content.
  • Sanitization: There is no evidence of sanitization, escaping, or filtering of the OCR-extracted text before it is analyzed by the agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 16, 2026, 06:16 PM
Security Audit — agent-trust-hub — vibium-browser-automation