visual-review

Pass

Audited by Gen Agent Trust Hub on Jul 9, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill provides instructions to download and install the Playwright library and the Chromium browser using npx playwright install chromium. Playwright is a well-known, established tool maintained by Microsoft.
  • [COMMAND_EXECUTION]: The skill uses various shell commands for environmental setup (mkdir), browser management (npx playwright), and process control (kill, lsof). These are used within the scope of its visual review functionality.
  • [DATA_EXFILTRATION]: To allow users to view screenshots on remote machines, the skill instructs the agent to start a network listener using python3 -m http.server on a high port. While documented for user collaboration, this exposes the /tmp/visual-review/ directory to the network, which may contain sensitive UI details if used on private applications.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection.
  • Ingestion points: External web pages and local HTML files are captured as screenshots in SKILL.md (Phase 2).
  • Boundary markers: None. The agent is directed to directly read and analyze the captured visual content.
  • Capability inventory: The skill has capabilities for shell command execution, file system modification, and starting network listeners.
  • Sanitization: Absent. There is no mechanism to filter or sanitize malicious instructions that might be visually embedded in the captured screenshots.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 9, 2026, 01:18 AM
Security Audit — agent-trust-hub — visual-review