localhost-screenshots

Pass

Audited by Gen Agent Trust Hub on Sep 16, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill captures page content (ARIA snapshots, DOM, text) which can contain instructions intended to influence the agent.
  • Ingestion points: Structured page data is captured in assets/scripts/screenshot-a11y.js and references/ai-snapshots.md via ariaSnapshot and DOM serialization.
  • Boundary markers: The skill uses an untrusted-page-content JSON envelope and HTML comment headers to explicitly delimit captured data from instructions.
  • Capability inventory: The skill possesses file-writing capabilities (fs.writeFileSync), screenshot capture, and browser navigation across all helper scripts.
  • Sanitization: DOM snapshots in references/ai-snapshots.md include logic to strip executable handlers and elements such as <script>, onclick, and iframe before persistence.
  • [EXTERNAL_DOWNLOADS]: The skill requires the installation of the Playwright browser automation framework and its associated Chromium binary.
  • Evidence: SKILL.md and references/playwright-patterns.md provide instructions to install playwright@1.62.0 and its browser dependencies. These are standard tools from a well-known service.
  • [COMMAND_EXECUTION]: The skill utilizes shell commands to manage dependencies, serve local files, and execute helper scripts.
  • Evidence: Instructions are provided for npm install, npx serve, and running Node.js scripts like assets/scripts/multi-breakpoint.js. The skill includes explicit warnings against using elevated privileges (sudo) for these operations.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 16, 2026, 07:52 AM
Security Audit — agent-trust-hub — localhost-screenshots