pixel-by-pixel

Warn

Audited by Gen Agent Trust Hub on Jul 23, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill performs dynamic loading of Node.js modules from paths computed at runtime. In scripts/lib/runtime-dependencies.mjs, it uses createRequire and resolve on strings assembled from environment variables and the current working directory to locate its internal dependency resolver.
  • [COMMAND_EXECUTION]: The skill executes external commands during setup and operation. In scripts/setup-runtime.mjs, it invokes npm install and the Playwright CLI to install the Chromium browser. Additionally, several scripts like scripts/run-fidelity-loop.mjs use spawn and execFile to coordinate the execution of multiple Node.js utility scripts included with the skill.
  • [EXTERNAL_DOWNLOADS]: During its setup process, the skill downloads the Chromium browser from official package repositories via Playwright. Its core functionality involves downloading HTML, CSS, and other media assets from arbitrary third-party URLs provided by the user for comparison and replication purposes.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted data from external websites.
  • Ingestion points: User-supplied URLs are processed in inspect-page.mjs, compare-pages.mjs, and bootstrap-static-replica.mjs.
  • Boundary markers: Access to external URLs is restricted via a custom validating proxy (validating-proxy.mjs) that blocks requests to private and reserved IP ranges (SSRF protection).
  • Capability inventory: The skill has capabilities to write files to the local system and execute internal scripts via spawn or execFile.
  • Sanitization: The skill actively sanitizes captured content by removing <script> and <iframe> tags and applies a restrictive Content Security Policy (CSP) to replicated pages.
  • [DATA_EXFILTRATION]: While not performing malicious exfiltration, the skill contains logic that could interact with sensitive data on targets. To mitigate risk, it includes a safety library (scripts/lib/network-safety.mjs) that identifies and redacts credential-like tokens (API keys, tokens, JWTs) and PII (emails, phone numbers) from generated reports and snapshots.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 23, 2026, 07:11 PM
Security Audit — agent-trust-hub — pixel-by-pixel