diff
Pass
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill facilitates the execution of Node.js-based scripts (
content-diff.mjsandvisual-diff.mjs) to perform automated browser testing. These scripts utilize Playwright to launch and control browser instances for page analysis. - [EXTERNAL_DOWNLOADS]: The tool navigates to and renders external web pages provided as arguments to compare prototypes and builds. It includes sophisticated browser configuration (UA spoofing, header management, and stealth modes) to handle sites protected by bot-management services like Akamai or Cloudflare, ensuring successful testing of production-like environments.
- [DATA_EXPOSURE]: The skill captures screenshots and extracts structural data (headings, CTAs, colors) from web pages. These results are stored locally in the
qa/vdiffdirectory and printed to the console for the user to review. No unauthorized data exfiltration patterns were identified. - [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests and processes untrusted text content from external web pages.
- Ingestion points: The
grabandcapturefunctions in the script directory ingest data from target URLs viapage.goto. - Boundary markers: Data is structured into inventories before being summarized, providing a layer of separation, though no explicit 'ignore instructions' markers are used in the summary output.
- Capability inventory: The skill has capabilities for local file writing (screenshots) and further network requests (Playwright navigation).
- Sanitization: Content is normalized for comparison purposes (whitespace and punctuation handling) but not specifically filtered for natural language instructions. The risk is evaluated as safe/low given the tool's specialized analytical purpose.
Audit Metadata