extract
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill is configured to download website assets, including favicons (
scripts/crawl.mjs), images, and font files (SKILL.mdPhase 2), from the target domains being crawled to populate the localstardust/current/assets/directory. - [COMMAND_EXECUTION]: The setup procedure and operational phases involve running Node.js scripts (
scripts/crawl.mjs,scripts/style-census.mjs,scripts/thumb.mjs) and potentially installing necessary NPM packages (playwright,pixelmatch,pngjs,cheerio) to facilitate website rendering and analysis. - [INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection as it processes untrusted data from external websites.
- Ingestion points:
scripts/crawl.mjs(via sitemap XML parsing and HTML link extraction). - Boundary markers: Absent in the raw text extraction; however, the
reference/brand-review-template.mdfor reports enforces structural boundaries by embedding only CSS and forbidding external JavaScript. - Capability inventory:
scripts/crawl.mjs(network operations, file writes),scripts/style-census.mjs(network operations, file writes), andscripts/thumb.mjs(file writes). - Sanitization:
scripts/crawl.mjsimplements visibility filters and an interstitial/error heuristic to exclude transient DOM elements and common overlays from the captured content. - [DYNAMIC_EXECUTION]: The
scripts/style-census.mjsscript utilizes dynamicimport()to load itslive-session.mjsdependency from relative local paths, ensuring modularity while limiting loading to known internal script locations.
Audit Metadata