swatches
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a local Node.js script (
scripts/sheet.mjs) to automate the process of photographing HTML design options. - [COMMAND_EXECUTION]: The skill uses
npx viteto serve the generated swatch files on a local port (5173by default) for user interaction. - [EXTERNAL_DOWNLOADS]: The skill utilizes
npxto fetch and runvite, a well-known and trusted development server tool. This is a standard practice in modern web development workflows. - [COMMAND_EXECUTION]: The
sheet.mjsscript spawns a headless Chrome or Chromium process to render the generated HTML files and take screenshots. It includes logic to locate the browser in standard system paths and Playwright cache directories. - [DATA_EXPOSURE]: The skill reads from a local
.swatches/picks.mdfile to incorporate previous user decisions into the current design round. This is a legitimate use of local context to improve agent performance. - [INDIRECT_PROMPT_INJECTION]: The skill ingests data from
.swatches/picks.mdand reads metadata from agent-generated HTML files. While this represents an attack surface for instructions embedded in data, the script includes basic HTML entity escaping for captions to mitigate risks during the generation of the comparison sheet.
Audit Metadata