website-builder
Pass
Audited by Gen Agent Trust Hub on Jul 20, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads and executes the
create-next-apppackage from the NPM registry vianpxand uses the Vercel CLI for cloud deployment. - [COMMAND_EXECUTION]: The
scripts/build-and-deploy.shscript executes shell commands to create directories, generate project files via heredocs, and run thevercel deploycommand. - [PROMPT_INJECTION]: The skill autonomously researches trends on public platforms like GitHub, Reddit, and HackerNews to determine the project's purpose. This creates a surface for indirect prompt injection, where content retrieved from the web could influence the generated site's content.
- Ingestion points: Research data fetched via
web_searchandghCLI (SKILL.md). - Boundary markers: None identified; research results are interpolated directly into the project setup.
- Capability inventory: File system writes (
cat), command execution (npx), and network deployment (vercel). - Sanitization: No explicit sanitization of the research-derived "idea" string before inclusion in the website scaffold.
Audit Metadata