creativeclaw-render-html-video
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references and downloads external resources, including the GSAP animation library from JSDelivr (
cdn.jsdelivr.net) and Three.js. These are well-known, trusted services for web development.\n- [INDIRECT_PROMPT_INJECTION]: The skill architecture involves processing external code (HTML, CSS, JS) and full project ZIP archives, which creates a potential surface for indirect prompt injection if those external files contain malicious instructions.\n - Ingestion points: Data is ingested through
get_example(example catalog) andproject_url(user-supplied or remote ZIP files).\n - Boundary markers: The skill documentation explicitly warns the agent in
references/project-packaging.mdandreferences/example-discovery.mdto treat source instructions and build scripts as untrusted data and avoid local execution.\n - Capability inventory: The agent can write/modify local ZIP files and initiate remote rendering jobs via the
render_html_videotool.\n - Sanitization: The skill notes that the remote HyperFrames renderer implements safety controls such as disabling npm lifecycle scripts and restricting network access to a whitelist of package registries and CDNs.
Audit Metadata