website-to-video
Pass
Audited by Gen Agent Trust Hub on Nov 14, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill relies on the
hyperframesCLI tool, which is executed vianpx. This involves downloading and executing packages from the NPM registry at runtime. It also references external libraries such as GSAP and Three.js from well-known CDNs like jsdelivr. - [COMMAND_EXECUTION]: The instructions call for the execution of various shell commands to perform its core functions. These include
npx hyperframes capturefor site ingestion,npx hyperframes ttsandtranscribefor audio processing, andnpx hyperframes renderfor video production. It also utilizescurlto interact with external APIs (ElevenLabs and HeyGen). - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-provided URLs using the capture pipeline. This data (extracted text, asset descriptions, etc.) is subsequently processed by the AI agent to inform the video strategy and script. This represents an indirect prompt injection surface where malicious content on a target website could attempt to influence the agent's creative or logical output.
- [CREDENTIALS_UNSAFE]: The skill manages several API keys (Gemini, ElevenLabs, HeyGen). While it follows best practices by recommending the use of
.envfiles, it also contains instructions to accept keys directly from user input in chat if provided, which increases the risk of credential exposure in chat history or logs. - [DYNAMIC_EXECUTION]: During the build phase (Step 5), the skill generates local HTML files containing JavaScript (GSAP timelines) which are then snapshotted and validated by the agent. This involve script generation based on agent-composed logic.
Audit Metadata