website-to-video
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill captures data from external websites via the
npx hyperframes capturecommand. This data (extracted text, asset descriptions, design tokens) is then used as primary context for the agent to generate scripts and storyboards. A malicious website could contain hidden instructions designed to manipulate the agent's output or behavior during the video generation process. - Ingestion points:
capture/extracted/visible-text.txt,capture/extracted/asset-descriptions.md, andcapture/extracted/tokens.json. - Boundary markers: Absent. The instructions direct the agent to read these files to derive brand identity and messaging without explicit delimiters or warnings to ignore embedded instructions.
- Capability inventory: The skill utilizes shell command execution (
npx,curl,python3), file system modifications (writingDESIGN.md,STORYBOARD.md, and HTML compositions), and network operations (API calls to HeyGen and ElevenLabs). - Sanitization: No sanitization or validation of the captured website content is mentioned.
- [COMMAND_EXECUTION]: The skill relies extensively on shell commands for its core functionality. User-supplied inputs, including URLs and narration scripts, are interpolated into commands such as
npx hyperframes capture <URL>andnpx hyperframes tts <text>. While these are part of the intended workflow, they represent a potential command injection surface if inputs are not properly handled by the underlying tools. - [CREDENTIALS_UNSAFE]: The skill instructions (e.g., in
references/step-4-vo.md) guide the user to provide and store sensitive API keys for Google Gemini, HeyGen, and ElevenLabs in.envfiles. Furthermore, it provides examplecurlcommands that pass these secrets as plaintext headers (e.g.,x-api-key,xi-api-key), which may be exposed in process monitoring logs or history files. - [EXTERNAL_DOWNLOADS]: The skill dynamically downloads and executes the
hyperframesCLI tool vianpxand usescurlto fetch audio assets from remote AI services (HeyGen and ElevenLabs). While these are well-known technology providers, the skill depends on the availability and integrity of these external endpoints for its runtime execution.
Audit Metadata