talking-head-recut
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided video content to generate a transcript, which it then uses to design and populate graphic cards. This represents an attack surface where content in the video (e.g., spoken instructions) could potentially influence the agent's behavior during the design phase.\n
- Ingestion points: Video transcript processed in Step 4 and 5 of the workflow (transcript.json).\n
- Boundary markers: No explicit prompt delimiters or instructions are used to distinguish transcribed text from design commands.\n
- Capability inventory: Local file writing, system command execution (FFmpeg), and video rendering via the hyperframes CLI.\n
- Sanitization: The skill relies on the internal sanitization logic of the hyperframes rendering engine.\n- [DYNAMIC_EXECUTION]: The skill dynamically generates HTML fragments, CSS styles, and GSAP JavaScript animation timelines at runtime. These elements are assembled into a final composition (index.html) and rendered into an MP4 file using a headless browser environment.\n- [EXTERNAL_DOWNLOADS]: The skill includes instructions for a silent auto-update mechanism using
npx hyperframes skills update talking-head-recut. This command fetches and executes the latest version of the skill and its core dependencies from the NPM registry.\n- [COMMAND_EXECUTION]: The workflow involves several shell command executions, includingffprobefor metadata extraction,ffmpegfor audio extraction and video re-encoding, and variousnpx hyperframessubcommands for transcription and rendering.
Audit Metadata