video-edit

Fail

Audited by Gen Agent Trust Hub on Oct 16, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill uses npx --yes hyperframes@0.8.34 to download and execute the hyperframes rendering engine at runtime. It also performs automated package installation using pip to set up a Python virtual environment containing faster-whisper, av, and pillow.
  • [EXTERNAL_DOWNLOADS]: During its setup phase, the skill downloads sound effects (MP3 files) from a remote GitHub repository (heygen-com/hyperframes). It also triggers the download of large machine learning models for speech transcription (Whisper models) when first invoked.
  • [COMMAND_EXECUTION]: The skill makes extensive use of subprocess.run and subprocess.Popen to execute system tools such as ffmpeg, ffprobe, node, and npx, as well as its own suite of Python scripts for video processing, transcription, and QA.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input in the form of user-provided video clips and external style reference links (Instagram/Pinterest). Transcripts generated from these clips are used to rebuild the script and generate captions, creating a surface where maliciously crafted audio content could potentially influence agent behavior.
  • Ingestion points: Raw video files provided by the user and external URLs for style references processed in scripts/ingest.py.
  • Boundary markers: Absent. No specific delimiters or safety warnings are used to isolate transcription data from the agent's instructional context.
  • Capability inventory: The skill can execute shell commands (ffmpeg, npx), write files to the local file system (project assets, rendering scripts), and perform network operations to download assets and models.
  • Sanitization: Absent. Transcription output is directly incorporated into the project's edit decision list (EDL) and captioning metadata.
  • [DYNAMIC_EXECUTION]: The skill generates and adapts build.py scripts based on project-specific metadata and look requirements. These scripts, which contain animation logic, are dynamically written to the project folder and subsequently executed to drive the rendering process.
Recommendations
  • HIGH: Downloads and executes remote code from: unknown (check file) - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 16, 2026, 03:20 PM