motion-design-html
Warn
Audited by Gen Agent Trust Hub on Nov 20, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill extensively uses the Python
subprocessmodule to execute system-level commands for media processing and rendering. scripts/analyze_song.pyrunsffmpegto analyze audio energy and estimate BPM.scripts/audio_template.pyandscripts/remake/remake_sync.pyexecuteffmpegfor audio muxing, volume normalization, and video encoding.scripts/render_template.pyandscripts/remake/remake_render.pylaunch Chromium viaplaywrightto capture frames and useffmpegto assemble them into video files.- [CREDENTIALS_UNSAFE]: The client script for the 21st.dev service retrieves a secret key from a hardcoded local file path.
scripts/mcp21_client.pyattempts to read a key from~/.config/21st.keyto use as anx-api-keyheader in network requests.- [EXTERNAL_DOWNLOADS]: The skill downloads various media assets and design components from external domains.
- Fetches music and sound effects from Mixkit (
assets.mixkit.co). - Downloads stock photos from Unsplash and Pexels APIs.
- Retrieves SVG brand logos from the
svgl.appAPI. - Interacts with the 21st.dev API to search for and retrieve UI component code.
- [INDIRECT_PROMPT_INJECTION]: The skill implements a complex multi-agent workflow where outputs from one agent (like shot scripts or beat maps) are consumed as instructions by subsequent agents.
- Data ingested from external sources (API responses, metadata) is interpolated into scripts that are then executed or read by other agents.
- The instructions recommend using boundary markers (e.g., "Example data" labels) and a critique loop to mitigate errors, but the automated nature of the pipeline presents a surface for instruction leakage.
Audit Metadata