product-motion-ads
Pass
Audited by Gen Agent Trust Hub on Nov 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple external sources to plan and build video content.
- Ingestion points: The workflow involves reading product source code (main pages, styles, routes), website captures, screenshots, creator briefs, and physical product spec sheets (SKILL.md, Workflow Step 1).
- Boundary markers: No explicit delimiters or instructions to ignore embedded commands are defined for the agent when it processes these external inputs.
- Capability inventory: The skill possesses extensive capabilities, including executing shell commands (
ffmpeg,hyperframes), running Node.js scripts, and performing network operations such as uploading content to YouTube (SKILL.md, Workflow Steps 4-8). - Sanitization: While the skill instructs the agent to replace secrets with stand-ins, there is no mention of sanitizing or escaping the data ingested from external code or briefs to prevent instructions within that data from being executed by the agent.
- [COMMAND_EXECUTION]: The workflow relies heavily on the execution of shell commands and local scripts to process media.
- The skill uses
ffmpegfor tasks like black frame detection, freeze detection, and complex audio/video mixing and filtering (SKILL.md, Workflow Steps 6 and 7). - The
hyperframesCLI tool is used for composition validation, generating snapshots, and performing the final video render (SKILL.md, Workflow Steps 5 and 6). - Several local Node.js scripts are executed as part of the workflow, including
new-edit.mjs,init-workspace.mjs, andsfx-sync.mjs(SKILL.md, Workflow Steps 4 and 7). - [DYNAMIC_EXECUTION]: The skill generates and executes code at runtime to create video content.
- The agent builds a HyperFrames composition consisting of HTML, CSS, and GSAP-based JavaScript. This code is dynamically generated based on the ad plan and then executed by the renderer to produce the final video frames (references/motion.md).
- [EXTERNAL_DOWNLOADS]: The skill interacts with external services and downloads additional resources.
- The command
hyperframes ttsis used to download voice models for local text-to-speech, and the skill also mentions using the external ElevenLabs API for voiceover generation (SKILL.md, Workflow Step 3). - The workflow includes instructions for uploading finished video files directly to YouTube (SKILL.md, Workflow Step 8).
Audit Metadata