The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted user prompts and convert them into structured code (React/Remotion) and configuration files.
Ingestion points: User prompts provided during conversation (SKILL.md Stage 0).
Boundary markers: Uses structured 'Ledgers' (Spec, Transformation, Element) to isolate and validate instructions before implementation.
Capability inventory: The skill instructs the agent to create/modify project files and execute shell commands (npm run render, npm run check, ffmpeg, ffprobe) if file/shell tools are available (SKILL.md Stage 6).
Sanitization: Includes a 'Design Jury' (Stage 6) with a specific ban list for generic or 'slop' content, which also functions as a self-review mechanism to ensure alignment with the initial brief.
[COMMAND_EXECUTION]: The skill includes a shell script (scripts/check.sh) and a Node.js script (scripts/score.mjs). These are used for legitimate project tasks (audio synthesis and video verification) and do not contain malicious patterns or command injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The package.json file references standard, well-known libraries from the official NPM registry (React, Remotion). No downloads from untrusted or obscure sources were detected.
[SAFE]: The skill follows professional development practices, including strict constant management in spec.ts and modularized rendering logic in motion.tsx.

motion-design-prompt-generator