The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands for standard build and verification workflows. In SKILL.md, it instructs the user to run pnpm run brand:audit and npx tsx to execute the verification script. These are routine operations for a developer-oriented tool.
[EXTERNAL_DOWNLOADS]: The skill uses external AI services for video verification. In scripts/verify-composition.ts, the code sends rendered video data to Google's Generative AI (Gemini) API for motion and pacing evaluation. This is a documented feature and requires the user to provide their own API key via environment variables.
[DYNAMIC_EXECUTION]: The verification script scripts/verify-composition.ts utilizes dynamic loading to interact with the local Remotion installation. It uses createRequire to resolve and import @remotion/bundler and @remotion/renderer from the project's node_modules. This approach ensures the verification script uses the same version of Remotion as the project it is analyzing.
[DATA_EXFILTRATION]: While the script sends video data to an external API (Google Gemini), this is restricted to the video content being verified and is a core functional requirement for the 'Layer 3' animation flow evaluation. No sensitive local files (like SSH keys or credentials) are accessed or transmitted.
[PROMPT_INJECTION]: The prompts provided for AI-assisted evaluation (prompts/motion-flow-evaluation.md and prompts/static-frame-evaluation.md) use clear, instructional language to guide the AI in performing objective brand and motion audits. They do not contain instructions to bypass safety filters or ignore system constraints.

motion-video