The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill guides the agent to install numerous official Remotion packages (e.g., @remotion/media, @remotion/captions, @remotion/three) and established third-party libraries like zod and mapbox-gl. These are fetched from standard public registries (NPM) and well-known services.
[COMMAND_EXECUTION]: Instructions include the use of npx for project scaffolding (create-video), local previewing (remotion studio), and rendering operations. It also utilizes FFmpeg for media processing tasks like trimming and silence detection via npx remotion ffmpeg.
[REMOTE_CODE_EXECUTION]: In rules/transcribe-captions.md, the skill describes using the @remotion/install-whisper-cpp utility to download and install the Whisper.cpp binary and associated models. This is a standard feature provided by the trusted vendor for local audio-to-text transcription.
[SAFE]: The skill exhibits potential surfaces for indirect prompt injection because it instructs the agent to fetch and process external data files such as JSON, SRT, and Lottie animations to drive video content.
Ingestion points: Dynamic fetching of data in rules/calculate-metadata.md, rules/compositions.md, rules/display-captions.md, and rules/lottie.md.
Boundary markers: None explicitly mentioned in the documentation for separating data from instructions.
Capability inventory: Shell command execution via Remotion CLI and FFmpeg, as well as file system write access for saving generated assets.
Sanitization: Not explicitly detailed; instructions focus on the functional implementation within the React/Remotion environment.
Context: These data ingestion patterns are fundamental to the primary purpose of dynamic video creation and are considered safe within the intended use-case of the framework.
[SAFE]: Sensitive information like API keys (Mapbox, ElevenLabs) are managed via environment variables and .env files, which is consistent with secure development best practices.

remotion-best-practices