The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data sources, including .srt subtitle files and video/audio media files, to generate caption objects. If these external files contain malicious instructions, they could potentially influence the agent's behavior during the transcription or parsing process.
Ingestion points: import-srt-captions.md (fetches subtitles via fetch() or staticFile()) and transcribe-captions.md (reads media files from the public/ directory).
Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore potential commands embedded in the caption text.
Capability inventory: The skill utilizes file system writes (node:fs/promises), network requests (fetch), and package installation commands (npx).
Sanitization: While the skill uses a dedicated SRT parser (parseSrt), it does not implement specific sanitization for AI-targeted prompt injection attacks.
[DYNAMIC_EXECUTION]: The skill provides boilerplate Node.js code in transcribe-captions.md for users or agents to create and run local scripts that perform audio decoding and transcription. This is a standard developer workflow for this framework.
[EXTERNAL_DOWNLOADS]: The skill involves several legitimate external resource fetches:
Installs framework-specific packages (@remotion/captions, @remotion/whisper-webgpu) and dependencies (@huggingface/transformers, mediabunny) from the npm registry.
Downloads pre-trained Whisper models from Hugging Face for local GPU-accelerated transcription.
Fetches component source code and documentation from the official Remotion website (www.remotion.dev).

remotion-captions