The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's primary functionality involves downloading pre-trained AI models (Kokoro-82M, Whisper, and U2Net) upon first execution of the tts, transcribe, or remove-background commands. These models are cached locally in the ~/.cache/hyperframes/ directory.
[COMMAND_EXECUTION]: The instructions direct the agent to install Python dependencies (kokoro-onnx, soundfile) and execute CLI commands via npx hyperframes. These tools are authored by the skill's vendor and are required for the stated media processing tasks.
[PROMPT_INJECTION]: The skill describes an attack surface for indirect prompt injection as it processes untrusted external data.
Ingestion points: Processes external text files (script.txt), audio/video files (audio.mp3, video.mp4), subtitle files (.srt, .vtt), and JSON data (openai-response.json) in SKILL.md.
Boundary markers: Absent; there are no instructions for the agent to ignore or delimit embedded commands within the processed media content.
Capability inventory: Subprocess execution of CLI tools (npx hyperframes) documented throughout SKILL.md.
Sanitization: No sanitization or validation of the input media content or resulting transcription text is mentioned.

hyperframes-media