The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use npx hyperframes to perform various media processing tasks.
Evidence includes commands like npx hyperframes tts, npx hyperframes transcribe, and npx hyperframes remove-background.
[EXTERNAL_DOWNLOADS]: The skill documentation notes that the CLI tool automatically downloads machine learning models (Kokoro, Whisper, and u2net) to a local cache folder during the first run.
Evidence: Models are cached in ~/.cache/hyperframes/ and include weights for Kokoro (~338 MB), Whisper (up to 3.1 GB depending on model size), and u2net (~168 MB).
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes untrusted media files, subtitle formats, and text inputs to generate outputs.
Ingestion points: The transcribe and tts commands ingest files such as audio.mp3, video.mp4, subtitles.srt, subtitles.vtt, openai-response.json, and script.txt (found in SKILL.md).
Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore embedded instructions in the processed media metadata or text files.
Capability inventory: The skill uses npx for command execution and accesses the local file system for reading/writing media assets (found in SKILL.md).
Sanitization: No sanitization or validation of the contents of the processed media or text files is described.

hyperframes-media