The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes transcript data from external sources (Whisper JSON, SRT, VTT) to drive automated caption authoring and animation generation. Maliciously crafted transcripts could attempt to influence the agent's logic during the styling and grouping phases.
Ingestion points: references/captions/authoring.md and references/captions/transcript-handling.md specify reading external transcript files.
Boundary markers: The instructions lack explicit boundary markers or system prompts to ignore embedded instructions in the transcript data.
Capability inventory: The skill utilizes spawnSync for ffmpeg operations in scripts/heygen-tts.mjs and facilitates the generation of browser-based animation code.
Sanitization: references/captions/transcript-handling.md provides a cleaning function to remove non-word tokens (e.g., music symbols and short filler words).
[COMMAND_EXECUTION]: The utility script scripts/heygen-tts.mjs executes the ffmpeg system command to perform audio transcoding.
Evidence: scripts/heygen-tts.mjs calls spawnSync to invoke ffmpeg for converting MP3 responses from the TTS provider into WAV format.
[EXTERNAL_DOWNLOADS]: The skill fetches several machine learning models and configuration files from external repositories and well-known services.
Evidence: references/requirements.md documents downloads for Whisper models, Kokoro-82M, MusicGen, and U2Net human segmentation models from sources including Hugging Face. These are well-known resources in the machine learning ecosystem.

hyperframes-media