The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from user-provided narration text and processes word-level timestamps returned from external transcription APIs.
Ingestion points: User-controlled text in audio_request.json and transcription output from api.transcribeAudio in scripts/lib/tts.mjs.
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are implemented within the narration processing pipeline.
Capability inventory: The skill can execute ffprobe via spawnSync, perform file system writes in defined project asset directories, and make outbound network requests to configured AI services.
Sanitization: The engine validates line IDs against a safe alphanumeric regex and filters transcription word objects for consistent timing data.
[COMMAND_EXECUTION]: The script scripts/audio.mjs and its helpers use node:child_process.spawnSync to execute ffprobe for audio stream validation. While the arguments are programmatically constrained, the skill relies on the presence of this binary in the system's PATH.
[EXTERNAL_DOWNLOADS]: The skill is configured to interact with well-known services including fish.audio and openai.com for media generation tasks. It also provides documentation for integrating with Pexels, Pixabay, and Unsplash APIs.

hyperframes-media