The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on subprocess.run to execute command-line tools for video and audio processing. This is a core functional requirement for the skill's purpose.
Evidence: Multiple scripts, including templates/normalize_all.py, templates/cut_segments.py, and templates/build_final.py, execute ffmpeg and ffprobe. The script templates/gfx/render_gfx.py executes npx hyperframes to render graphics.
[EXTERNAL_DOWNLOADS]: The skill fetches resources from well-known and reputable services to support its rendering and transcription features.
Evidence:
HTML templates in templates/gfx/compositions/ download the GSAP animation library from the jsdelivr CDN.
The templates/transcribe_all.py script utilizes the faster-whisper library, which fetches pre-trained models from the Hugging Face Hub.
The graphics pipeline uses npx to execute tools from the hyperframes ecosystem.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data (audio from uploaded video clips) which is transcribed and then used by the agent to structure the video edit, creating a surface for indirect prompt injection.
Ingestion points: Raw audio is ingested and converted to text via templates/transcribe_all.py.
Boundary markers: The instructions do not define specific delimiters or guardrails to distinguish between user audio content and task instructions during the weaving phase.
Capability inventory: The skill has extensive capabilities to execute shell commands (ffmpeg, npx) and modify the local file system.
Sanitization: No sanitization or filtering of the transcribed text is implemented before the agent processes it for thematic weaving.

blessing-video-weaver