The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches required AI models for transcription and subject matting. scripts/transcribe.cjs triggers model downloads via the WhisperX framework, and scripts/matte.cjs downloads the u2net_human_seg model to a local cache directory.
[REMOTE_CODE_EXECUTION]: Executes remote Python packages and scripts. scripts/transcribe.cjs uses uvx to run a specific version of whisperx for transcription. Additionally, various HTML templates include the GSAP animation library via the JSDelivr CDN.
[COMMAND_EXECUTION]: Heavily relies on command-line tools for core functionality. It spawns ffmpeg and ffprobe for video processing and uses puppeteer to conduct pixel-perfect layout measurements and overflow checks. scripts/render-theme.sh also contains logic to execute a local _postfx.sh script if provided within a project directory.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data that could theoretically contain malicious instructions. Evidence chain: 1. Ingestion points: transcript.json and project configuration files. 2. Boundary markers: Absent. 3. Capability inventory: Broad file-system access and shell command execution (child_process). 4. Sanitization: Includes HTML escaping for rendered text, though it can execute shell scripts provided in the project folder.

hyperframes-embedded-captions