The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingestion surface involves user-provided video files and automatically generated transcripts. While this represents a potential vector, the risk is addressed through HTML entity escaping (sanitization) in the rendering scripts and the fact that transcript data is used exclusively for time-synced caption display. 1. Ingestion points: scripts/transcribe.cjs. 2. Boundary markers: Absent. 3. Capability inventory: Local execution of ffmpeg, node scripts, and puppeteer. 4. Sanitization: scripts/make-composition.cjs escapes HTML tags via the escBr function.
[EXTERNAL_DOWNLOADS]: The skill manages the installation of legitimate dependencies (sharp, puppeteer, gsap) from the NPM registry and the execution of the whisperx transcription tool via uvx. These operations target well-known services and are necessary for the skill core functionality.
[DYNAMIC_EXECUTION]: The workflow involves using Headless Chromium to load and animate local HTML compositions for video frame generation, which is a standard part of the build and render process for this use case.

embedded-captions