The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands using ffmpeg and ffprobe to extract audio, retrieve video metadata (duration, resolution, FPS), and re-encode video files for frame-accurate seeking.
[EXTERNAL_DOWNLOADS]: Uses npx to invoke the hyperframes CLI, which facilitates the download and execution of packages from the NPM registry. It also includes a self-update mechanism: npx hyperframes skills update talking-head-recut.
[DYNAMIC_EXECUTION]: The core workflow involves the agent dynamically generating HTML fragments and a master index.html file containing GSAP (GreenSock Animation Platform) JavaScript logic. This code is executed at runtime within a headless browser during the hyperframes render process.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect injection because it processes transcript.json (ASR output from potentially untrusted video files) to drive the content and timing of graphic overlays.
Ingestion points: transcript.json containing per-word timestamps and text.
Boundary markers: No explicit delimiters or instructions are used to separate the transcript data from the agent's code generation logic.
Capability inventory: The skill can execute shell commands (ffmpeg, npx), write files to the local workspace, and perform network operations via the hyperframes CLI.
Sanitization: The agent interprets the transcript to generate HTML, but there is no programmatic sanitization of the transcript text before it is used to populate card templates.

talking-head-recut