heygen-video

This skill generates HeyGen presenter (talking-head) videos through HeyGen's v3 Video Agent pipeline. It solves the problem of producing high-quality AI presenter videos by orchestrating the steps that raw API calls skip: aspect-ratio and framing correction ("Frame Check"), prompt engineering, avatar resolution, voice selection, and asset routing. It explicitly forbids calling deprecated v1/v2 endpoints directly, routing instead through MCP, the OpenClaw plugin, or the `heygen` CLI, and returns a shareable video URL plus a session URL for iteration.

The skill acts as a guided "video producer." It reads saved avatar identity files (AVATAR-<NAME>.md, plus role symlinks) to load an avatar's group and voice IDs, appends framing-correction notes to the prompt based on avatar orientation and background type, and classifies user-supplied assets into two paths — contextualize the content into the script, or upload the raw file to HeyGen for on-screen B-roll. It downloads voice preview audio to a temporary directory for playback, appends one JSON line per generated video to a local learning log, and follows strict UX rules (be concise, hide internal pipeline jargon, poll silently for completion). A companion `update-check.sh` script is explicitly opt-in and is not run automatically on invocation. Authentication uses a HEYGEN_API_KEY environment variable.

Target users are people creating personalized video messages, outreach, announcements, pitches, explainers, tutorials, or product demos with a human presenter. When a request also involves creating or designing an avatar, the skill chains to the companion heygen-avatar skill first. It is not for avatar creation, cinematic b-roll without a presenter, video translation, TTS-only output, or streaming avatars.