heygen-avatar

This skill creates and manages persistent HeyGen avatars: reusable face-plus-voice identities for the agent itself, the user, or named characters, powered by HeyGen's Avatar V technology. It solves the setup step that must precede video generation by establishing a consistent identity, handling identity extraction, avatar generation, and voice selection, and saving the result to an AVATAR-<NAME>.md file so the same look and voice can be reused across videos. Creation is prompt-based by default (a description drives generation), with optional photo upload for real-person digital twins. It returns an avatar_id and voice_id intended to be passed to a companion heygen-video skill.

Its capabilities include language awareness (detecting the user's language and communicating in it while keeping API enum values in English), voice design via semantic search over HeyGen's voice library (which consumes no quota and generates no new voices) or manual catalog browsing as a fallback, playable voice previews before committing, and role aliasing through AVATAR-AGENT and AVATAR-USER symlinks. The skill is explicit about file access: it reads SOUL.md, IDENTITY.md, and existing AVATAR files, writes new AVATAR files, uses a temp path for downloaded voice previews that are deleted after the session, and uploads user-provided photos to HeyGen only when the user explicitly supplies them. A troubleshooting reference covers issues like unnatural pauses in generated videos.

It targets creators and users who want to appear in or produce HeyGen videos, give the agent a presenter identity, or build named character presenters. It declares Bash, WebFetch, Read, Write, and the HeyGen MCP tools; uploads go only to HeyGen and only with explicit user consent.