HeyGen Skills - Setup Guide¶

Source: heygen-com/skills via skills.sh - 20.2K combined installs across 15 indexed listings; first documented November 3, 2026 (evening sweep) GitHub: 538 stars, license MIT, last pushed 2026-07-14, 4 SKILL.md directories Category: Avatar video for agents: creator avatars, scripted video generation and translation through the HeyGen v3 pipeline Quality Tier: 🟡 Yellow - MIT, 538 stars and a genuine install spec, but only one of the five sampled listings passes all three auditors and the other four carry Snyk warnings

HeyGen publishes this repository as AI agent skills for creating avatar videos, described in the README as giving an agent a face, a voice, and the ability to send video like a message. It works with Claude Code, OpenClaw, Codex, Cursor and other coding agents.

The tree ships three skills, heygen-avatar, heygen-video and heygen-translate, plus a fourth SKILL.md under platforms/nanoclaw/heygen for that platform. skills.sh indexes 15 listing names against those four directories, so most of the table below is listing names without a folder of their own. The README is explicit that the repository ships three skills and that the three are auto-discovered at heygen-avatar/SKILL.md, heygen-video/SKILL.md and heygen-translate/SKILL.md after a clone.

Installation¶

The README's preferred path is to paste a prompt into the agent rather than run commands, because the install spec lives in the repository:

Read https://raw.githubusercontent.com/heygen-com/skills/master/INSTALL_FOR_AGENTS.md and follow it.
Ask me for any API keys you need.

That document is what clones the repository, asks for the key, picks a transport, runs a verify test and ends with a working video. The manual routes the README lists alongside it:

gh skill install heygen-com/skills heygen-avatar
gh skill install heygen-com/skills heygen-video
gh skill install heygen-com/skills heygen-translate

gh skill install requires GitHub CLI v2.90 or newer and writes to the directory your agent reads. The README's other three routes are ClawHub (clawhub install heygen-skills), an OpenClaw plugin that bundles MCP support (openclaw plugins install clawhub:@heygen/openclaw-plugin-heygen), and a plain clone into ~/.claude/skills/heygen-skills or ~/.openclaw/workspace/skills/heygen-skills. For a Hermes profile, the clone route maps onto the profile skills directory:

git clone --single-branch --depth 1 https://github.com/heygen-com/skills.git /tmp/heygen-skills
cp -R /tmp/heygen-skills/heygen-avatar /tmp/heygen-skills/heygen-video /tmp/heygen-skills/heygen-translate ~/.hermes/profiles/corpusiq/skills/

Three steps follow the clone. Generate an API key at app.heygen.com/api, install the CLI with curl -fsSL https://static.heygen.ai/cli/install.sh | bash, then export HEYGEN_API_KEY in the shell or in a shell profile so it survives the session. The README verifies with heygen --version and heygen auth status, and notes that the CLI is a single static binary with no runtime dependencies. It also documents a remote MCP alternative that uses OAuth and needs no API key when a HeyGen plan with credits is already in place; the skills auto-detect MCP when no key is set.

What It Provides¶

Skill	Installs	Use For
heygen-video	3,133	Scripted video generation through the v3 Video Agent pipeline; one of the three shipped skill directories.
heygen-avatar	2,736	Building the agent's own creator avatar from identity files, then reusing it across videos; a shipped directory.
heygen-translate	2,140	Translating and re-voicing an existing video into another language; a shipped directory.
heygen	1,655	Listing name for the HeyGen platform surface. No directory of its own; the shipped path is the `platforms/nanoclaw/heygen` tree.
text-to-speech	1,189	Speech synthesis as a standalone listing. No directory in the tree; voice work runs through the video and avatar skills.
video-translate	1,154	Second translation listing alongside heygen-translate, which is the shipped directory.
video-understand	1,088	Understanding an uploaded video. No directory in the tree and no matching capability in the README's three-skill list.
video-edit	1,029	Editing an existing video. No directory of its own; the nearest shipped skill is heygen-video.
avatar-video	994	Avatar video as a combined listing. The two shipped halves are heygen-avatar and heygen-video.
ai-video-gen	987	Generic generation listing with no directory; heygen-video is the shipped entry point for generation.
create-video	945	Another generation alias with no directory of its own.
video-download	930	Retrieving a finished render. No directory in the tree; the README's flow ends by delivering a share link.
visual-style	921	Style selection for a render. The README describes the video skill picking the best style, with no style skill of its own.
faceswap	903	Face replacement listing with no directory in the tree and no mention in the README's skill list.
heygen-skills	381	The ClawHub package name for the whole set, not a capability; it is the string `clawhub install` takes.

All 15 indexed listings appear in the table above.

The gap between 15 listing names and four directories is the main thing to read here. Three of the fifteen, heygen-video, heygen-avatar and heygen-translate, are real skill folders that the README documents and auto-discovers. heygen-skills is the bundle name. The remaining eleven are index aliases around the same three capabilities, several in near-duplicate pairs: video-translate against heygen-translate, avatar-video against heygen-avatar, and create-video, ai-video-gen and heygen-video all pointing at generation.

Why This Matters for Hermes Agents¶

A link is not a deliverable in most agent workflows, and video is the format that changes that. This pack turns a script into a rendered file with a face and a voice, which is the difference between an agent that drafts a launch announcement and one that ships it. The README's framing is blunt about the intended use: send video like a message.

The install pattern is worth stealing even if you never make an avatar. One prompt tells the agent to fetch INSTALL_FOR_AGENTS.md and follow it, and that file holds the clone path, the key prompt, the transport choice and a verify step. The install logic lives in the repository, so it can be corrected without asking every user to re-paste new instructions. For a builder shipping skills to other agents, that is a better contract than a README full of commands that go stale.

The three skills divide cleanly by job. heygen-avatar builds the identity asset once, heygen-video turns a script into a render, and heygen-translate re-voices an existing video into another language. An agent can plan those as three steps with an artifact between each, which is far easier to retry than one opaque generate call.

Two transports are supported, and the choice matters operationally. An API key is a static secret in the environment; the MCP route is OAuth against an existing plan. The skills auto-detect MCP when no key is present, so a profile can move between the two without editing skill files.

Usage¶

You say	What happens
"Create your own avatar and make a 30-second intro video"	heygen-avatar reads the agent's identity files, asks only what is missing and builds the avatar, then heygen-video renders the intro.
"Turn this blog post into a two-minute avatar video"	heygen-video takes the script through the v3 pipeline, generates the render and returns a share link.
"Dub this video into Spanish with my own voice"	heygen-translate translates and re-voices the existing video instead of regenerating it.
"Make a video of me presenting this deck"	heygen-avatar supplies the creator avatar and heygen-video drives the script and delivery.
"Check whether my HeyGen key is working"	The agent runs `heygen auth status` and reports the transport in use, API key or MCP.
"Use my HeyGen plan over OAuth, I have no API key"	The skills detect no key, fall back to the remote MCP server and authenticate through OAuth.
"Generate the same video in English and German"	heygen-video renders the source, then heygen-translate produces the German variant from it.

Verification¶

The raw path below is built from the layout in this pack and returned 200 on branch master when this guide was written on 2026-11-03. Note the branch, which is master for this repository rather than main:

curl -sL -o /dev/null -w "%{http_code}\n" \
  "https://raw.githubusercontent.com/heygen-com/skills/master/heygen-video/SKILL.md"

After a clone, confirm the directory count and the install spec the README tells agents to fetch, then check the CLI and the credential separately:

git clone --single-branch --depth 1 https://github.com/heygen-com/skills.git /tmp/heygen-skills
find /tmp/heygen-skills -name SKILL.md | wc -l
ls /tmp/heygen-skills/INSTALL_FOR_AGENTS.md
heygen --version && heygen auth status

The count should be 4, matching the pack layout: the three top-level skills plus platforms/nanoclaw/heygen. The second command confirms the install spec exists, because the paste-in install prompt fails silently if that file is missing. The last line separates the two failure modes that look identical in a render error: a CLI that is not installed, and a CLI that is installed but unauthenticated.

Security¶

skills.sh verdicts for sampled skills (verified November 3, 2026) - re-check the security pages on skills.sh before production use:

Skill	Gen Agent Trust Hub	Socket	Snyk
heygen-video	Pass	Pass	Pass
heygen	Pass	Pass	Warn
video-edit	Pass	Pass	Pass
video-download	Pass	Pass	Warn
heygen-skills	Pass	Pass	Warn

One of the five sampled listings passes all three auditors. heygen-video, the skill that does the actual generation, is the clean one, which is reassuring for the core path. The three warnings cluster on heygen, video-download and heygen-skills, and the first of those is the listing with no directory of its own, so a warning on it does not map to a folder you can read.

The install path deserves its own note. The README asks you to paste a prompt that tells an agent to fetch a remote Markdown file and follow it, which is convenient and also the widest grant in this pack: the agent clones a repository, asks for a key and runs a verify. A remote install spec can change after you read it, so pin it if that matters to you, or take the manual route and read the three SKILL.md files before installing.

The credential is a real one. A HeyGen API key bills a paid account, and the README's own instruction is to export it into a shell profile, where it persists for every later session. The MCP route exists precisely to avoid that: OAuth against the plan, no long-lived key in the environment. Either way, video generation spends credits, so an agent with this skill and no budget ceiling can spend real money one render at a time.

Integration with CorpusIQ¶

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

Limitations¶

Snapshot data, verified November 3, 2026: 20,185 combined installs across 15 indexed listings; 538 GitHub stars; license MIT; last pushed 2026-07-14. Counts drift over time.
15 indexed listing names behind four skill directories. Only heygen-video, heygen-avatar and heygen-translate are documented, auto-discovered skills, and several listings are near-duplicates of one another.
video-understand and faceswap have no directory in the tree and no entry in the README's three-skill list. Treat them as index artifacts rather than capabilities until the repository shows otherwise.
The paste-in install prompt fetches a remote spec from the master branch, so the commands it runs are not fixed at the moment you read the README.
Avatar video spends account credits. The README documents no budget guard, and the CLI auth check tells you the key works, not what it will cost.
Face and voice generation is a likeness and consent question before it is a technical one. Nothing in the repository states a consent policy for building a face from identity files.

Skills Catalog | Skills Marketplace

Pro Tips¶

Start with the MCP route if you already pay for HeyGen. It authenticates by OAuth and keeps a billable API key out of your shell profile.
Run heygen auth status before you debug a failed render. Auth failures and generation errors return the same kind of message, and the status call settles which one you have.
Build the avatar once and reuse it. heygen-avatar is the slow step, and rendering from an existing avatar is what makes a five-video series practical.
Translate rather than regenerate when you need another language. heygen-translate works from the finished video, so the picture and timing stay fixed while the audio changes.
Copy only the three skill folders into a profile. The platforms/nanoclaw tree is for that platform, and pulling it in adds a duplicate heygen skill you will not call.

← Skills Catalog | Skills Marketplace → Powered by CorpusIQ