HeyGen

Use this skill for HeyGen Direct API work: avatar and voice discovery, guarded avatar video generation from approved scripts, video translation with lip-sync, and polling asynchronous generation or translation status.

Scope

• list available avatars, voices, and video-translation target languages
• create gateway-proxied requests for POST /v2/videos
• create gateway-proxied requests for POST /v2/video_translate
• poll GET /v1/video_status.get and GET /v2/video_translate/{id}
• classify and retry HeyGen 429, quota, and transient upstream responses with bounded Retry-After aware backoff
• block private or internal media URLs before asking HeyGen to fetch them
• require explicit operator grant before credit-consuming generate/translate calls

Custom avatar training, live streaming, real-time avatars, public publishing, and distribution to external channels are outside this adapter slice.

Credential Rules

HeyGen Direct API uses an API key sent as X-API-KEY. Store it in HybridClaw encrypted runtime secrets; never paste it into the prompt.

Recommended setup order:

1. Browser admin: open the active HybridClaw admin URL ending in /admin/secrets and set HEYGEN_API_KEY.
2. Browser /chat or TUI fallback: /secret set HEYGEN_API_KEY "<api-key>".
3. Local console fallback:

hybridclaw secret set HEYGEN_API_KEY "<api-key>"

For live API calls inside HybridClaw, prefer the helper's request command. It sends the secret-backed payload through the local gateway, retries bounded 429/5xx responses, summarizes large asset lists by default, and fails clearly if the gateway truncates a response.

Use http-request only when you need to inspect or hand off the generated payload wrapper to the built-in http_request tool. The helper sets secretHeaders: [{ "name": "X-API-KEY", "secretName": "HEYGEN_API_KEY", "prefix": "" }] so the gateway injects the API key server-side.

Do not try to verify HEYGEN_API_KEY with bash, environment inspection, or by asking the model whether the secret exists. The model cannot inspect the gateway secret store. If the operator says the secret was set, attempt the http_request call with the emitted secretHeaders. Only say the secret is missing if http_request returns a gateway error that explicitly says HEYGEN_API_KEY is not set, unavailable, missing, or unresolved.

Default Workflow

1. Start with request list-avatars and request list-voices unless the avatar and voice ids are already known. These commands return concise summaries by default; use --limit <count> to show more candidates. Asset ids are opaque values; never use display names as --avatar-id or --voice-id.
2. Run the script through /brand-voice before generating avatar video from marketing, sales, training, or public-facing copy.
3. Use plan for natural-language generation or translation requests so the stakes tier and required grant are explicit.
4. Do not emit generate-video or translate-video requests unless the user has granted that exact credit-consuming operation in the current task or through an approved F8 escalation.
5. Poll status with request video-status --watch or request translation-status --watch. Use a bounded polling budget, respect Retry-After, avoid parallel polling loops, and stop if HeyGen reports quota, credit, or rate-limit failure.
6. Treat auto-publish to public channels as red tier and escalate before any upload, share link, social post, or public channel message.

Command Contract

Run the colocated helper with Node:

node skills/heygen/heygen.cjs --help

Plan a request without contacting HeyGen:

node skills/heygen/heygen.cjs plan "Create an approved onboarding video from this script"

List avatars, voices, or translation languages. Asset lists use compact summary output by default:

node skills/heygen/heygen.cjs request list-avatars --limit 10
node skills/heygen/heygen.cjs request list-voices --limit 10
node skills/heygen/heygen.cjs request list-translation-languages

Generate an avatar video only after explicit operator grant:

node skills/heygen/heygen.cjs http-request generate-video \
  --avatar-id avatar_123 \
  --voice-id voice_123 \
  --script "Approved script text" \
  --title "Quarterly enablement update" \
  --resolution 1080p \
  --aspect-ratio 16:9 \
  --operator-grant

Translate a video only after explicit operator grant:

node skills/heygen/heygen.cjs http-request translate-video \
  --video-url https://example.com/source.mp4 \
  --output-language de \
  --mode fast \
  --operator-grant

Poll status:

node skills/heygen/heygen.cjs request video-status --video-id <video-id>
node skills/heygen/heygen.cjs request video-status --video-id <video-id> --watch --max-polls 10 --interval-seconds 30
node skills/heygen/heygen.cjs request translation-status --video-translate-id <translation-id>

If you truly need a raw gateway payload for the built-in http_request tool:

node skills/heygen/heygen.cjs http-request list-avatars

Classify retry behavior after an API response:

node skills/heygen/heygen.cjs classify-rate-limit \
  --status 429 \
  --retry-after 5 \
  --body-json '{"message":"Too Many Requests"}'

Stakes

Green tier:

• list-avatars
• list-voices
• list-translation-languages
• video-status
• translation-status

Amber tier:

• generate-video
• translate-video

Red tier:

• public auto-publish, social posting, public share-link distribution, or uploading generated media to a public channel

Working Rules

• Never print or ask for the HeyGen API key.
• Prefer request for live API calls. Use http-request plus the built-in http_request tool only when a raw payload is necessary.
• Do not hand-author HeyGen http_request payloads. Use the helper output unchanged so skillName, secret injection, response limits, and cost metadata stay intact.
• Do not pipe asset-list responses through head or ask for the full JSON by default. Use request list-avatars --limit <count> or request list-voices --limit <count>. These commands cache concise asset summaries locally so generate-video can catch display names and stale ids before contacting HeyGen.
• Pass --skip-cache-validation only for a known private/custom asset id supplied by the operator and not present in the cached asset list.
• Do not claim HEYGEN_API_KEY is missing unless http_request returns an explicit missing/forbidden/unresolved secret error.
• Keep scripts at or below 5000 characters for the direct video endpoint.
• Do not pass localhost, loopback, RFC1918, link-local, or other internal media URLs to HeyGen.
• Honor Retry-After on HTTP 429 and back off on quota/rate-limit messages.
• Poll with --watch only when the user expects you to wait. Keep polling bounded with --max-polls and --interval-seconds; otherwise perform one status check and report the id so the session can resume later.
• Use request or skills/heygen/client.cjs when direct execution is needed; it sends the generated httpRequest through /api/http/request, never adds cleartext HeyGen credentials to process output, retries bounded 429/5xx responses, and refuses truncated gateway bodies instead of parsing partial JSON.
• Avoid concurrent generate/translate bursts; HeyGen quotas are account and plan dependent.
• Cost per assistant run is recorded by HybridClaw UsageTotals; helper output includes costMeasurement.system = "UsageTotals" so evals can verify the accounting contract.

References

• HeyGen Quick Start: https://docs.heygen.com/
• Generate video endpoint: https://docs.heygen.com/reference/create-video-1
• Translate video endpoint: https://docs.heygen.com/reference/video-translate
• API limits and costs: https://docs.heygen.com/reference/limits

Validation

Run:

python3 skills/skill-creator/scripts/quick_validate.py skills/heygen
node skills/heygen/heygen.cjs --help
node skills/heygen/heygen.cjs eval-scenarios