원클릭으로 Manus에서 모든 스킬 실행

heygen-avatar

스타1

포크0

업데이트2026년 6월 23일 22:43

HeyGen avatar-video provider reference for Synthex. Documents the real env-driven HeyGen v2 client and its ElevenLabs voice pairing. NEVER call createAvatarVideo without a consent record. NEVER hard-code HEYGEN_API_KEY (Vercel env only). NEVER silently mock a live call — no key hard-blocks unless allowMockFallback is explicitly set. ALWAYS prefer an ElevenLabs audio_url (lip-sync) over HeyGen text TTS. Activate on ANY request to generate, poll, or debug HeyGen avatar videos, or wire avatar+voice generation.

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

Manus에서 실행

출처

CleanExpo

CleanExpo/Synthex

GitHub 저장소 열기 Creator 저장소 보기

다운로드

Manus에서 실행

파일 탐색기

2 개 파일

SKILL.md

readonly

name	heygen-avatar
description	HeyGen avatar-video provider reference for Synthex. Documents the real env-driven HeyGen v2 client and its ElevenLabs voice pairing. NEVER call createAvatarVideo without a consent record. NEVER hard-code HEYGEN_API_KEY (Vercel env only). NEVER silently mock a live call — no key hard-blocks unless allowMockFallback is explicitly set. ALWAYS prefer an ElevenLabs audio_url (lip-sync) over HeyGen text TTS. Activate on ANY request to generate, poll, or debug HeyGen avatar videos, or wire avatar+voice generation.
metadata	{"author":"synthex","version":"1.0","type":"reference-skill","epic":"SYN-1005","triggers":["heygen","avatar video","avatar lip-sync","generate avatar","video_status","elevenlabs voice clone"],"requires":[]}
context	fork

HeyGen Avatar Provider

Client

lib/marketing-agency/heygen/client.ts → createHeyGenClient(config?) returns:

createAvatarVideo(request) → { id, provider:'heygen', status }
getVideoStatus(videoId) → { status: queued|processing|completed|failed, videoUrl? }
waitForCompletion(videoId, { timeoutMs, intervalMs, sleep? }) → final job (bounded poll)

config: { apiKey?, baseUrl?, fetchImpl?, allowMockFallback? }. apiKey defaults to process.env.HEYGEN_API_KEY. fetchImpl is injectable for tests.

Request shape

{
  avatarId: string,
  script: string,                 // used only for the text→TTS path
  consent: HeyGenConsentMetadata, // REQUIRED — { subjectName, sourceRef, confirmedAt }
  audioUrl?: string,              // PREFERRED — ElevenLabs audio for lip-sync
  voiceId?: string,               // HeyGen voice for the text path (no audioUrl)
  dimension?: { width, height },  // default 1280×720
  title?: string,
}

Voice pairing (preferred path)

ElevenLabs is the voice source (lib/services/ai/voice-generation.ts → generateSpeech returns audioBase64). The composition helper lib/marketing-agency/studio/avatar-video.ts (generateAvatarVideo) does: generateSpeech → upload to a public URL → createAvatarVideo({ audioUrl }). Prefer this (real cloned voice + lip-sync) over HeyGen's built-in text TTS.

Hard rules (NEVER)

NEVER call createAvatarVideo without consent — it throws HeyGenConsentError by design.
NEVER hard-code the API key — HEYGEN_API_KEY lives in Vercel env only.
NEVER make a silent mock stand in for a live call. No key → HeyGenConfigurationError (matches the Artlist provider gate). The mock job is returned only when allowMockFallback: true is explicitly passed (dev/preview).
NEVER publish a video still in queued/processing — poll to completed first.

Status mapping

HeyGen completed|success → completed; failed|error → failed; pending|waiting|processing → processing.

Cost awareness

HeyGen renders + ElevenLabs synthesis are billed per use. Track cost-of-generation per client (margin vs retainer) — see client-content-studio. Don't loop renders without a cap.

Testing

Inject fetchImpl (and sleep for waitForCompletion) — never hit the network in unit tests. See tests/unit/lib/heygen-client.test.ts.

Anti-patterns

❌ Throwing away the consent gate to "simplify" — it's a legal/likeness control.
❌ Using HeyGen text TTS when an ElevenLabs cloned voice is available (off-brand voice).
❌ Blocking a serverless function on a long synchronous poll — prefer queue + status check.

Foundation & Gate Wiring (SYN-1050)

Adopted from the senior-skill standard so every artefact this skill produces is checked against the locked foundation before it lands.

Reads at every invocation (never cached — re-read each run):

.claude/memory/ceo-foundation.md — consent records, Aid Rule (Q3.1.1) on RestoreAssist, cross-client boundary (Phase 3.4), no fabricated client metrics.
.claude/memory/verification-gates.md — gate state for any claim referenced.

Output gate: every client-facing artefact this skill produces routes through brand-voice-enforce before the CEO batched-review queue. A REJECT blocks the artefact until the quoted offending string is fixed.

Evidence standard: every quantitative or factual claim carries exactly one tag — [VERIFIED] / [INFERENCE] / [UNCONFIRMED]. Untagged = defect (.claude/rules/fabel-evidence-standard.md). Never state a projected result as fact.

Spec: see spec.md in this skill directory.

이 저장소의 다른 Skills

같은 저장소

north-star

CleanExpo/Synthex

The engineering navigation layer for Synthex — holds the mission, the current course, the rules of the ship, and routes work to the right existing specialist. Load at the START of any planning, prioritisation, "what's next", roadmap, scope, or new-feature decision. NEVER let work drift off the critical path or add capabilities the ship already carries (see dependency-discipline). NEVER claim done without the verification gate. NEVER merge to production or change prod data without a human gate. ALWAYS trace a task to the NorthStar, the live Linear epic, and an existing system before writing code.

2026-06-241

playwright-cli

CleanExpo/Synthex

Authenticated browser automation for Synthex — drive real Chrome to log in, screenshot, and audit dashboard/integration surfaces. Use for live visual verification, "audit the dashboard", "check what's connected", or closing the verification gate on /dashboard/* routes. Two backends: the Playwright MCP server (interactive, in-session tools) and the committed CLI scripts (scripts/browser/*.mjs, reliable + headless). RUNS WHERE THE BROWSER + NETWORK + AUTH EXIST — i.e. a local machine, not a network-restricted sandbox.

2026-06-241

api-testing

CleanExpo/Synthex

Synthex API testing enforcer. NEVER mock the database in integration tests — use real Supabase. NEVER skip the 401 (unauthenticated) or 403 (wrong org) test cases. NEVER use pnpm. ALWAYS structure tests as: 401 → 403 → 400 → 200/201 happy path. Activate on ANY request to write API tests, validate endpoints, add test coverage, or check test quality.

2026-06-231

database-prisma

CleanExpo/Synthex

Synthex database operations enforcer. NEVER use prisma db push for schema changes — use migrate diff + db execute. NEVER add non-nullable columns without defaults. NEVER write Prisma queries without organizationId scope. ALWAYS use backward-compatible migrations and validate with prisma validate first. Activate on ANY request to change schema, write a query, create a migration, add a model, or modify database operations.

2026-06-231

design

CleanExpo/Synthex

Synthex design system enforcer. NEVER use Inter as a heading font, purple (#8B5CF6) gradients on white, or generic glassmorphism without Synthex tokens. ALWAYS use Space Grotesk headings, #f97316 brand orange, #0f172a slate background, and the Synthex glass token set. Activate on ANY request involving UI, components, styling, layout, visual design, colour, typography, spacing, shadows, animations, or anything a user will see on screen.

2026-06-231

security-hardener

CleanExpo/Synthex

Synthex security posture enforcer. NEVER apply generic OWASP checklists without grounding in Synthex's specific threat model. NEVER suggest any auth system other than Supabase. ALWAYS audit: SSRF via validateExternalUrl, JWT tier elevation via resolveVerifiedTier, CORS via CORS_ORIGIN exact-match, org-scope bypass in Prisma. Activate on ANY request to harden security, audit vulnerabilities, review auth patterns, check for injection risks, or assess an attack surface.

2026-06-231