heygen-com__skills/CHANGELOG.md

Changelog

3.2.0 (2026-05-13)

Features

• add heygen-translate skill (video translation / dubbing) (#82) (89d28b9)

3.1.0 (2026-04-27)

Features

• reference HeyGen ChatGPT app for Codex plugin (replaces bundled MCP) (10533a9)

3.0.0 (2026-04-27)

⚠ BREAKING CHANGES

• Direct-clone consumers who cached root SKILL.md will lose access to the meta-router skill. Contents migrated into both heygen-avatar/SKILL.md and heygen-video/SKILL.md so the skills remain fully functional standalone. Re-clone or git pull picks up the new layout.

Features

• eliminate root SKILL.md + root references/ — skills are now independent (#79) (d70bfc4)
• make heygen-avatar + heygen-video install cleanly via gh skill (#77) (e66e716)

2.3.1 (2026-04-27)

Bug Fixes

• ci: use PAT for release-please so tag push fires publish.yml (#74) (256896e)

2.3.0 (2026-04-27)

Features

• avatar: role-based symlinks (AVATAR-AGENT.md, AVATAR-USER.md) for generic lookup (#73) (5ec0927)

2.2.1 (2026-04-27)

Bug Fixes

• install: make avatar creation a first-class install step (#71) (64ab640)

2.2.0 (2026-04-27)

Features

• add Codex and Cursor plugin manifests (#65) (df9458c)
• INSTALL_FOR_AGENTS.md + 2-line paste-prompt install (#68) (d58fa24)
• route video generation through @heygen/openclaw-plugin-heygen when available (#67) (a2526d0)

2.1.2 (2026-04-17)

Bug Fixes

• frame-check: specify AI Image tool for generative fill (#62) (af77568)

2.1.1 (2026-04-17)

Bug Fixes

• heygen-avatar: agent-first default + prompt-based as primary path (#55) (57c319b)
• heygen-avatar: Before You Start ran user-centric logic before Phase 0 (#56) (0c1e9d3)
• prevent batch-asking and fix agent onboarding flow (#51) (e5e5db3)
• remove api-reference.md, enforce MCP xor CLI (no cross-lookup) (#59) (75f1ee3)
• silent mode detection, no CLI-broken narration (#54) (ff7b6a3)
• SKILL: add UX rules against narrating skill internals, next-action, and transport choice (#58) (3e8b06d)
• SKILL: HEYGEN_API_KEY presence → use CLI, short-circuit MCP (#60) (808dd77)

2.1.0 (2026-04-17)

Features

• MCP-first API mode with CLI fallback (#47) (4b052ae)

Bug Fixes

• remove prompt injection triggers from README (#44) (94981e0)

v1.3.7 (2026-04-13) — Auto-Proceed Fix

Fixed

• Video Agent review checkpoint — The HeyGen v3 Video Agent API supports an auto_proceed flag in the request body. This skill documents passing "auto_proceed": true as a server-side API parameter that tells the HeyGen backend to skip its own internal review checkpoint — it is not granting the agent discretion to submit jobs unilaterally. The agent still requires the user to initiate a video generation request before any API call is made; no video jobs are submitted autonomously.
• Batch submission throttle — Capped parallel video submissions at 2–3 max to prevent queue congestion.
• Simplified polling status flow: thinking → generating → completed (no checkpoint approval step)

---

v1.3.2 (2026-04-11) — Multi-Platform Distribution

Added

• Claude Code Marketplace — Submission docs with plugin metadata, skill listing, and install instructions (docs/claude-code-submission.md)
• Hermes Agent support — hermes metadata block in root SKILL.md frontmatter (tags + category) and install guide (docs/hermes-install.md)
• Manus support — Import workflow docs (docs/manus-install.md)
• NanoClaw container skill — Compact 112-line SKILL.md at platforms/nanoclaw/heygen/SKILL.md covering avatar discovery, voice selection, script writing, v3 video generation, polling, and delivery
• NanoClaw install guide — docs/nanoclaw-install.md

Fixed

• Synced .claude-plugin/plugin.json and .claude-plugin/marketplace.json version from 1.2.7 → 1.3.2

---

v1.3.0 (2026-04-10) — i18n Support

Added

• Language Detection — Skills detect user's language from input and communicate in that language
• Language-aware voice selection — Voice API calls filter by detected language; voice_settings.locale guidance added
• Non-English script generation — Scripts and narration generated in the video language
• i18n eval suite — Test scenarios across English, Japanese, Spanish, Korean with scoring rubric
• Japanese example prompt in prompt-craft reference
• Language consistency criterion (#11) in reviewer prompt

Changed

• Mode Detection is now explicitly language-agnostic (semantic intent matching, not English keyword matching)
• User-facing messages converted from hardcoded English strings to semantic descriptions the LLM adapts to user's language
• Buddy voice construction pattern uses {video_language} instead of hardcoded "English"
• Buddy sign-off lines converted to semantic descriptions instead of English idioms
• Test video prompt (Phase 5) generates greeting in video language instead of hardcoded English

Documentation

• Added > Language note callouts to motion-vocabulary.md and prompt-styles.md explaining English-only directives
• Updated api-reference.md with voice_settings.locale guidance and language filter examples
• All reference files document the content-language vs directive-language separation

---

v1.2.7 (2026-04-09)

Bug Fixes

• Synced version numbers across all files (SKILL.md frontmatter, User-Agent headers, plugin.json, marketplace.json) to match VERSION file
• Fixed Quick Shot avatar_id rule in heygen-video to use AVATAR file when available instead of always omitting
• Completed Frame Check correction matrix with Aspect Ratio column and ratio-fix corrections (F, G) across root SKILL.md and frame-check.md
• Fixed frame-check.md correction stacking matrix: removed stale 4-column header, corrected intro sentence (photo_avatar never gets background correction C)
• Replaced macOS-incompatible readlink -f in heygen-video preamble with POSIX-compatible path resolution

Architecture

• Trimmed root SKILL.md from 399 to ~215 lines by extracting duplicated Script, Prompt Craft, and Generate content into the producer sub-skill where it belongs
• Fixed stale path reference: identity/SKILL.md -> heygen-avatar/SKILL.md
• Registered buddy-to-avatar skill in marketplace.json

Documentation

• Added buddy-to-avatar to README "What's Inside" section

v1.1.0 (2026-04-06)

heygen-video

• Prompt-only Frame Check architecture (no external image generation, preserves face identity)
• submit-video.sh wrapper enforces aspect ratio checks before every API call
• Phase naming overhaul: action verbs replace numbered phases (Discovery, Script, Prompt Craft, Frame Check, Generate, Deliver)
• Style-adaptive Phase 3.5: 3D, animated, and photorealistic avatars get matching fill directives
• ATO lane carving: distinct tool descriptions for agent discoverability vs built-in video_generate
• Version check system with cache TTLs and snooze backoff
• Inline MP4 delivery (downloads video, sends as media attachment)
• Hard gates at all user decision points

heygen-avatar

• Voice Design endpoint (POST /v3/voices) with semantic search, seed pagination
• Reference photo nudge on first-time avatar creation
• Inline audio previews for voice selection
• Hard gates: voice selection and avatar approval require explicit user confirmation
• UX Rules: interactive at checkpoints, silent everywhere else
• Moved into heygen-stack monorepo

Infrastructure

• submit-video.sh: auto-validates avatar dimensions, appends FRAMING NOTE if mismatch detected
• update-check script moved from bin/ to scripts/
• Branch protection: 1 approval required, CODEOWNERS enforced
• README trimmed to essentials (Quick Start, What's Inside, How It Works)

v1.0.0 (2026-04-01)

Initial release. Five-phase video production pipeline with avatar discovery, prompt engineering, aspect ratio corrections, and HeyGen Video Agent API integration. 22 eval rounds, 80+ test videos generated.