Script Writer Agent Skills: The Missing Link Between Blog Content and Video Production
Video content creation is one of the hottest AI application scenarios, but there's a clear gap in our skill system: blog posts have blog-content-authoring, music has songwriting-and-ai-music, video frameworks have remotion-best-practices, but the structured creation capability from "idea" to "video script" is missing.
Script Writing Projects on GitHub
Three relevant projects found, none being a standalone high-star agent skill:
YouTube Script Writer (rahulanand1103, 22 stars) — The most directly borrowable project. Uses a 4-step LangGraph pipeline:
Input(title/language/tone/duration) →
1. CreateBlueprint(generate initial outline) →
2. InternetResearch(web research) →
3. RefinedBlueprint(refine outline based on research) →
4. GenerateScript(write script section by section) →
Output(complete script + YouTube Description)
The core highlight is the 4-role prompt chain: Content Strategist generates structured outline (with time allocation) → Research Analyst generates search queries per section → Script Architect integrates research into refined outline → Script Writer writes per section, strictly following research content.
OpenLucid (agidesigner, 23 stars) — A complete marketing knowledge model platform where Script Writer is one of 9 core modules. Uses a "platform × persona × structure" 3D preset matrix, with each platform/persona/narrative structure prompt as an independent markdown file. Links with knowledge base (Brand Kit + Knowledge Base + Strategy Units), output connects to digital avatar video generation.
AIYOU (yubowen123, 112 stars) — Chinese AI short drama platform: 5-50 episode framework → per-episode scripts → character dialogue + scene descriptions → storyboards → AI video generation. React+Express architecture with node-based drag-and-drop.
Capability Gap Matrix
| Dimension | YouTube Script Writer | OpenLucid | AIYOU | Our Skills |
|---|---|---|---|---|
| Video script structuring | Multi-level outline + time allocation | Platform preset + structure templates | Episode/scene/storyboard | None |
| Web research augmentation | Per-section search + citations | Knowledge base driven | Partial | None |
| Tone/style control | Multi-tone presets | Brand Kit + persona presets | Limited | humanizer (de-AI only) |
| Multi-platform adaptation | YouTube only | Xiaohongshu/WeChat/Douyin etc. | Short drama only | blog only |
| Script→video handoff | None | Digital avatar video gen | Storyboard→AI video | remotion (framework only) |
| Multi-role collaborative writing | 4-agent role chain | MCP Agent collaboration | Node-based | None |
Core gap: Structured video/audio script creation capability is completely absent.
Borrowable Patterns
Pattern A: Multi-Role Pipeline (YouTube Script Writer)
The 4-role prompt chain is the most borrowable design. Each role has an independent system prompt, structured I/O. The Researcher role auto-generates search queries, injecting results into subsequent writing. This can be implemented as a multi-step flow within a skill, without LangGraph.
Pattern B: Platform × Persona × Structure Preset Matrix (OpenLucid)
Each platform has independent prompt templates, Brand Kit provides brand tone constraints. Can be combined with humanizer, embedding multi-platform prompt templates in the skill.
Pattern C: Structured Output Schema
Pydantic Schema defines BluePrint → Section structure, each section containing section_title / description / time / pointers. We can use JSON/YAML to define script structure, outputting markdown that connects to remotion.
Self-Build Approach
Recommend creating a new script-writer skill with core design:
| Design Dimension | Approach |
|---|---|
| Positioning | Video/audio/presentation script structured creation skill |
| Input | Topic + platform(YouTube/Xiaohongshu/Douyin/TikTok/presentation) + tone + duration |
| Flow | Strategist(outline) → Researcher(web supplement) → Architect(refine) → Writer(draft) |
| Output structure | JSON Schema: title → sections[{title, description, time, pointers, content}] |
| Output format | Markdown script (connectable to remotion) |
| Platform presets | One prompt template per platform (markdown file) |
| API dependencies | No paid APIs (use Hermes local LLM + existing web search) |
| Implementation | Pure prompt + template + rules, zero external dependencies |
Synergy with existing skills: Generate script → humanizer polish → remotion video production. This completes the blog → script → humanize → remotion content creation chain.
Sources:
- YouTube Script Writer: https://github.com/rahulanand1103/youtube-script-writer (MIT, 22 stars)
- OpenLucid: https://github.com/agidesigner/OpenLucid (MIT, 23 stars)
- AIYOU: https://github.com/yubowen123/AIYOU_open-ai-video-drama-generator (MIT, 112 stars)
- songwriting-and-ai-music: Hermes Agent built-in skill
- remotion-best-practices: Hermes Agent built-in skill
- humanizer: Hermes Agent built-in skill