English | 简体中文
A Claude Code skill that orchestrates end-to-end writing workflows — from idea to polished, illustrated, published article. Built-in topic management, viral benchmarking, experience tracking, multi-platform publishing, and autonomous mode for fully hands-off execution.
Ask Claude Code:
Install the writing-assistant skill from https://github.com/VegetaPn/writing-assistant-skill to my project directory
Manual installation
curl -L https://github.com/VegetaPn/writing-assistant-skill/archive/refs/heads/main.zip -o writing-assistant-skill.zip
mkdir -p .claude/skills
unzip writing-assistant-skill.zip -d .claude/skills/
mv .claude/skills/writing-assistant-skill-main .claude/skills/writing-assistant
rm writing-assistant-skill.zip/writing-assistant
Or just talk to Claude:
I want to write an article about the attention economy, for Xiaohongshu
That's it. The skill walks you through the entire process interactively.
Want to walk away and come back to a finished article? Just say:
Autonomous mode, write an article about the attention economy for Xiaohongshu
The AI runs the entire workflow start-to-finish — no interaction needed. It makes every decision autonomously, logs everything, and only stops when a strict Completion Gate checklist is fully satisfied.
More examples:
全自动写一篇公众号文章，主题是注意力经济，写完发到公众号
Write three XHS articles on different angles of AI productivity
自主模式，把 inbox 里的选题都写了
| Phase | What happens |
|---|---|
| Prep | Environment check, progress tracker, platform selection |
| Research | Search reference library, analyze platform trends, select writing techniques |
| Draft | Interactive Q&A to clarify intent → structured draft with technique application |
| Refine | Element-level optimization (title, opening, structure, hooks) |
| Polish | Professional content refinement via content-research-writer |
| Illustrate | Auto-generate images via baoyu-xhs-images + generate-image |
| Publish | Review → platform adaptation → publish to WeChat / Xiaohongshu / X |
Three starting modes: topic idea, raw materials, or existing draft.
Two execution modes: interactive (default, step-by-step with user) or autonomous (fully hands-off).
| Command | What it does |
|---|---|
| "Record a topic: {idea}" | Save an idea to inbox |
| "Show topics" | View topic pipeline |
| "Develop topic: {topic}" | Research topic with benchmarks + outline |
| Command | What it does |
|---|---|
| "Analyze viral post" + URL | Deep-analyze a viral piece |
| "Monitor trends" / "Show trends" | Scan trending content across platforms |
| "Start trend monitoring" | Start background monitoring |
| Command | What it does |
|---|---|
| "Generate titles" | Generate platform-optimized title candidates |
| Command | What it does |
|---|---|
| "Show experience" | View lessons and recent cases |
| "Summarize experience" | Re-distill rules from all cases |
Autonomous Mode (New in 2.1)
- Send one instruction, walk away, come back to a finished article
- AI makes all decisions autonomously — platform, title, structure, style — and logs every choice
- Failure-resilient: tool failures are recorded and skipped, never blocking the workflow
- Stuck detection: auto-detects and escapes hung operations
- Completion Gate: strict 7-point checklist (G1-G7) that must ALL pass before stopping — guarantees nothing is left unfinished
- Supports batch execution: "write all topics in inbox", "write 3 articles", etc.
- Full transparency: Autonomous Decision Log + execution summary for post-hoc review
- Images skipped by default in autonomous mode (opt-in with explicit request)
Content Creation
- Multi-mode input: topic idea / raw materials / existing draft
- Interactive clarification — asks smart questions, not generic templates
- Psychology-based writing techniques (content funnel, emotional hooks) applied throughout
- Platform-aware from the start — length, tone, and structure adapt to target platform
Reference & Learning System
- Three-level content hierarchy (system → user → project) with automatic merging
- Reference library: author styles, title patterns, opening techniques, structure templates
- Viral benchmarking: analyze hits, extract patterns, grow your library organically
- Experience tracking: auto-records corrections, distills lessons, prevents repeated mistakes
Production Pipeline
- Professional polishing via content-research-writer skill
- Auto-generated illustrations via baoyu-xhs-images + generate-image
- Per-session progress tracker with step-by-step checklists
- Multi-platform publishing: WeChat Official Account, Xiaohongshu, X/Twitter
- One article → multiple platform adaptations with per-platform optimization
To generate actual images (not just descriptions), set up an OPENROUTER API key:
- Get a key from OpenRouter
- Add to
.envin your project root:OPENROUTER_API_KEY=your_key_here - Add
.envto.gitignore
Without this, the workflow still works — Step 7 will produce image descriptions that you can create manually.
All dependencies are bundled in the repository and auto-installed on first run:
| Skill | Purpose |
|---|---|
| content-research-writer | Content polishing |
| baoyu-xhs-images | Illustration generation |
| generate-image | AI image generation (needs OPENROUTER API) |
| xiaohongshu | Xiaohongshu content creation, search & publish |
| wechat-article-search | WeChat article search |
| baoyu-post-to-wechat | WeChat publishing |
| baoyu-post-to-x | X/Twitter publishing |
Manual dependency installation
mkdir -p .claude/skills
cp -r dependencies/<skill-name> .claude/skills/File structure
writing-assistant-skill/
├── SKILL.md # Main workflow orchestrator
├── skills/ # Sub-skills (project-local)
│ ├── title-generator.md # Platform-optimized titles
│ ├── content-adapter.md # Multi-platform adaptation
│ ├── topic-manager.md # Topic lifecycle + benchmarking
│ └── experience-tracker.md # Correction tracking + lessons
├── assets/ # System-level defaults
├── references/ # System-level reference library
│ ├── authors/ # Author profiles & styles
│ ├── by-element/ # Writing elements (cases)
│ └── techniques/ # Methodologies (principles)
├── dependencies/ # Bundled dependency skills
└── outputs/ # Generated articles
Three-level content system
Assets and references use a three-level hierarchy. Content merges on read; lower levels override higher on conflict.
| Level | Location | Purpose |
|---|---|---|
| System | {skill-dir}/assets/, {skill-dir}/references/ |
Skill defaults (read-only) |
| User | {project-root}/assets/, {project-root}/references/ |
Your accumulated knowledge |
| Project | outputs/{topic-slug}/assets/, outputs/{topic-slug}/references/ |
Per-article overrides |
Output directory
outputs/{topic-slug}/
├── {topic-slug}-progress.md # Progress tracker
├── {topic-slug}.md # Initial draft
├── {topic-slug}-polished.md # Polished version
├── {topic-slug}-final.md # Final version
├── {topic-slug}-{platform}.md # Platform adaptation
└── xhs-images/ # Illustrations
Contributions, issues, and feature requests are welcome at the issues page.
MIT License
Autonomous Mode — send one instruction, walk away, come back to a finished article.
- Autonomous mode: fully hands-off execution from idea to final article
- Natural language completion conditions ("write 3 articles", "write all inbox topics", etc.)
- Completion Gate: mandatory 7-point checklist (G1-G7) enforced before stopping
- Autonomous Decision Log: every AI decision recorded with reasoning and alternatives
- Failure-resilient execution: tool failures logged and skipped, never blocking
- Stuck detection: auto-escapes hung operations with fallback strategies
- Autonomous capability boundary: auto-skips operations requiring user participation (login, scan, etc.)
- Images skipped by default in autonomous mode for speed (opt-in)
- Batch execution: multiple articles with independent progress trackers and cross-article learning
- Execution summary: structured post-hoc report with key decisions, outputs, failures, and review recommendations
Architecture overhaul — from linear writing tool to sustainable content creation system.
- Sub-skill architecture (title-generator, topic-manager, experience-tracker)
- Three-level content system with merge protocol
- Topic management: inbox → developing → writing workflow
- Viral benchmarking: single analysis, batch scanning, background monitoring
- Experience tracking: auto-detect corrections → cases → distilled lessons
- Writing methodology integration (psychology-based techniques)
- Real-time platform search before writing
- Multi-platform adaptation
- Progress tracker with execution log and retrospective
- Reference library system (author profiles, writing elements, techniques)
- Element-level refinement (titles, openings, structures, hooks)
- Bundled dependencies with auto-installation
- Image generation support (OPENROUTER)
- Initial release: three starting modes, content polishing, image generation, WeChat/X publishing