The First AI-Native Intellectual Property Skill for Agent Workflows -- with Direct Word + PPT Output
An Agent Skill that generates Chinese patent application documents, software copyright registration materials, and technical disclosure reports from AI project code, research papers, and design docs -- with domain-specific rules baked in, and outputs ready-to-use .docx + .pptx files by default. The patent path includes a built-in technical disclosure intermediate deliverable for handoff to patent agents.
Existing patent/copyright skills are all generic -- same templates for any domain. AI has unique IP challenges:
- Patentability: Is your algorithm even patentable under CNIPA guidelines?
- Claim drafting: How to write method+system+storage-medium triple claims for a neural network?
- Desensitization: How to replace "A100", "PyTorch", "7B parameters" with patent-safe language?
- Software copyright: Which source files to include? How to handle GPU/CUDA requirements?
- Paper-to-patent mapping: How to turn ablation studies into dependent claims?
- Output format: Can I get a Word file directly instead of Markdown?
- Internal communication: How to brief management or patent agents quickly?
- 3D vision / Generative AI / Embodied AI: Domain-specific claim structures and patentability traps?
This skill answers all of them.
| Path | Deliverables | Default Format |
|---|---|---|
| Patent | Disclosure + Claims + Specification + Abstract + Briefing PPT | .docx x2 (disclosure + patent) + .pptx |
| Software Copyright | Software Manual + Source Code Doc | 2 .docx files |
| Technical Disclosure | Full disclosure document | Single .docx file |
| Domain | Sub-Directions | Claim Template |
|---|---|---|
| Perception AI | 2D CV · 3D Vision & Graphics · Audio/Video · Sensor Fusion | 2.1 Architecture · 2.2 3D Vision |
| Cognitive & Language | LLM · Multimodal LLM · RAG · Knowledge Graph | 2.3 Training · 2.4 MLLM · 2.5 RAG |
| Generative AI | Diffusion & Controlled Gen · AIGC Watermark · Style Transfer | 2.6 Diffusion |
| Decision & Interaction | Agent · Embodied AI · Reinforcement Learning | 2.7 Agent · 2.8 VLA |
| AI Engineering | Training/Fine-tuning · Inference Optimization · Data Engineering · Edge/IoT | 2.9 Inference · 2.10 Data |
| AI Safety & Governance | Adversarial Defense · Watermark & Traceability · Federated Learning · Alignment | 2.11 Watermark |
| Industry Applications | Autonomous Driving · Industry QC · Healthcare · Finance · AI4Science · Digital Content | Domain-adapted |
- Patentability Pre-Assessment (Phase 0): 3-factor test + domain risk levels for 13 high-risk areas
- 15-Type Claim Templates: Architecture / 3D Vision / Training / MLLM / RAG / Diffusion / Agent / Embodied AI / Inference / Data / Watermark / World Model / SSM / Video Generation / AI-native Application
- Domain-Specific Dependent Claim Strategies: 5-layer general + 9 domain-specific expansion tables
- Patent Layout Advisory: Single vs. divisional filing; special rules for 3D visions and Embodied AI
- 6+14+4 Desensitization Rules: 6 general + 14 industry-specific + 4 3DGS/NeRF-specific replacements
- AI Project Auto-Detection: Decision tree covering 15+ project types + 9 industry categories
- Source Code Priority: 16-level priority with domain-specific mandatory files (3D vision, generative AI, embodied AI, RAG)
- Paper-to-Patent Mapping: 6-type general + 10 domain-specific mapping rules
- 5 Software Manual Templates: General / 3D Vision / Generative AI / Embodied AI / World Model
- 13-Type AI System Figure Requirements: From DL inference to AI watermark systems
- 6-Group CPC/IPC Classification: 30+ codes across AI core / perception / language / decision / safety / industry
- 100-Point Quality Scoring: Quantitative self-check for each output path
- Technical Disclosure as Intermediate (C1.3): Auto-generated from technical key points + search results
- Direct Word (.docx) Output (Phase F): Professional patent document formatting with docx-js
- Briefing PPT (.pptx) Output (Phase G): 5-8 slide executive briefing with AI domain tag
- Multi-Source Prior Art Search (Phase C1.1): CNIPA crawler + Google Patents API + arXiv + WebSearch fallback
- Structured Prior Art Comparison (Phase C1.1b): Feature comparison table with risk grading (Low/Medium/High)
- Patent Figure Auto-Generation (Phase H): Mermaid/Graphviz → PNG, 9 AI-domain-specific diagram types
- Full OA Response Workflow (Phase E.3): 6-step office action response with A33 compliance check
- Iteration Traceability: Timestamped deliverables + revision dialog log with diff tracking
- EPO/USPTO Differentiated Drafting (§14): 35 USC 101/102/103, Art. 52/56/83, multi-office claim formatting
- Confirmation Gates: Pause at each phase boundary for user review
- Iterative Correction: Modify specific sections without re-running the entire pipeline
- Quantitative Self-Check: Auto-score and auto-fix when score < 80
- Slash Commands:
/ai-copyrightor/AI知产 - Format Options:
docx(default) /md/both - PPT Option: Auto-generated for patent path;
--pptflag for other paths
Phase 0 Patentability Pre-Assessment + Domain Risk Detection (patent path only)
Phase A Requirement Diagnosis -> path selection + domain attribution + risk level
Phase B Project Analysis -> auto-detect AI project type (15+ types) + extract key points
Phase C Generation (branch by path)
├── C1 Patent (search -> layout -> disclosure -> claims (15 templates) -> specification -> abstract -> self-check)
├── C2 Software Copyright (manual (5 templates) -> source code doc -> self-check)
└── C3 Technical Disclosure (general + domain-specific mapping -> drafting -> self-check)
Phase D Confirmation Gate (user review at each phase end)
Phase E Iterative Correction (targeted fix, no full re-run)
Phase F Word Document Output (automatic, docx-js workflow)
Phase G Briefing PPT Output (patent path default; others optional)
Phase H Patent Figure Generation (Mermaid/Graphviz → PNG, auto-embedded in specification)
AI-Copyright-Skill/
├── SKILL.md # Main skill file (agent entry point)
├── README.md # This file
├── README_CN.md # Chinese documentation
├── LICENSE # MIT License
├── tools/ # Prior art search scripts
│ ├── cnipa_search.py # CNIPA patent search (Playwright)
│ └── google_patents_search.py # Google Patents API search
├── examples/ # Example cases
│ └── example-3dgs-denoise/ # 3DGS denoising example
└── references/ # Domain knowledge base
├── ai-patent-special.md # Patentability + domain risks + layout + mapping + desensitization + CPC + OA + EPO/USPTO (14 sections)
├── ai-software-copyright-guide.md # Project detection (15+ types) + source priority (16 levels) + 5 manual templates + 14 pitfalls
├── ai-patent-claims-guide.md # 15 claim templates + 5-layer + 9 domain-dependent claim expansion
└── ai-prior-art-search.md # Multi-source prior art search + structured comparison + risk grading
Copy the entire AI-Copyright-Skill/ directory to your skills folder:
macOS / Linux:
# TeleAI Super Agent
cp -r AI-Copyright-Skill/ ~/.config/teleai-super-agent/skills/AI-Copyright-Skill/
# OpenClaw / Claude Code
cp -r AI-Copyright-Skill/ .claude/skills/AI-Copyright-Skill/Windows (PowerShell):
# TeleAI Super Agent
Copy-Item -Recurse AI-Copyright-Skill/ "$env:USERPROFILE\.config\teleai-super-agent\skills\AI-Copyright-Skill"
# OpenClaw / Claude Code
Copy-Item -Recurse AI-Copyright-Skill/ .\.claude\skills\AI-Copyright-Skill# Install to TeleAI Super Agent skills directory
clawhub install ai-copyright-skill --dir ~/.config/teleai-super-agent/skills
# Or install to current directory (then copy manually)
clawhub install ai-copyright-skillNote: The ClawHub slug is
ai-copyright-skill(lowercase). If you haven't installedclawhubCLI yet, see ClawHub for setup instructions.
Once installed, simply mention patent/copyright-related keywords in your agent chat:
- "Help me write a patent for my 3DGS reconstruction method"
- "Generate software copyright materials for this embodied AI project"
- "Convert my diffusion model paper into a technical disclosure document"
/ai-copyrightor/AI知产
User: 帮我基于我的3DGS项目写一份发明专利申请
Agent (Phase 0): Running patentability assessment...
✓ Technical problem: "Real-time rendering bottleneck in large-scale 3D scene reconstruction"
✓ Technical means: "Gaussian splatting with adaptive density control"
✓ Technical effect: "Inference latency reduced from 120ms to 35ms"
⚠ Domain risk: MEDIUM (3D vision - ensure rendering formula in claims)
→ Patentability: PASS (with domain-specific drafting rules applied)
Agent (Phase B): Auto-detecting project type...
Entry file: render.py + gaussian.py → 3D Vision (3DGS)
Domain: Perception AI > 3D Vision & Graphics
Claim template: 2.2 (3D Vision)
Agent (Phase C1): Generating patent documents...
C1.1 Prior art search: Found 3 related patents, differentiated
C1.2 Layout advice: Single application (reconstruction+rendering coupled)
C1.3 Technical disclosure: Generated (for patent agent handoff)
C1.4 Claims: Independent (method, 3D vision 4-stage) + 5 dependent + Independent (system) + storage media
C1.5 Specification: 5 chapters, 3 embodiments, 3DGS-specific desensitization applied
C1.6 Abstract: 280 characters
C1.7 Self-check: 92/100 → PASS
Agent (Phase F): Generating Word documents...
→ 3dgs-adaptive-density_disclosure.docx (technical disclosure)
→ 3dgs-adaptive-density.docx (cover + claims + specification + abstract)
Agent (Phase G): Generating briefing PPT...
→ 3dgs-adaptive-density_简介.pptx (8 slides, domain: Perception AI)
Output: outputs/3dgs-adaptive-density/
└── patent/
├── 3dgs-adaptive-density_disclosure.docx ← For patent agent
├── 3dgs-adaptive-density.docx ← Formal application
├── 3dgs-adaptive-density_简介.pptx ← Management briefing
├── 3dgs-adaptive-density_claims.md
├── 3dgs-adaptive-density_specification.md
└── 3dgs-adaptive-density_abstract.md
User: 生成这个AI项目的软著登记材料
Agent (Phase B): Auto-detecting project type...
Entry file: app.py + serve.py → AI Service (Web API)
Framework: FastAPI + PyTorch
GPU required: Yes (CUDA 11.8+)
Manual template: 3.1 General type
Agent (Phase C2): Generating software copyright materials...
C2.1 Manual: 12 pages, 8 screenshots, AI-capability description included
C2.2 Source code: model.py(340 lines) + train.py(280 lines) + ...
Total: 1560 lines → All submitted (< 3000 lines)
C2.3 Self-check: 88/100 → PASS
Agent (Phase F): Generating Word documents...
→ xxx_软件说明书.docx
→ xxx_源代码文档.docx
User: 把我这篇论文转成技术交底书
Agent (Phase B): Detecting domain...
Paper topic: Diffusion model with ControlNet
Domain: Generative AI > Diffusion & Controlled Generation
Mapping: Domain-specific (noise schedule → claim step, condition injection → sub-claim)
Agent (Phase C3 → Phase F): Mapping paper→disclosure→Word...
Research Problem → Technical Problem ✓
Contribution → Technical Solution ✓
Ablation Study → Alternative Embodiments ✓
Diffusion-specific: Noise schedule → dependent claim, ControlNet injection → dependent claim ✓
→ xxx_disclosure.docx generated
| Feature | AI-Copyright-Skill v3.0 | Fokkyp/SoftwareCopyright-Skill | patent-disclosure-skill | patent-software-ip (v1.0) |
|---|---|---|---|---|
| Patent application | Full | -- | -- | Full |
| Software copyright | Full | Full | -- | Full |
| Technical disclosure | Full | -- | Full | -- |
| 7 AI domains x 22 sub-directions | Yes | -- | -- | -- |
| 15 claim templates | Yes | -- | -- | -- |
| Domain-dependent claim strategies | Yes (9 domains) | -- | -- | -- |
| Domain risk assessment | Yes (13 risk areas) | -- | -- | -- |
| 13-type figure requirements | Yes | -- | -- | -- |
| 5 software manual templates | Yes | -- | -- | -- |
| Disclosure as patent intermediate | Yes | -- | -- | -- |
| Direct Word (.docx) output | Yes | -- | -- | -- |
| Briefing PPT (.pptx) output | Yes | -- | -- | -- |
| AI patentability check | Built-in | -- | -- | -- |
| AI project auto-detection | Built-in (15+ types) | -- | -- | -- |
| AI desensitization rules | 6+14+4 rules | -- | -- | Partial |
| Paper-to-patent mapping | 6-type + 10 domain | -- | Partial | -- |
| Patent layout advisory | Built-in | -- | -- | -- |
| Confirmation gates | Yes | Yes | -- | -- |
| 100-point quality scoring | Yes | -- | -- | -- |
| Quantitative self-check | Yes | -- | -- | Checklist only |
| Iterative correction | Yes | -- | Yes | Yes |
| Multi-source prior art search | CNIPA+Google Patents+arXiv | -- | CNIPA only | -- |
| Structured prior art comparison | Feature table + risk grading | -- | -- | -- |
| Patent figure auto-generation | Mermaid/Graphviz → PNG | -- | Mermaid | -- |
| Full OA response workflow | 6-step + A33 check | -- | -- | -- |
| EPO/USPTO drafting | §14 differential rules | -- | -- | -- |
| Iteration traceability | Timestamped + dialog log | -- | Timestamped | -- |
| Reference File | Sections |
|---|---|
ai-patent-special.md |
§1 Patentability + 13 domain risks / §2 Layout / §3 Mapping / §4 Figures / §5 Desensitization / §6 CPC / §7 Quick reference / §8 2026 guidelines / §9 Innovation mining / §10 OA response / §11 Domain algorithms / §12 Terminology / §13 PCT / §14 EPO/USPTO |
ai-software-copyright-guide.md |
§1 15+ type decision tree + 9 industry detection / §2 16-level source priority / §3 5 manual templates / §4 14 pitfalls |
ai-patent-claims-guide.md |
§1 Triple claim / §2 15 claim templates (incl. world-model/SSM/video/AI-native) / §3 5-layer + 9 domain-dependent expansion |
ai-prior-art-search.md |
§1 Multi-source search (CNIPA/Google/arXiv) / §2 Structured comparison + risk grading / §3 Report format |
PptxGenJS is required for Phase G (briefing PPT). The skill auto-installs it in the working directory, but if network or permission issues prevent this:
# Manual install in the skill's temp directory
cd .temp/{case-name}/ && npm install pptxgenjs
# Or install globally
npm install -g pptxgenjsThis is a PptxGenJS API pitfall. When using addText with bullet lists, you must pass an array of objects, not an array of strings:
// Wrong - will throw TypeError
slide.addText(["item1", "item2"], { bullet: true, ... });
// Correct - array of text objects
slide.addText([{ text: "item1\n" }, { text: "item2" }], { bullet: true, ... });The docx npm package is required for Phase F (Word output). It is usually pre-installed in TeleAI Super Agent. If missing:
npm install docxThe ClawHub slug is lowercase: ai-copyright-skill. Using AI-Copyright-Skill (mixed case) will not match.
# Correct
clawhub install ai-copyright-skill
# Wrong
clawhub install AI-Copyright-SkillMake sure the skill folder is in the correct directory and the folder name matches exactly:
| Agent | Skills Directory |
|---|---|
| TeleAI Super Agent | ~/.config/teleai-super-agent/skills/AI-Copyright-Skill/ (macOS/Linux) |
%USERPROFILE%\.config\teleai-super-agent\skills\AI-Copyright-Skill\ (Windows) |
|
| OpenClaw / Claude Code | .claude/skills/AI-Copyright-Skill/ |
Verify the skill is loaded by checking if SKILL.md exists inside the folder.
The skill auto-fixes when the score is 60-80. If it stays below 60 after retries, the most common causes are:
- Insufficient technical detail: Add more quantified comparison data (e.g., "latency reduced from 120ms to 35ms")
- Missing dependent claims: Ensure the independent claim is followed by at least 5 dependent claims
- Incomplete desensitization: Check for brand names (A100, PyTorch) or exact parameter values (7B, 1.5B)
The default font is Microsoft YaHei (微软雅黑). If it is not installed on your system (common on non-Chinese Windows or some Linux distros), install it or switch to a compatible CJK font:
// Replace in the PptxGenJS script
fontFace: "Noto Sans CJK SC" // or "Source Han Sans CN", "SimHei"- Added Phase H: Patent figure auto-generation (Mermaid/Graphviz → PNG, 9 AI-domain diagram types)
- Added multi-source prior art search (CNIPA crawler + Google Patents API + arXiv + fallback)
- Added structured prior art comparison with feature table and Low/Medium/High risk grading
- Added full OA response workflow (6-step: parse → compare → strategy → A33 check → argument → replacement pages)
- Added iteration traceability (timestamped deliverables + revision dialog log)
- Added EPO/USPTO differentiated drafting rules (§14: 35 USC 101/102/103, Art. 52/56/83)
- Added
tools/directory with CNIPA search and Google Patents search scripts - Added
examples/with 3DGS denoising example case - Added
references/ai-prior-art-search.mdknowledge base - Expanded claim templates from 11 to 15 (added: World Model, SSM, Video Generation, AI-native Application)
- Expanded domain risks from 8 to 13 areas
- Expanded software manual templates from 4 to 5 (added: World Model/Simulation)
- Expanded source code priority from 12 to 16 levels
- Synchronized README numbers with SKILL.md across all metrics
- Expanded from 6 claim types to 11 claim templates covering 7 AI domains x 22 sub-directions
- Added domain risk assessment with 8 high/medium-risk areas and drafting countermeasures
- Added 5 new claim templates: 3D Vision, Diffusion/Generative AI, RAG, Embodied AI (VLA), AI Watermark
- Added 5 domain-specific dependent claim expansion strategies
- Expanded desensitization from 6 to 6+14+4 rules (general + industry + 3D-specific)
- Expanded project detection from 6 to 11 types + 6 industry categories
- Added 4 software manual templates (general/3D-vision/generative-AI/embodied-AI)
- Expanded figure requirements from 5 to 13 AI system types
- Expanded CPC/IPC from 6 to 6 groups / 30+ codes
- Added §7 domain quick reference table to ai-patent-special.md
- Initial public release with 6 claim templates and 3 patent paths
MIT License -- see LICENSE
jaccen -- GitHub | AI-Copyright-Skill | Awesome-Gaussian-Skills
This skill was designed by analyzing and learning from these excellent projects:
- Fokkyp/SoftwareCopyright-Skill -- Confirmation gate interaction design
- patent-disclosure-skill -- Prompt molecule architecture + disclosure workflow
- na57/chinese-copyright-application-skill -- AI project type detection
- Earl000333/paperforge -- Paper-to-patent mapping methodology
- hapi-ds/mPAPA -- Structured multi-step workflow
- IGTA-Tech/provisional-patent-skills -- 100-point quality scoring
- handsomestWei/patent-disclosure-skill -- CNIPA crawler + Mermaid figure generation + iteration traceability
- yycyyv/M-Cube -- OA response workflow + multimodal figure analysis + multi-verification
- RobThePCGuy/Claude-Patent-Creator -- Multi-source patent search + compliance check architecture
- kimlawtech/korean-patent-diagram -- Patent diagram auto-generation from specification