handsomestWei/patent-disclosure-skill

📚 What this project does

handsomestWei/patent-disclosure-skill is a collection of Agent‑Skills that turn a large‑language model (LLM) into a personal patent‑assistant. The skill set can:

Core capability	What it actually produces
Patent‑disclosure	Takes raw project material, extracts protectable ideas, runs a novelty search and drafts a complete disclosure (invention, utility model or design) ready for filing.
Patent‑application	Converts a disclosure into the formal filing bundle – claims, description, abstract and drawings – and generates a checklist of open questions.
Docket‑assistant	Simulates a back‑and‑forth between an inventor and a patent attorney, iteratively refining the disclosure and application while keeping a full conversation log.
Patent‑reader	Turns a PDF or publication number into a plain‑language summary and a set of linked notes/graphs that can be imported into Obsidian.
Patent‑map	Visualises a personal patent knowledge base as a “map”: semantic terrain, applicant quadrants, citation networks and effectiveness matrices, all viewable in a browser.
OA‑assistant	Helps answer office‑action remarks by breaking down examiner comments, drafting response drafts and re‑using anonymised past cases.
Claim‑chart	Produces an Excel matrix that aligns each claim feature with evidence from prior art, products or standards.
Search	Generates search strings from a name, classification code, product image or claim text and runs a CNIPA (China National Intellectual Property Administration) search via Playwright.
Policy‑brief	Scrapes the CNIPA website for the latest examination policy updates and produces a short human‑readable briefing.

All of these are packaged as AgentSkills (a framework for building LLM‑driven agents) and rely on Playwright to automate the CNIPA web portal.

---

⚙️ How it works (high‑level)

1. LLM orchestration – Each skill defines a prompt template and a set of tool calls (web‑scraping, file I/O, image‑to‑text). The LLM decides which tool to invoke based on the user’s command (e.g., “交底书”).
2. Web automation – Playwright drives the CNIPA search pages, extracts bibliographic data, and optionally downloads PDFs.
3. Document generation – The model fills Jinja‑style templates for disclosures, claims, and application drafts, then writes the results to a timestamped folder under outputs/.
4. Knowledge‑graph export – Summaries are emitted as Markdown files with Obsidian‑compatible front‑matter and Mermaid diagrams, enabling the user to build a private knowledge graph.
5. Iterative workflow – The docket skill keeps a conversation history so you can ask follow‑up questions (“Add a dependent claim”) and get updated files without losing earlier versions.

---

🚀 Getting started (quick steps)

1. Clone & install – Follow INSTALL.md (Python 3.9+, pip install -r requirements.txt).
2. Configure the LLM – Set your OpenAI/Claude API key in .env (the skill uses the default model configured for AgentSkills).
3. Run a skill – python -m skills.patent_disclosure.run and type a trigger phrase like 交底书. The agent will ask for the source material and then output a folder with:
   • disclosure.docx
   • mermaid_diagram.mmd
   • conversation.log
4. Import into Obsidian – Drop the generated Markdown notes into your vault; the built‑in links and Canvas view let you explore the patent map.

---

🎯 Who would benefit?

• Inventors & R&D teams who write code/hardware but lack patent‑drafting expertise.
• SME legal departments looking to automate first‑draft disclosures and keep a searchable internal patent knowledge base.
• Patent analysts who want to quickly turn public patents into linked notes for comparative studies.

---

📦 What’s inside the repo

• skills/ – Sub‑folders for each AgentSkill (disclosure, application, docket, etc.) with their own README and prompt files.
• docs/ – Sample screenshots showing generated outputs.
• INSTALL.md – Detailed environment setup and Playwright browser installation.
• SKILL.md – Entry‑point description and usage examples.
• MIT license.

---

🛠️ Limitations & safety notes

• The tool does not provide legal advice; all generated drafts should be reviewed by a qualified patent attorney.
• It currently only supports Chinese CNIPA workflows; adapting to USPTO/EPO would require new web‑scraping modules.
• The LLM may hallucinate technical details if the input material is sparse – the docket skill mitigates this by refusing to fabricate when facts are missing.

---

🙏 Support the author

If the skill helps you file patents faster, you can “buy the author a coffee” via the link in the README. Stars are also tracked with a star‑history badge.

---

All information above is taken directly from the repository’s README; no external assumptions have been added.