Skip to content
Get packs

Guide · Agent skill

How to use the Patent Novelty Check agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

The patent-novelty-check skill is an instruction file that tells an AI agent how to assess whether an invention could qualify for a patent. It checks each prior-art reference for anticipation (one document disclosing every claim element) and for obviousness (a combination of references plus a reason to combine them). It is written for inventors, small teams, patent agents and students who want a structured first-pass screen. After this guide you will know what the skill does, how to install it, how to trigger it and what it needs to run.

Patent Novelty Checkwanshuiyin/Auto-claude-code-research-in-sleep · MIT

Quick start

  1. 1Download and unzip. Download the ZIP from this page and unzip it so the patent-novelty-check folder containing SKILL.md is available.
  2. 2Place the folder. Move the folder into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder.
  3. 3Check the prior-art report. Confirm that patent/PRIOR_ART_REPORT.md exists, since the skill depends on it.
  4. 4Configure the reviewer. Set up Codex MCP or another reviewer backend and confirm you have access to gpt-6-astra.
  5. 5Make a natural request. Ask for a patentability check or novelty check, or use the trigger phrases in the description.
  6. 6Review the output. Read patent/NOVELTY_ASSESSMENT.md and check the anticipation and obviousness tables against the cited references.

What is an agent skill?

An agent skill is a folder containing a SKILL.md file. The file gives an AI agent a set of instructions for one kind of task, and the agent loads it when a request matches the skill's description. This skill is distributed as a single SKILL.md file inside the skills/patent-novelty-check folder of the source repository.

The skill is designed for Claude Code or another SKILL.md-compatible agent that has the tools it lists.

What the skill makes the agent do

The skill walks the agent through six steps. It first extracts the technical features that make the invention novel and drafts preliminary claim language for two or three independent claims, covering method and system claims. It then tests each claim against every prior-art reference in patent/PRIOR_ART_REPORT.md, asking whether a single reference discloses all claim elements.

If the claim survives that test, the agent builds an obviousness matrix that pairs a primary reference with secondary references and records the motivation to combine them. It then asks a separate reviewer model to act as a patent examiner, and gives a PASS or FAIL view under US, Chinese and European rules. The output is written to patent/NOVELTY_ASSESSMENT.md, covering overall assessment, anticipation and obviousness analysis, reviewer feedback, suggested amendments and risk factors.

  • Step 1: define claim elements and draft preliminary claims
  • Step 2: anticipation (novelty) analysis per reference
  • Step 3: obviousness analysis across reference combinations
  • Step 4: cross-model examiner verification
  • Step 5: jurisdiction-specific PASS/FAIL (US 35 USC 102/103, CN Article 22, EP Articles 54/56)
  • Step 6: write patent/NOVELTY_ASSESSMENT.md

How to install the skill

You can install the skill from the ZIP download on this page. Unzip it so that the patent-novelty-check folder, containing SKILL.md, sits in one of these locations:

For a personal install, place it in ~/.claude/skills/. For a single project, place it in the project's .claude/skills/ folder.

Alternatively, in the Claude apps, open Settings, go to Capabilities, and upload the skill under Skills.

  • Personal: ~/.claude/skills/patent-novelty-check/SKILL.md
  • Project: .claude/skills/patent-novelty-check/SKILL.md
  • Claude apps: Settings → Capabilities → Skills → upload

How to trigger the skill with natural requests

The skill loads when your request matches its description. The source lists these trigger phrases: 专利查新, patent novelty, 可专利性评估, patentability check, or any request to evaluate whether an invention is patentable. You can also pass an invention description or brief path as the argument, since the skill's argument hint is [invention-description-or-brief-path].

Before running, make sure a prior-art report exists at patent/PRIOR_ART_REPORT.md. The skill expects this file as output of a separate /prior-art-search step and does not create it.

  • "Run a patent novelty check on this invention: [short description]"
  • "Is this invention patentable? Use the prior art in patent/PRIOR_ART_REPORT.md."
  • "可专利性评估：请根据现有的对比文件评估这项发明。"

What the skill needs to run

The skill lists several requirements. It needs Claude Code or a SKILL.md-compatible agent with the tools in its allowed-tools list. It needs the prior-art report file, patent/PRIOR_ART_REPORT.md. For the cross-model examiner step it needs Codex MCP (mcp__codex__codex) or another configured reviewer backend, and the file names a model, gpt-6-astra, that you must have access to.

Other reviewer MCP servers (antigravity, grok, claude-review) are listed as optional in allowed-tools. The skill also reads shared reference files from ../shared-references/, which the ZIP on this page does not document as included, so check that they are present in your setup.

  • Claude Code or a SKILL.md-compatible agent
  • patent/PRIOR_ART_REPORT.md from a prior-art search step
  • Codex MCP or another configured reviewer backend, with access to gpt-6-astra
  • Optional: antigravity, grok or claude-review MCP servers

Safety notes: scripts and network access

The skill runs scripts and uses the network. Its allowed-tools list includes Bash(*), WebSearch and WebFetch, and it calls external reviewer models through MCP. Review the SKILL.md file and your MCP configuration before running it, and be aware that invention descriptions and prior-art content may be sent to those reviewer services.

If a selected reviewer is unavailable, the skill records REVIEW_UNAVAILABLE and continues only as an explicitly unreviewed draft. It does not switch to another provider on its own.

  • Bash(*), WebSearch and WebFetch are in allowed-tools
  • Reviewer calls go to the MCP backend you configure
  • Unavailable reviewers are logged as REVIEW_UNAVAILABLE

Tips for better results

Give the agent a precise invention description and the features you consider inventive. The skill's anticipation test is only as good as the claim elements it extracts, and the prior-art report determines which references are checked.

Treat the output as a screen, not a decision. The skill itself notes that actual prosecution may reveal different prior art.

  • Describe the technical features in concrete terms before asking for the check.
  • Make sure the prior-art report covers the closest references before running the skill.
  • Use the obviousness matrix to decide which combinations need a closer look by an attorney.

What people use it for

  • Screening an invention before paying a patent attorney for a formal search and opinion.
  • Building a claim-by-claim anticipation matrix against a set of prior-art references.
  • Creating an obviousness matrix that pairs references and records motivation to combine.
  • Comparing the likely US, Chinese and European novelty and inventive-step outcomes for one invention.
  • Helping a student or patent agent practise a structured novelty and obviousness analysis.

Frequently asked questions

Is the patent-novelty-check skill free?

The source file is published under the MIT licence, which is stated on this page. The skill itself does not mention a price, but it relies on a reviewer model and tools you configure, which may have their own costs.

How do I install the patent novelty check skill?

Download the ZIP from this page and unzip it so the patent-novelty-check folder is in ~/.claude/skills/ or a project's .claude/skills/ folder. You can also upload it in the Claude apps under Settings, then Capabilities, then Skills.

What does the skill need to work?

It needs Claude Code or a SKILL.md-compatible agent, a prior-art report at patent/PRIOR_ART_REPORT.md, and a reviewer backend such as Codex MCP with access to gpt-6-astra. The file lists other reviewer servers as optional.

Can this skill replace a patent attorney?

No. The skill is advisory only and says prosecution may reveal different prior art. It is meant as a first-pass screen before a qualified patent attorney reviews the invention.

Why does the skill say it needs patent/PRIOR_ART_REPORT.md?

The anticipation and obviousness tests check the claims against each reference in that file. The skill does not produce the report itself, so it must come from a separate prior-art search step.

What is the difference between research novelty and patent novelty?

The skill states that research novelty asks whether anyone has published the idea, while patent novelty asks whether a single reference teaches every claim element. Any public disclosure before the priority date counts as prior art.

What should I do if the reviewer is unavailable?

The skill records REVIEW_UNAVAILABLE and continues only as an explicitly unreviewed draft. It does not switch providers automatically, so you need to fix the reviewer configuration to get the cross-model review.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you want a structured first-pass novelty and obviousness screen before paying for a formal search, or if you are learning the method as a student or patent agent. Pick a qualified patent attorney or a formal search service if you need a final patentability decision or filing strategy, since the skill explicitly does not replace them.

Download the skill