Skip to content
Get packs

Guide · Agent skill

How to use the Patent Examiner Review agent skill (patent-review)

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

The patent-review skill is an agent skill for Claude Code, or a compatible agent, that makes the agent act as a critical patent examiner for a patent application you are drafting. It sends your claims and specification to an external reviewer model and returns a formal-style office action covering clarity, support, enablement, novelty, obviousness and claim scope. This guide explains what the skill does, how to install it, how to trigger it, and what it needs and risks before you run it.

Patent Examiner Reviewwanshuiyin/Auto-claude-code-research-in-sleep · MIT

Quick start

  1. 1Download and unzip. Download the skill ZIP from this page and unzip it so SKILL.md is available.
  2. 2Place or upload the skill. Put the folder in ~/.claude/skills/ or a project's .claude/skills/, or upload the ZIP under Settings → Capabilities → Skills in the Claude apps.
  3. 3Register the Codex MCP server. Run the claude mcp add codex command, pointing the path to the codex-exec server.py in your ARIS clone.
  4. 4Organise your patent folder. Create patent/ with CLAIMS.md, specification/, figures/numeral_index.md, PRIOR_ART_REPORT.md and INVENTION_DISCLOSURE.md.
  5. 5Ask for an examiner review. Use a request such as 'patent review' or 'examiner review' and name the scope you want checked.
  6. 6Apply fixes and re-review. Work through CRITICAL and MAJOR issues, then let the agent submit the revised text for round 2.
  7. 7Read PATENT_REVIEW.md. Check the issue table, the scores and the recommendations checklist before deciding next steps.

What the patent-review skill makes the agent do

The skill runs a multi-round review of a patent application. The agent first gathers context: all claims, the specification sections, the prior art report and the invention structure. It then sends that material to an external reviewer model, GPT-6-Astra, through the Codex MCP server, with an instruction to act as a senior patent examiner.

The reviewer is asked to cover clarity (112(b)/Art 84), written description and enablement (112(a)/Art 83), novelty (102/Art 54), non-obviousness (103/Art 56), claim scope and specification quality. The source says the output is a formal-style office action with grounds of rejection, suggested amendments and a patentability score from 1 to 10.

The agent then proposes fixes, re-submits the revised text for a second round, and writes a report to patent/PATENT_REVIEW.md with an issue table and scores for each round.

Install the skill (ZIP download or upload)

Download the skill ZIP from this page and unzip it. The skill is a single file, SKILL.md. For a personal setup, place the unzipped folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.

If you use the Claude apps instead of Claude Code, upload the ZIP in the app under Settings → Capabilities → Skills.

  • Personal install: ~/.claude/skills/
  • Project install: .claude/skills/ inside your project
  • Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Set up the Codex MCP server

The skill will not run its review without the Codex MCP server. The source gives this registration command, where the path must point to your own ARIS clone:

  • Replace the path with the location of your ARIS repository clone.
  • The review needs access to the GPT-6-Astra model through the Codex MCP setup, which requires an external API account.
claude mcp add codex -s user -- python3 "$HOME/aris_repo/mcp-servers/codex-exec/server.py"

Prepare the patent folder layout

The workflow expects a specific folder structure. Before you trigger the skill, make sure these files exist, because the skill reads them as inputs. If they are missing, the source says the workflow will struggle.

  • patent/CLAIMS.md: all drafted claims
  • patent/specification/: all specification sections
  • patent/figures/numeral_index.md: reference numeral mapping
  • patent/PRIOR_ART_REPORT.md: known prior art
  • patent/INVENTION_DISCLOSURE.md: invention structure

How to trigger the skill with natural requests

The skill activates when you use phrases from its description, such as 專利審查 (专利审查), patent review, 审查意见, examiner review, or when you ask for critical feedback on your claims and specification. You can also pass an optional argument, a patent directory or scope, as the skill's argument hint describes.

Be specific about the scope. Asking for an examiner review of the claims in patent/ is clearer than a vague request for feedback.

  • "Run a patent review on my draft in patent/ and tell me the rejection grounds."
  • "Do an examiner review of claims 1 to 5 against the prior art report."
  • "Give me critical feedback on my claims and specification before I file."

What the output looks like and what to do next

The skill runs two review rounds. After round 1, the agent sorts issues into CRITICAL (such as 102 rejection, 112 indefiniteness or missing enablement), MAJOR (such as 103 obviousness or weak claim scope) and MINOR (language and numeral consistency). CRITICAL issues must be fixed before the next step, and each fix is shown as an old-to-new claim change with its rationale.

The final report includes a checklist. The source says that if the score is below 5/10 after round 2, you should consider significant rework before filing. The review is advisory, and actual prosecution may go differently.

  • Read the issue table in PATENT_REVIEW.md first, starting with CRITICAL rows.
  • Check every claim amendment against the specification to confirm support exists.
  • Treat the score as a rough signal, not a verdict.

What people use it for

  • Stress-testing independent claims for indefiniteness or missing support before filing.
  • Getting a checklist of likely rejection grounds to discuss with a patent attorney.
  • Checking that reference numerals in the specification match the figures index.
  • Comparing claim scope against a known prior art report to find the closest reference.
  • Tracking every claim change with its rationale in a single review report.

Frequently asked questions

Is the patent-review skill free?

The skill is published under the MIT licence, as shown on this page. The review itself uses GPT-6-Astra through the Codex MCP setup, which the source says requires an external API account, so you should check the cost of that account yourself.

What do I need to install patent-review?

You need Claude Code or a compatible agent with skill support, a configured Codex MCP server, and access to the GPT-6-Astra model through that setup. The source also lists optional reviewer backends such as Antigravity, Grok and claude-review MCP if you change the reviewer routing.

Can the patent-review skill replace a patent attorney?

No. The source describes the output as an AI simulation of an examiner, not legal advice or a substitute for a registered patent attorney. It may miss jurisdiction-specific practice or misstate statute details.

Is it safe to send my unpublished invention to this skill?

The skill sends your full claims and specification to an external model provider through the Codex MCP server. The source lists confidential inventions as a case where you should not use it unless you are comfortable with that. Check your confidentiality obligations first.

Why does the skill say it cannot find my files?

The workflow assumes a specific patent/ layout with CLAIMS.md, specification/ and PRIOR_ART_REPORT.md. If those files are missing or named differently, the source says the skill will struggle, so match the expected structure first.

How does patent-review compare to a general AI prompt for patent feedback?

The skill adds a fixed workflow: a two-round review with a defined examiner persona, a structured issue table with severity levels, and a saved PATENT_REVIEW.md report. A general prompt does not give you that repeatable structure, though it still needs the same care and legal checking.

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 patent-review if you are an inventor, drafter or patent agent who wants a structured stress test of claims before filing, and you already have Codex MCP configured. Pick something else if you need legal advice, if your files do not follow the expected patent/ layout, or if you cannot send confidential claims to an external model provider.

Download the skill