How to use the claims-drafting agent skill for patent claims

What the claims-drafting skill makes the agent do

Claims define the legal scope of protection in a patent. The skill tells the agent to treat the claims as the most critical part of the application, with the specification, figures and abstract existing to support them.

The skill works from three files in a patent/ folder: an invention disclosure with core, supporting and optional features, a prior art report, and a novelty assessment with suggested amendments. It sets the target jurisdiction from the disclosure or from your request.

• Chooses claim style: CN (two-part form with 其特征在于), US (open 'comprising' form) or EP (two-part form with 'characterised in that').
• Drafts independent claims, typically a method claim and a system claim, and for Chinese utility models only apparatus claims.
• Drafts 5-10 dependent claims per independent claim, each adding a meaningful limitation.
• Checks each independent claim against a quality list, including antecedent basis, no relative terms without definition, and no result-to-be-achieved limitations.
• Builds a claim-to-specification mapping to check that every claim element is described in the specification.
• Sends the draft to an external reviewer model for revision rounds, up to three by default.

Install the claims-drafting skill

Download the ZIP for skills/claims-drafting from the directory page and unzip it. The skill is a single file, SKILL.md, so the folder you unzip should contain that file.

Place the folder in one of two locations. For a personal install, use ~/.claude/skills/. For a single project, use the .claude/skills/ folder inside that project.

Alternatively, upload the skill in the Claude apps under Settings → Capabilities → Skills.

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

How to trigger the skill with natural requests

You do not need a special command. The skill's description lists trigger phrases, so the agent loads it when your request matches. The listed phrases are 撰写权利要求, draft claims, 写权利要求书 and claim drafting, and it also loads when you ask to create patent claims.

Give the agent the invention disclosure path and the target jurisdiction in the same message, because the skill expects a path argument and uses the jurisdiction to choose the claim format.

• 'Draft claims for the invention in patent/INVENTION_DISCLOSURE.md, targeting the US.'
• '撰写权利要求，目标司法管辖区为中国，专利类型为实用新型。'
• 'Draft CN, US and EP claims from my disclosure and prior art files.'

Requirements before you start

The skill needs more than the skill file. It expects a patent/ folder with three files, and it loads shared reference files from ../shared-references. Those reference files are not included in the skill's ZIP, so you must supply them for the skill to follow its cited patent writing and format guides.

The skill can also call optional external reviewer servers (Codex, Grok, Antigravity or claude-review) listed in its allowed tools, along with a reviewer configuration. Without them, the external review step cannot run as written.

• Claude Code or another SKILL.md-compatible agent
• patent/INVENTION_DISCLOSURE.md, patent/PRIOR_ART_REPORT.md and patent/NOVELTY_ASSESSMENT.md
• Shared references: patent-writing-principles.md, patent-format-cn.md, patent-format-us.md and patent-format-ep.md in ../shared-references
• Optional: reviewer MCP servers and a reviewer config
• The skill runs scripts and uses the network, as described in the safety notes

Safety notes: scripts and network access

The skill's allowed tools include Bash(*), file reading and writing, web search, web fetch and several MCP reviewer tools. In practice, this means it can run commands, read and write files in your project, and send claim text to external reviewer models. Check what you are sharing before you start, especially unpublished invention details.

The skill is MIT licensed, and its description states that its output is a draft. Do not treat the output as a filing-ready claim set.

• Review the allowed tools before enabling the skill in a sensitive project.
• Keep unpublished invention material in a project you control.
• Confirm which reviewer backend is configured, since the skill sends claim text to it for review.

Tips for better claim drafts

The quality of the draft depends mostly on the inputs. The skill draws its core features from the 'Core Inventive Concept' section of the disclosure, so a clear, structured disclosure produces a more useful first draft than a loose description.

Ask for one jurisdiction at a time when you are learning the skill. The output is easier to check, and you can then request the other formats.

• Name the core inventive features explicitly in the disclosure.
• Make sure the prior art report lists the references you want the claims to avoid.
• Use the novelty assessment's suggested amendments as input for narrowing claims.
• Have a qualified patent attorney review every claim before filing.