Hi, I'm Jon. I'm the founder of ArcPrime. Before this, I was patent counsel at Meta and Fish & Richardson. I've spent my career doing patent work end to end: drafting claims from disclosures, reviewing applications, searching prior art, responding to office actions, building claim charts, filing continuations, and making portfolio decisions across thousands of patents.

These prompts are how I actually do patent work. Drafting claims from a messy disclosure, figuring out which continuations to file, mapping claims to a competitor's product, deciding what to prune. These are real workflows I've run across real patents and real portfolios. They work for me.

They won't replace your judgment. A prompt can give you a strong first draft of claims, a solid prior art analysis, a continuation strategy worth discussing. But you still need to review, refine, and make the final call. The work's not done when the output lands. But a lot of it is, and you're starting from a much better place than a blank page.

Two caveats:

1. Domain. These prompts come from my practice, which is mostly software and hardware. If you work in pharma, biotech, or other fields, you'll want to adjust the claim drafting conventions and legal frameworks to match your domain.
2. Depth. These are the single-prompt versions, designed for easy use. In production at ArcPrime, these run as multi-stage pipelines with anti-hallucination checks, specification cross-referencing, and output validation. The multi-stage versions are where the last mile of accuracy comes from.

Fork them. Improve them. Make them yours.

Who this is for:

• Patent attorneys and agents who want to 100x their throughput on drafting, prosecution, and portfolio analysis
• In-house patent counsel making strategic decisions about large portfolios: pruning, continuations, categorization
• Patent professionals at any level who want to see what AI can actually do for patent work, with real prompts that produce real output

If you're in-house counsel applying AI to patent work, I run a WhatsApp group of 40+ practitioners sharing what works. Send me an email at jon@arcprime.com or connect with me on LinkedIn to join.

1. Pick a prompt from the Prompt Catalog below
2. Open the prompt.md file and copy the prompt
3. Replace all {{PLACEHOLDER}} values with your patent data
4. Paste into any capable LLM

Want to see one in action first? Every prompt has a Try it link, a pre-filled version using our sample patent and sample disclosure. Just copy and paste into any LLM.

🌐 = requires web search    🧪 = pre-filled example you can paste into any LLM

All prompts use {{PLACEHOLDER}} format for user inputs. Each prompt's documentation includes a table describing what to put in each placeholder. Common placeholders:

• Use full text, not summaries. These prompts work best with complete patent text (full claims, full specification). Truncated inputs produce lower-quality outputs.
• More context is better. When analyzing against a product, include technical specifications, architecture details, and documentation -- not just marketing copy.
• Iterate. Run the prompt, review the output, and re-run with adjustments. Consider modifying the product description or adding context the model missed.
• Model selection. These prompts work with any capable LLM. For complex tasks (claim drafting, claim charts), stronger models produce noticeably better results.
• Verify everything. LLM outputs are analytical aids, not final work product. Always have a qualified patent professional review the output.

Some prompts (marked with 🌐) work best when the LLM can search the internet — for example, to find prior art references or gather technical details about an accused product. If your LLM has built-in web search, it can search directly. Otherwise, paste your own search results into the {{WEB_SEARCH_RESULTS}} placeholder. Each prompt's documentation explains what to search for.

These prompts are available as installable skills for AI coding agents. Once installed, you can invoke them directly from your agent (e.g., /patent-prompts:claims-drafting).

Install as a plugin:

/plugin install patent-prompts@arcprime-ip/patent-prompts

Or add manually to a project:

git clone https://github.com/arcprime-ip/patent-prompts.git
claude --plugin-dir ./patent-prompts

After installation, all prompts are available as slash commands:

Codex automatically reads the AGENTS.md file in this repository. Clone the repo into your project or working directory:

git clone https://github.com/arcprime-ip/patent-prompts.git

Then reference the prompts in your Codex sessions — Codex will use the AGENTS.md instructions to locate and apply the relevant prompt files.

The SKILL.md format is an open standard supported by multiple agents. Clone the repo and point your agent at the skills/ directory, or copy individual SKILL.md files into your agent's skills directory.

These prompts are written for US patent practice (USPTO procedures, 35 U.S.C. references, MPEP conventions). Many of the underlying concepts transfer to other jurisdictions, but the specific legal frameworks, claim drafting conventions, and procedural references are US-centric.

We welcome contributions adapting these prompts for EPO, CNIPA, JPO, KIPO, and other patent offices. See CONTRIBUTING.md for how to submit.

These prompts are provided for educational and informational purposes only. They do not constitute legal advice. Patent law is complex and jurisdiction-specific. Always consult a qualified patent attorney or agent before making decisions based on LLM-generated patent analysis. The authors and contributors are not responsible for any actions taken based on outputs generated using these prompts.

MIT License. See LICENSE for details.