patent

Overview of patent skill

What the patent skill does

The patent skill is a patent prior-art and landscape intelligence workflow for Claude-style agent environments. It is built for structured patent searching, not broad legal Q&A. Its main job is to turn an invention, product, target company, or disputed patent into a routed search plan, collect evidence from patent sources, deduplicate patent families, and produce a decision-oriented .docx report with citations, ranked results, strategy notes, and an audit trail.

Best-fit users and legal use cases

This patent skill is best suited for founders, R&D teams, product counsel, IP operations teams, competitive intelligence analysts, and legal-adjacent researchers who need a defensible first-pass search package before attorney review. It supports five distinct work modes: novelty search, freedom-to-operate, competitive landscape, acquisition diligence, and litigation prior-art. That routing matters because each mode asks a different legal or business question.

What makes this different from a generic prompt

A generic patent prompt often mixes search strategies and produces plausible but poorly scoped results. This skill forces a sub-use-case commitment before search, then adapts queries, ranking, and report emphasis. It also includes references for CPC/IPC class follow-up, legal disclaimer discipline, and sub-use-case routing, plus helper scripts for citation tracking, patent-family resolution, and deterministic routing.

Boundaries to understand before install

The skill produces search signal, not legal advice. It should not be treated as a patentability opinion, infringement opinion, validity opinion, or clearance opinion. It is also not designed for trademark, copyright, trade-secret, contract, or general legal research. For Legal teams, the value is better intake and evidence organization before professional review, not replacement of counsel.

How to Use patent skill

patent install and environment requirements

Install from the repository path with:

npx skills add alirezarezvani/claude-skills --skill patent

The skill is located at research/patent/skills/patent. Before relying on it, inspect SKILL.md, then read references/sub_use_case_routing.md, references/cpc_classification_canon.md, and references/legal_disclaimer_discipline.md. The workflow expects web access for Google Patents, Espacenet, USPTO, and optionally Lens.org. Node.js with the docx package is useful for the final Word report, and Lens.org requires your own API key if you want citation-graph signals.

Inputs the patent skill needs

Give the agent enough information to choose the correct route and avoid vague search results. Useful inputs include:

• invention or product description in plain technical language
• intended use case: novelty, FTO, landscape, diligence, or litigation
• target jurisdictions such as US, EP, CN, JP, or global
• known patents, products, assignees, inventors, or competitors
• priority date, launch date, filing target, or acquisition context
• risk setting: strict review versus signal-gathering
• output preference: short memo, full .docx, table-heavy report, or attorney handoff

Weak prompt: “Do a patent search for my AI medical device.”

Stronger prompt: “Use the patent skill for a novelty search. The invention is a wearable ECG patch that uses an on-device transformer model to detect atrial fibrillation and transmit only abnormal segments. Focus on US and EP publications. Known competitors: iRhythm, AliveCor, Philips. I need closest prior art ranked by claim similarity, CPC class follow-up, and a .docx report for attorney review.”

Recommended patent usage workflow

Start by stating the sub-use-case. If you are unsure, ask the agent to run the router conceptually before searching. For novelty, emphasize claim-like invention elements. For FTO, describe the product as sold or planned, including jurisdictions and launch timing. For landscape, name the technology area and competitor set. For diligence, provide the target company and assets. For litigation prior-art, provide the challenged patent and relevant priority date.

After intake, expect the workflow to search multiple sources, record counts, identify CPC/IPC classes, resolve patent families, rank closest art, and produce an audit trail. Review the ranked hits manually before acting on conclusions.

Repository files worth reading first

Read SKILL.md for the end-to-end workflow. Use references/sub_use_case_routing.md to understand why novelty, FTO, landscape, diligence, and litigation searches are not interchangeable. Use references/cpc_classification_canon.md to understand why class-based searching catches terminology gaps. Use scripts/sub_use_case_router.py for route logic, scripts/family_resolver.py for deduplication behavior, and scripts/citation_tracker.py for audit expectations.

patent skill FAQ

Is the patent skill suitable for legal teams?

Yes, when used as a research and triage tool. The patent skill for Legal workflows is most useful for building an organized evidence pack: closest art, patent-family grouping, jurisdiction coverage, FTO flags where relevant, and a reviewable audit log. It should feed attorney analysis rather than replace it.

When should I not use patent?

Do not use it for non-patent IP issues such as trademarks, copyright, trade secrets, licensing language, or contract interpretation. Avoid using it when you need a formal legal opinion, claim chart for litigation filing, prosecution strategy, or final clearance decision without qualified counsel.

How is this better than searching Google Patents manually?

Manual search can be faster for a quick lookup, but it often misses adjacent terminology, class-based results, family duplication, and mode-specific ranking. This skill adds disciplined intake, CPC/IPC follow-up, source tracking, family resolution, and report structure, which are useful when the output must be reviewed or shared.

Is this beginner-friendly?

It is usable by beginners if they can describe the invention and choose a goal. However, users should expect patent terminology such as claims, priority date, assignee, CPC, IPC, patent family, and FTO. Beginners get better results by supplying concrete product details and asking the agent to explain assumptions before search.

How to Improve patent skill

Improve patent results with better invention framing

The most important improvement is turning a broad idea into searchable technical elements. Include components, data flows, materials, steps, constraints, and what is new compared with known systems. For software inventions, describe inputs, model behavior, outputs, deployment context, and performance constraints. For hardware, include structure, arrangement, materials, and operating conditions.

Avoid common failure modes

Common failures include treating FTO like novelty, ignoring jurisdiction and patent status, overcounting the same patent family, relying only on keywords, and missing the priority-date cutoff for litigation prior-art. The repository’s routing reference and family resolver exist to reduce those errors, but the user must still provide the correct business question.

Iterate after the first output

After the first report, do not stop at the executive verdict. Ask for: “expand the top five claim comparisons,” “rerun with broader CPC classes,” “separate active from expired patents,” “show family members by jurisdiction,” or “identify search terms that produced no useful hits.” These follow-ups improve confidence and expose blind spots.

Add local review standards for your team

For production Legal or IP workflows, add internal requirements: preferred jurisdictions, attorney disclaimer wording, required databases, citation format, risk thresholds, and report sections. If your organization uses Lens.org, private docket data, or paid patent tools, document how those sources should supplement Google Patents, Espacenet, and USPTO results.