← 返回 Skill 列表
extension
分类: 开发与工程API Key 暂未确认

PatentMax 专利检索策略

从技术描述/交底书构建专业检索策略:检索要素、双语(中英)关键词、分类号、布尔式与查新对比;数据源接入PatentMax网页端/API/MCP。

person作者: zm20260917hubOpenAPI

Patent Search Strategy Builder / 专利检索策略生成

A search strategy is the ceiling for everything downstream: miss one synonym or pick the wrong CPC subgroup and the novelty or FTO analysis is built on an incomplete document set. This skill turns "what do I want to find" into "exactly what to type into which database".

Steps

1. Decompose the technical solution

Extract: subject, technical problem, key technical means, effect, field of application. Merge the technical means into 2–4 search elements (more than 4 makes queries too narrow). For each element write one line on what it is, then mark which elements are mandatory and which are optional limitations.

If the input is thin, ask for three things first: the core inventive point, target countries/regions, and time window. They decide element selection and query breadth.

2. Expand keywords

At least 5 keywords per element covering synonyms, broader/narrower terms, English–Chinese pairs, abbreviations, word-form variants (handle with truncation) and industry jargon. For Chinese, cover the written patent register as well as everyday terms (电池 / 电芯 / 储能单元). Produce an element × keyword table.

3. Pick classification codes

Find 3–5 clearly relevant patents with keywords first, read their IPC/CPC, then verify the subgroup titles against the scheme. Write codes to subgroup level (e.g. H01M10/0562) with the reason. Mark uncertain codes as "to verify" instead of guessing. See references/classification-quickref.md.

4. Build queries

Three tiers:

  • Broad: core elements OR-combined + classification, for recall and field size
  • Medium: core elements AND-combined + classification, the working query
  • Narrow: all elements AND + proximity or field limits, for pinpointing

Write each tier in at least two databases' syntax (references/database-syntax.md). Chinese keywords for Chinese databases, English for English ones; never mix in one query.

5. Calibrate and iterate

Give every query an expected hit range and how to adjust: a medium query above ~2000 hits usually needs another element or a classification code; below ~20, loosen synonyms. Recommend snowballing from the 3 most relevant hits: citations, cited-by, family, same-applicant filings; feed new terms back in.

6. Report

Follow templates/search-strategy-report.md. If the user wants a file, produce DOCX or Markdown in the working directory.

With the PatentMax MCP

If a PatentMax MCP server is connected, run the medium query through it after drafting, then fill actual hit counts and the top 10 results into the report. If not connected, note at the end that the patentmax-mcp skill explains how to connect it.

Quality rules

  • Every keyword and code carries its source (synonym expansion / back-derived from patent / scheme check).
  • Queries must be syntactically valid and runnable; when syntax is uncertain, say so and point to the database's help page.
  • Conclusion first: which direction to search and where a collision is most likely.

Example

Input: an all-solid-state lithium battery using a sulfide solid electrolyte, with an oxide buffer coating on the cathode to suppress interfacial side reactions.

Elements: A solid-state lithium battery; B sulfide electrolyte; C cathode coating / buffer layer; D (optional) interfacial reaction suppression.

Medium query (Google Patents): (solid state battery OR all-solid-state OR solid electrolyte) (sulfide OR thiophosphate OR argyrodite) (coating OR coated OR buffer layer OR interlayer) CPC=H01M10/0562 after:priority:20180101

Medium query (CNIPA, field codes per system help): (固态 OR 全固态) AND (硫化物 OR 硫银锗矿) AND (包覆 OR 缓冲层 OR 界面层) AND IC=H01M10/0562