patent-pre-evaluation-report

Patent Pre-Application Evaluation Report

Overview

Use this skill to turn a technical proposal into a structured Chinese patent pre-application evaluation report, output as a complete HTML file whose CSS and layout exactly match the CUMT-IP-PRE-2026-0001.html reference template (deep-blue/gold color scheme, A4 portrait cover page, section-number circles, progress bars, conclusion quick-view page).

Treat the report as an iterative artifact: draft from the proposal, run real PatSnap/智慧芽 searches via the installed MCP tools, incorporate evidence, reassess risks, and update the conclusion.

Load references/report-workflow.md when drafting or revising a report.

Core Workflow

Step 1 — Clarify Input

• If the user provides only a technology idea, extract: invention title, field, problem, solution, technical effects, application scenario, and likely inventors/applicant.

• If the user provides a disclosure or prior report, preserve existing facts; mark uncertain or missing content as 待补充.

• If the user asks to iterate, identify the specific sections affected by new evidence.

Step 2 — Extract Technical Structure (MCP)

Call the following MCP tools from hub-mcp-gateway-novelty-search in order:

1. novelty_summary — extract tech problem / solution / efficacy (three elements)

2. novelty_feature_extract — classify technical features and build the feature table

3. Convert strongest feature combinations into numbered 查新点 (3–6 points). Keep each 查新点 concrete: component, data, rule, parameter, workflow, model, or interaction logic — not broad effects.

Step 3 — Build Search Strategy (MCP)

Call in order:

1. novelty_keywords_extract — generate Chinese/English keyword groups + IPC classification

2. novelty_keywords_extend — expand synonyms, hypernyms, hyponyms for each keyword

3. novelty_query_planner — plan multi-round Boolean retrieval queries

Output of this step populates the 检索范围与策略 chapter (databases, languages, date range, keyword table, IPC table, Boolean query strings).

Step 4 — Execute Real Search (MCP)

For each 查新点, call:

1. novelty_search_agent (preferred, full-pipeline AI search entry point) — OR split into:

   • novelty_semantic_search (semantic similarity search)

   • novelty_patent_search (keyword Boolean search)

   • novelty_paper_search (academic paper search)

2. novelty_fetch_patent_data — fetch title, abstract, assignee, publication date, abstract figure for top hits

3. novelty_abstract_figure_similarity — evaluate abstract figure similarity against the input solution (optional, when figures are available)

Record for every search: database, query string, date range, total hits, screened count, and selected references. Cite these in the report. Do not invent publication numbers, dates, applicants, or similarity scores.

Step 5 — Feature Comparison (MCP)

1. novelty_feature_comparison (or novelty_feature_comparison_async + novelty_cc_result for large sets) — compare each close reference feature-by-feature against 查新点

2. novelty_rl_predict — predict novelty / inventiveness score for top references

3. novelty_report_generate — generate the comparison narrative paragraph

Classify each feature as: 相同 / 相近 / 部分公开 / 未见 / 待复核. Do not claim novelty from absence of evidence alone; phrase as "在当前检索样本中未见相同公开".

Step 6 — Evaluate & Iterate

• Map each search result to one or more 查新点.

• Classify relevance: 密切相关 / 相关 / 一般相关.

• When evidence weakens a 查新点, suggest narrowing, recombination, dependent-claim placement, or additional experimental support.

• When evidence supports a strong conclusion, explain which distinguishing features carry novelty/inventiveness.

• For market/industry context (转化价值评估 chapter), optionally call novelty_website_search.

Step 7 — Generate HTML Report

After all MCP results are collected, generate a single self-contained HTML file with the following requirements:

Style requirements (must match reference template exactly):

• CSS variables: --primary: #1a3a6b, --primary-light: #2a5298, --accent: #c8a94b, --accent-light: #f5e6b8, --danger: #c0392b, --warning: #e67e22, --success: #27ae60, --info: #2980b9, --gray: #f4f6f9, --border: #d0d7e3

• Cover page: A4 portrait ( width: 210mm; max-width: 210mm; min-height: 297mm; margin: 0 auto), linear-gradient(145deg, #0d1f4a, #1a3a6b, #2a5298) deep-blue gradient background; see Cover Page Layout below

• Main content area: width: 210mm; max-width: 210mm; margin: 0 auto; padding: 24px 0 60px (no left/right padding so content width equals cover width)

• Section number circles: #1a3a6b background, white text

• Section titles: border-bottom: 2px solid #c8a94b gold underline

• Left-border highlights: border-left: 4px solid #c8a94b

• Table headers: #1a3a6b background, white text

• Similarity progress bars: red (≥70%) / orange (40–69%) / green (<40%)

• Conclusion quick-view page: deep-blue gradient background + gold-border cards

• Fixed print button: bottom-right corner

• Footer: #1a3a6b background

Cover Page Layout (竖版A4封面，从上到下):

1. Top deep-blue bar ( height: 14mm, linear-gradient(90deg, #0d1f4a, #1a3a6b, #2a5298))

2. Gold decorative line ( height: 4px, gradient gold)

3. Institution header: left = "中国矿业大学" (17pt, bold, letter-spacing 3px) + English full name; right = circular gold-border "矿" badge

4. Divider line (blue-gold gradient)

5. Confidentiality badge (gold-border ellipse label: "内部保密 · 申请前预评估 · PatSnap 智慧芽支持")

6. Chinese main title (21pt, deep-blue bold, centered, two lines)

7. English subtitle (gray uppercase)

8. Report type box (deep-blue gradient background + gold text "专利 申 请 前 评 估 报 告")

9. Basic info table (2-col 8-row: report number, date, unit, field, inventors, IPC, confidentiality level, recommendation)

10. Red risk alert bar (red background, key risk conclusion + recommendation)

11. Confidentiality notice (light-gray dashed border, small text, usage restrictions)

12. Bottom deep-blue bar: left = institution name, right = report number + date (gold text)

Appendix card style (附件区块样式):

• Section title uses same gold-underline style as other chapters

• Each appendix item rendered as a rounded card: border: 1px solid #d0d7e3, border-radius: 8px, background: #fff, padding: 20px, margin-bottom: 12px

• Left gold circle badge: background: #c8a94b, color: #fff, font-weight: bold, border-radius: 50%, width: 32px, height: 32px, display inline-flex, align-center

• Appendix title: color: #1a3a6b, font-weight: bold, font-size: 1.05em

• Description text: color: #666, font-size: 0.9em, margin-top: 6px

Report chapters (in order):

1. 封面 & 基本信息（竖版A4，见 Cover Page Layout）

2. 数据安全与保密声明（固定模板，见下方说明，放在政策背景之前）

3. 政策背景（固定模板文字，含职称改革政策，见下方说明）

4. 技术方案要点（来自 Step 2 MCP 结果）

5. 查新点与查新要求（来自 Step 2 MCP 结果）

6. 检索范围与策略（来自 Step 3 MCP 结果）

7. 检索结果与相关文献（来自 Step 4 MCP 真实检索结果）

8. 逐项技术特征比对（来自 Step 5 MCP 比对结果）

9. 可专利性风险判断（来自 Step 5 RL评分 + AI分析）

10. 非正常申请风险排查（AI生成）

11. 申请文件质量评估（AI生成）

12. 转化价值评估（AI生成，可选调用 novelty_website_search）

13. 申请策略建议（AI综合生成）

14. 综合结论快览页（AI综合，含评分卡片）

数据安全与保密声明（固定内容，每份报告必须包含）：

• 章节编号圆圈内仅显示 🔒 emoji，不含文字（避免溢出圆圈）

• 章节标题：数据安全与保密声明

• 使用与正文其他章节完全一致的 section / section-header / section-num / section-title CSS class，不加内联样式覆盖

• 4条声明内容（以表格/卡片形式展示）：

  1. 内部使用与权限控制 ：本报告仅供中国矿业大学科研院知识产权管理办公室、项目发明人及经授权代理机构使用，未经授权不得外传。

  2. 脱敏检索与最小披露 ：外部检索仅使用关键词、IPC分类号和抽象技术特征，不上传技术交底书原文、核心参数完整表或未公开实验数据。

  3. 平台安全说明 ：智慧芽（PatSnap）安全与合规能力可参见官方网站： https://www.zhihuiya.com/security-center。

  4. 留痕与复核 ：提交、检索、修改、导出、盖章等环节应在知产办流程中留痕；AI辅助结论须经知识产权管理人员复核后使用。

政策背景（固定内容，每份报告必须包含，共5条政策）：

1. 教育部、科技部《关于规范高等学校SCI论文相关指标使用 树立正确评价导向的若干意见》（教科技〔2020〕2号）

2. 教育部《关于加强高校有组织科研 推动高水平自立自强的若干意见》（教科技〔2022〕1号）

3. 国务院办公厅《专利转化运用专项行动方案（2023—2025年）》

4. 人力资源和社会保障部、教育部《关于深化高等学校教师职称制度改革的指导意见》（人社部发〔2020〕100号） ：明确将专利成果转化情况纳入职称评审，鼓励以专利转化实绩替代论文数量要求，推动"以用促创"。

5. 教育部《破除"唯论文"不良导向若干措施》及配套政策 ：支持将发明专利授权数量、许可转让收益、产学研合作纳入职称评定，以专利转化金额及经济效益作为职称晋升依据。

附件章节（必须生成，紧接第14章之后，缺失任何一个视为报告不完整）：

附件1：检索式完整记录

• 内容：各查新点在智慧芽全球专利数据库中使用的完整检索式，含布尔逻辑运算符、字段限定和IPC分类号限定；以及CNKI/万方补充检索式（如未执行则标注"待填"）

• 数据来源：Step 3 novelty_query_planner 输出 + Step 4 实际执行的检索式

附件2：密切相关专利文献题录及摘要

• 内容：D1—DN所有密切相关专利的完整题录（申请号、公开号、申请人、发明人、IPC分类、公开日、摘要全文）

• 数据来源：Step 4 novelty_fetch_patent_data 返回结果；期刊/学位论文未经CNKI/万方检索则标注"待补充"

附件3：相似专利清单（含智慧芽相似度）

• 内容：P1—PN所有相似专利列表，含公开号、申请人类型、IPC分类、智慧芽相似度得分、关联查新点编号

• 数据来源：Step 4 novelty_search_agent / novelty_semantic_search 返回的相似度评分；不得手动填写评分

附件4：技术特征比对明细

• 内容：各查新点所有技术子特征（F1-1、F1-2…FN-M）与最接近对比文献的逐项对照，含相同点、差异点说明及比对结论

• 数据来源：Step 5 novelty_feature_comparison 输出；与正文第八章保持一致

附件5：委托人提供资料清单

• 内容：发明人提交的技术交底书、实验数据、图纸、论文草稿等材料目录（用户未提供则标注"待补充"）

• 数据来源：Step 1 用户输入整理

附件6：政策文件参考目录

• 内容：报告引用的政策文件列表，包括但不限于：教科技〔2020〕1号、国务院办公厅《专利转化运用专项行动方案（2023—2025年）》、国家知识产权局令第77号、人社部发〔2020〕100号、教育部破除唯论文配套政策、江苏省专利申请预审规范及教师职称评价相关文件

• 数据来源：固定政策模板 + 正文第三章政策背景引用

Output path: @session/reports/[报告编号].html Use files.begin_write → repeated files.append → files.finish_write for large HTML files (never write the entire file in one payload).

Step 8 — Final Decision

Give one of: 建议申请 / 修改后申请 / 暂缓申请 / 不建议申请. Include: application type, independent-claim focus, dependent-claim candidates, evidence gaps, and next actions. Add review caveat: AI辅助结论及PatSnap检索结果，正式申请前须经知识产权专员或专利代理人复核。

Output Rules

• Write primarily in Chinese unless the user asks otherwise.

• Professional report tone, not marketing copy.

• Use tables for: search strategy, close references, feature comparison, patentability risk, abnormal application risk.

• Preserve provenance: cite search date, database, query, result count, screened count, and evidence source for every search conclusion.

• Mark uncertain bibliographic data as 待复核; never invent publication numbers, applicants, dates, or similarity scores.

• Keep confidentiality language visible when the report is based on unpublished technology.

• Final output must always be a complete HTML file saved to @session/reports/, not a Markdown block in chat.

• 附件1—附件6必须全部生成 ，每个附件用独立卡片渲染，缺失任何一个附件视为报告不完整。

• 数据安全与保密声明必须生成 ，位于政策背景章节之前。

• section-num 圆圈内仅放数字或单个emoji ，不放文字，避免溢出。

References

• references/report-workflow.md: report section skeleton, MCP tool call sequence, PatSnap evidence schema, risk scoring guidance, HTML style guide, and iteration checklist.

使用前配置

本 Skill 依赖智慧芽开放平台 MCP 服务：

• 完成安装、初次使用时需进行自检，参见 README.md

• 用户需完成账号授权，并确保 Agent 环境已启用对应 MCP 工具

• 若未完成配置，本 Skill 只能提供分析框架，无法检索实时数据或生成基于数据库的结论

• 缺少MCP配置时，引导用户参照 README.md 在 [ open.zhihuiya.com ]( https://open.zhihuiya.com/ ) 获取MCP。