agentleFS
hxhx-winhxhx-win/zhuanli-flow/skills/cn-patent-disclosure-draft/SKILL.md

cn-patent-disclosure-draft

Skill4 starsChanged 3 months agoMITView on GitHub

当用户已完成 cn-patent-mainline-analysis 与 cn-patent-prior-art-search,需要在进入正式稿起草前做起草前评审、生成技术版交底书时使用。承载编排器 step 3 的两个子阶段:评审 + 技术交底书。用户分流决策与 Gate A 段(起草前决策 + 确认)由编排器执行。不挑主线、不重做检索、不起草权要。触发词:起草前评审、技术交底书、disclosure draft、pre-draft review。

The file

Files
SKILL.md

起草前评审与技术交底书 skill

承载编排器 step 3 的两个子阶段:评审(review)+ 技术交底书(disclosure-generation)。两类 subagent 共用本 SKILL.md 骨架,按 stage 各自加载对应 references。

适用边界

  • 仅在编排模式下使用;状态文件 patent/<patent-slug>/state/patent-iteration-state.json 必须已含 prior_art_search.status = completed
  • 不重新挑主线(用 cn-patent-mainline-analysis)
  • 不重新检索(用 cn-patent-prior-art-search)
  • 不起草权要(评审通过后由 cn-patent-formal-drafting 起草)
  • 不做起草前决策与 Gate A 确认(由编排器自执行,见 cn-patent-domain-runtime/references/drafting-decisions.md)

子阶段工作流

子阶段视角动作主要产物写入字段
review资深专利代理师 / 起草前评审者:以稳妥起草为目标,对发明点的合规性、因果链、区别特征反例、可面试度做闸前体检,宁可标 revise-required 也不放含糊点过闸派 subagent 隔离执行:subagent 加载 references/review-protocol.md + 读 4 份输入 + 三道复审 + 判定 decision_readiness 软字段 + 生成 proposal_summary_oneliner(≤60 字)+ Write pre-draft-review.md + 回 ≤200 字 summarypatent/<patent-slug>/reviews/pre-draft-review.mdstep_3.pre_draft_review.*(含 verdict + decision_readiness + chain_check + proposal_summary_oneliner)
disclosure-generation研发人员 + 专利工程师双视角:以研发口径把技术方案讲清楚(机制 / 数据流 / 实验),同时以专利工程师口径锚定权利要求范围与从权层次,不替代代理师抠条款派 subagent 加载 references/drafting-discipline.md + assets/disclosure-template.md + 读 4 份输入 + 写技术版交底书主体八节;"题目"/"保护客体"/"权利要求范围"字段允许占位"(待用户确认)"(由编排器在起草前决策阶段填入 state)patent/<patent-slug>/disclosure/disclosure-draft.mdstep_3.disclosure_draft.*

起草前决策(原 3b)与 Gate A 确认由编排器执行,见 cn-patent-domain-runtime/references/drafting-decisions.md。用户分流决策(原 3d)在 step 3 末尾由编排器 AskUserQuestion 五选执行。

出口分支(disclosure-generation 完成后)

disclosure-generation 完成后,不直接进 step-3.post-disclosure-decision,而是先经 step-3.inventor-review:

出口 stage执行者
step-3.inventor-reviewcn-patent-disclosure-review skill(发明人 3 阶段审核)
step-3.post-disclosure-decision由编排器在 inventor-review gate_passed=true 后进入

REQUIRED CONTEXT: 见 cn-patent-disclosure-review/SKILL.md。

修订模式:当 cn-patent-disclosure-review 累积反馈触发整体修订时,编排器以 revision_mode: true + prior_draft_path + inventor_feedback 重新派单 disclosure-generation subagent。subagent 复用本 SKILL.md 的 drafting-discipline.md 准绳。

核心铁律

  • 四档 verdict(go / revise-recommended / revise-required / stop-recommended)是软闸门,不存在硬阻断;最终决策权在用户(编排器侧处理)
  • 唯一硬阻断(review subagent 侧):复审 4 五栏因果链一致性自检中任一 IP 命中 chain_broken → verdict 直接置为 revise-required,跳过软合成。chain_check 字段必须随 verdict 一并写入 state。详见 references/review-protocol.md 「预检阻断硬规则」段
  • 评审 subagent 必须执行 IP 三句话可面试测:对每个独权核心 / 独权步骤 IP,写出"做什么 / 差异 / 因果"三句话;写不出标 interview_failed
  • 评审 subagent 必须生成 proposal_summary_oneliner(≤60 字中文)并写入 state;若任一 IP interview_failed,oneliner 必须为 null,不得编造
  • 评审 subagent 必须输出 decision_readiness 软字段(ready / needs_supplement),判定规则见 references/review-protocol.md
  • 评审与技术交底书都派 subagent;subagent 直接 Write 落盘,主 agent 只收 ≤600 字摘要;铁律:写到磁盘的内容不返回给父 agent
  • 技术交底书主体八节字段严格对齐公司官方模板,不得新增非官方字段、不得调整字段顺序;推荐项、决策依据、对照表、风险登记、特征分层等编排/分析层信息只在上游产物中存在,不进交底书
  • 修订次数上限 3 次(用户在 step 3 用户分流决策选"修订交底书"超过 3 次时编排器提示考虑"调整主线")
  • subagent 不直接 Read PDF / docx / 源码文件 / 截图 / 实验数据 / 外链网页等大体量原始资料;所需信息必须由上游 subagent 在 evidence-matrix.md / mainline-analysis.md / prior-art-search-report.md 中预吸收,缺失项记为待确认事项由编排器在派单前补齐

subagent 入口分路

subagent 启动后,根据主编排器派单 prompt 中的 stage 字段加载对应纪律,不要读另一份:

stage必读纪律必读输入产物写入产物
reviewreferences/review-protocol.mddf-rationale-signals.yml + evidence-matrix.md + mainline-analysis.md + prior-art-search-report.mdpatent/<patent-slug>/reviews/pre-draft-review.md
disclosure-generationreferences/drafting-discipline.md + assets/disclosure-template.md必读列表 6 份(按顺序 Read,前 2 份是模板与纪律,后 4 份是上游产物):(1) assets/disclosure-template.md (2) references/drafting-discipline.md (3) pre-draft-review.md (4) mainline-analysis.md (5) evidence-matrix.md (6) prior-art-search-report.mdpatent/<patent-slug>/disclosure/disclosure-draft.md

所有输入与输出路径必须从主编排器派单 prompt 中"输入产物绝对路径列表"与"输出产物绝对路径"两个字段读取,禁止 subagent 自行猜路径。

review 与 disclosure-generation 共享 mainline-analysis.md / evidence-matrix.md / prior-art-search-report.md 三份输入,由两个隔离 subagent 各自 Read 一次(重复读取是隔离架构的清醒代价,token 成本只发生在 subagent 各自 context 内,不进主编排器 context)。

主编排器派 subagent prompt 契约

主编排器调用 Agent(subagent_type=general-purpose) 派 subagent 时,prompt 必须包含以下三段字段:

stage: review | disclosure-generation
skill_root: <skills_root>/cn-patent-disclosure-draft
patent_root: <patent root 绝对路径>
任务: subagent 内部通过 Skill 工具加载 cn-patent-disclosure-draft SKILL.md,按其「subagent 入口分路」段加载对应 stage 的纪律 references。
输入产物绝对路径列表(按列出顺序 Read,作为 Skill 加载失败时的硬路径兜底):
  - <绝对路径 1>
  - <绝对路径 2>
  - ...
输出产物绝对路径: <绝对路径>

主编排器侧 cn-patent-domain-runtime/references/agent-tool-mapping.md 中 disclosure-draft 派单条目须引用本 SKILL.md 的入口分路段与本契约段,避免两边漂移。

产物路径

  • 评审报告:patent/<patent-slug>/reviews/pre-draft-review.md
  • 技术交底书:patent/<patent-slug>/disclosure/disclosure-draft.md

状态字段回写

  • step_3.stage:每个子阶段开始时更新(review / disclosure-generation / completed)
  • step_3.pre_draft_review.*:评审完成后写入;含 verdict、decision_readiness、chain_check、proposal_summary_oneliner、risk_inputs 数组
  • step_3.disclosure_draft.*:技术交底书完成;含 draft_path
  • step_3.status ∈ {in-progress, completed, stopped};用户分流决策完成后由编排器写 completed

参考资料

  • 评审协议(评审 subagent 必读):references/review-protocol.md
  • 起草纪律(交底书 subagent 必读):references/drafting-discipline.md
  • 交底书模板:assets/disclosure-template.md
  • 类目配置:已迁到 cn-patent-domain-runtime/assets/decision-categories.json,本 skill 不再持有
  • 证据可信度信号 schema:../cn-patent-mainline-analysis/references/evidence-quality-signals-schema.md
  • DF 反例自检规则:../cn-patent-prior-art-search/references/creativity-screening.md(末尾段)

跨平台 fallback

Agent tool / 子 agent 机制是 Claude Code 特性。在 Cursor / Codex 等环境无子 agent 工具时:

  • 评审:主 agent 自行加载 references/review-protocol.md,按其指导直读 4 份输入、直写 pre-draft-review.md
  • 技术交底书:主 agent 自行加载 references/drafting-discipline.md + assets/disclosure-template.md,按其指导直读 4 份输入、直写 disclosure-draft.md;起草完成后用户手动 /compact
  • verbatim 回查原始资料:主 agent 自行 Read 单文件 + 立即丢弃中间态(用户手动 /compact)

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.