File contents 交底书编写
分步指令在 prompts/包内三个目录 ，不是三个可触发技能。
步骤
文件
Step 1
prompts/intake.md
Step 2
prompts/project_scan.md
Step 3–4
prompts/invention/ · utility_model/ · design/ 挖点
填表 / 线稿
prompts/fill_*、image_gen.md、*_lineart_*.md；外观视图口径 references/design_view_cnipa.md
Step 5
prompts/prior_art_search.md（轻量查新：3～4 个手段短语，公布模式每页 10 条，LLM 摘要精排）
Step 5.5
同文件「D1 锁定与区别特征 Fk 」：主比对钉一篇最接近 + 逐特征表（可 D2 补行）+ 三态门禁 ，落 查新与区别定位_*.md。三态都进 Step 6；不过 仍成文但创造性降级，禁止假 D1 / 空喊创新性强
Step 6
prompts/disclosure_preview.md
Step 7
对应类型 disclosure_builder.md + template_reference.md
Step 8
prompts/disclosure_self_check.md
迭代
iteration_context.md / merger.md / correction_handler.md
旁路 · 保护型 1+N
首篇定稿后 prompts/fence/guardrails.md；用户同意则 decompose.md → design_around.md → matrix.md → plan.md（family.yaml + 专利布局.md）→ score.md（立项后弱校验）→ 确认后 dispatch.md
交付对话
定稿回复末块 prompts/delivery_confirm.md（标题固定 交付后请确认 ）
查新工具：tools/crawl/cnipa_epub_search.py。整仓安装时路径为 skills/patent-disclosure/tools/crawl/cnipa_epub_search.py。著录检索不在本包，禁止 当查新引擎调用。不要 自动进入申请文件；用户点名并给出本目录后，再走 skills/patent-application/SKILL.md。## 交付后请确认prompts/delivery_confirm.md），不先打分。用户说「专利布局 / 专利围栏 / 族树 / 做围栏」可强开，但仍须分解→突围→矩阵再立项。立项由模型直写 outputs/{案件}/fence/专利布局.md（章节见 prompts/fence/plan.md），yaml 只给校验和分件；不要出 HTML、不要用脚本转 md。立项后弱校验默认表 references/scorecards/gbt42748_fence.yaml：作答与门槛回写说明稿「立项校验」；对话按过门槛/临界/未过门槛给推荐或不推荐，不叫高价值达标。口头确认「按 C1、P1 写」后再分件。--type 与 intake 一致；邻域召回：3～4 个手段短语 → EPUB_CLASS_HINT / IPC·LOC → --class（默认 1×1，公布模式每页 10 条）→ LLM 对照摘要精排 再锁 D1。不足 4 条则同分类号回补。第二轮导航失败保留第一轮，不要当成 0 条、不要因此降级 WebSearch。不要切列表模式、不要自建向量库。
线稿、CAD、公式、Word 出图用本包 tools/（browser.py、mermaid_render.py、md_to_docx.py 等）。
python skills/patent-disclosure/tools/fence/check_layout.py --decompose outputs/{案件}/fence/decompose.yaml
python skills/patent-disclosure/tools/fence/check_layout.py --family outputs/{案件}/fence/family.yaml --matrix outputs/{案件}/fence/matrix.yaml
python skills/patent-disclosure/tools/fence/check_scorecard.py -i outputs/{案件}/fence/scorecard.yaml --case-dir outputs/{案件}
1 --- 2 name: patent-disclosure 3 description: Patent Disclosure 4 --- 5 6 # 交底书编写 7 8 分步指令在 **`prompts/`**（本包内）。发明 / 实用新型 / 外观是**包内三个目录**，不是三个可触发技能。 9 10 | 步骤 | 文件 | 11 |------|------| 12 | Step 1 | `prompts/intake.md` | 13 | Step 2 | `prompts/project_scan.md` | 14 | Step 3–4 | `prompts/invention/` · `utility_model/` · `design/` 挖点 | 15 | 填表 / 线稿 | `prompts/fill_*`、`image_gen.md`、`*_lineart_*.md`；外观视图口径 `references/design_view_cnipa.md` | 16 | Step 5 | `prompts/prior_art_search.md`（轻量查新：3～4 个手段短语，公布模式每页 10 条，LLM 摘要精排） | 17 | Step 5.5 | 同文件「**D1 锁定与区别特征 Fk**」：主比对钉一篇最接近 + 逐特征表（可 D2 补行）+ **三态门禁**，落 `查新与区别定位_*.md`。三态都进 Step 6；**不过**仍成文但创造性降级，禁止假 D1 / 空喊创新性强 | 18 | Step 6 | `prompts/disclosure_preview.md` | 19 | Step 7 | 对应类型 `disclosure_builder.md` + `template_reference.md` | 20 | Step 8 | `prompts/disclosure_self_check.md` | 21 | 迭代 | `iteration_context.md` / `merger.md` / `correction_handler.md` | 22 | 旁路 · 保护型 1+N | 首篇定稿后 `prompts/fence/guardrails.md`；用户同意则 `decompose.md` → `design_around.md` → `matrix.md` → `plan.md`（family.yaml + 专利布局.md）→ `score.md`（立项后弱校验）→ 确认后 `dispatch.md` | 23 | 交付对话 | 定稿回复末块 `prompts/delivery_confirm.md`（标题固定 **交付后请确认**） | 24 25 查新工具：`tools/crawl/cnipa_epub_search.py`。整仓安装时路径为 `skills/patent-disclosure/tools/crawl/cnipa_epub_search.py`。著录检索不在本包，**禁止**当查新引擎调用。 26 交底交付后**不要**自动进入申请文件；用户点名并给出本目录后，再走 `skills/patent-application/SKILL.md`。 27 首篇 Step 8 之后，围栏只写在交付回复 **`## 交付后请确认`** 第 3 条（见 `prompts/delivery_confirm.md`），不先打分。用户说「专利布局 / 专利围栏 / 族树 / 做围栏」可强开，但仍须分解→突围→矩阵再立项。立项由模型直写 `outputs/{案件}/fence/专利布局.md`（章节见 `prompts/fence/plan.md`），yaml 只给校验和分件；不要出 HTML、不要用脚本转 md。立项后弱校验默认表 `references/scorecards/gbt42748_fence.yaml`：作答与门槛回写说明稿「立项校验」；对话按过门槛/临界/未过门槛给推荐或不推荐，不叫高价值达标。口头确认「按 C1、P1 写」后再分件。 28 `--type` 与 intake 一致；邻域召回：3～4 个手段短语 → `EPUB_CLASS_HINT` / IPC·LOC → `--class`（默认 1×1，公布模式每页 10 条）→ **LLM 对照摘要精排** 再锁 D1。不足 4 条则同分类号回补。第二轮导航失败保留第一轮，不要当成 0 条、不要因此降级 WebSearch。不要切列表模式、不要自建向量库。 29 30 线稿、CAD、公式、Word 出图用本包 `tools/`（`browser.py`、`mermaid_render.py`、`md_to_docx.py` 等）。 31 32 ```bash 33 python skills/patent-disclosure/tools/fence/check_layout.py --decompose outputs/{案件}/fence/decompose.yaml 34 python skills/patent-disclosure/tools/fence/check_layout.py --family outputs/{案件}/fence/family.yaml --matrix outputs/{案件}/fence/matrix.yaml 35 python skills/patent-disclosure/tools/fence/check_scorecard.py -i outputs/{案件}/fence/scorecard.yaml --case-dir outputs/{案件} 36 ```
handsomestwei/patent-disclosure-skill/tree/main/skills/patent-disclosure commit ca67dbe9e9
Frequently asked questions How do I install the Patent Disclosure skill? Run npx skillmds@latest add handsomestwei/patent-disclosure in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.
What does the Patent Disclosure skill do? Patent Disclosure It is listed under Coding & Dev Tools on SkillMD.
Is Patent Disclosure safe to use? This skill has not completed SkillMD's automated safety review yet. Independent scanners report: SkillSpector: PASS, Skill Scanner: PASS. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.
Which AI agents work with Patent Disclosure? This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.
Is Patent Disclosure free to use? Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.
Who published Patent Disclosure? handsomestWei (@handsomestwei) published this skill. Their other Agent Skills are listed on their SkillMD profile.
handsomestWei