用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/lawyerllsx-dotcom/Legal-Workflow-Chain --skill legal-citation-comprehensive命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
| name | legal-citation-comprehensive |
| description | 当用户需要【规范化法学引注/脚注】时使用。触发词:引注、脚注、引用格式、注释、参考文献格式。 输出:符合规范的脚注。不做:检索法条、事实核查。 |
加载后静默检查:若上下文存在 [Skill接力: legal-citation-comprehensive] → 接力模式。
| 用户意图 | 应交由 |
|---|---|
| 检索法条/案例(找法律依据) | legal-research |
| 撰写/起草法律文书正文 | 项目实际文书生成技能(case-pipeline E 层 / 02-文书草稿.md) |
| 核查文书事实准确性 | legal-fact-checker |
| 识别/梳理争议焦点 | dispute-issue-identification |
| 三段论推理/涵摄分析 | deductive-reasoning |
| 整理证据材料到模板 | evidence-catalog-generator |
| 编辑 .docx 文件格式 | docx-editing |
| 仅解决法条竞合/规范冲突 | conflict-resolution |
判定口诀:用户只关心"脚注格式/引注规范/引用格式"→ 本 skill;用户要"写正文/查法条/核事实/改格式"→ 不触发。
加载后立即执行:检查用户请求是否同时需要本 skill 不覆盖的能力。
如果用户请求包含"核查/检查案例是否真实/法规是否过时"等事实核查,但引注来源未经核实 → 必须先由 legal-fact-checker 核查:
[Skill接力: legal-fact-checker]
接力上下文: 需要先核查引注中引用的法规效力和案例真实性
待办: 核查完成后回到 legal-citation-comprehensive 格式化脚注
如果用户请求包含"撰写/写作/起草/写"等创作任务(如"写一份案例分析/法律意见书+脚注"),本 skill 只负责脚注格式,正文必须先由项目实际文书生成技能(case-pipeline E 层 / 02-文书草稿.md;项目外无该技能时由上游按项目规则生成草稿)完成:
[Skill接力: 项目实际文书生成技能]
接力上下文: 用户要求撰写法律文书,本 skill 只负责引注格式化,正文生成任务需先由文书生成技能完成
待办: 正文草稿完成后回到 legal-citation-comprehensive 格式化脚注
注意:legal-fact-checker 只核查不写作,写作任务不得接给它。
在格式化引注时,引注中的法条或案例引用出现异常时的分流:
| 引用异常 | 接力到 | 说明 |
|---|---|---|
| 引用的法条已失效或被修改 | legal-research | 检索现行有效版本后回到引注格式化 |
| 引用案例真实性存疑 | legal-fact-checker | 核查案号/当事人/文书类型后回到引注格式化 |
| 同一观点引用了多个冲突法条 | conflict-resolution | 确定优先引用哪个后回到引注格式化 |
判定关键词:写、撰写、起草、帮我写、帮我做一份
引注前必须区分两件事,不得混为一谈:
缺失要素标 [待补: 要素名],不得猜测出版社、年份、案号、页码或 URL;不得把「格式上看似完整」的引注当作「来源已核验」的引注。引注中出现失效法条或案例真实性问题时,先接 legal-research / legal-fact-checker 处理,不得直接格式化成看似完整的脚注。
完成后输出两部分:引注清单(已完成的引注)与未核验清单(标 [待补]/[待核实] 的引注及查找路径)。
若本技能在复杂深链中作为终点环节(环节 6),须把引注状态回写 pipeline/02-文书草稿.md(或对应报告文件),而不是只在聊天中返回文本。
□ 接力目标是否正确?(引注来源未核实 → fact-checker;法条失效 → legal-research;多法条冲突 → conflict-resolution) □ "待办"字段是否包含了需要核验的具体法条编号或案例名称? □ 本 skill 的核心任务(引注格式化)是否确实已完成或已确认需前置处理?
执行自检通过后在接力标记前输出确认行:✅ 接力自检通过: 目标正确([skill-name]) | 待办完整 | 本skill任务已完成
[Skill接力: next-skill-name]
接力上下文:
- 结论: {一句话核心结论}
- 依据:
- 法律依据: {名称/条号/原文或摘要}
- 事实依据: {事实 + 来源 + 状态}
- 来源标签: {[元典法规]/[元典案例]/[verify-需对照原文]/[待核实]}
- 核验状态: {已详情核验/仅候选/未核验}
- 检索日期/版本: {...}
- 置信度: {高/中/低 + 理由}
- 待核验项: {逐项列出;无则填"无"}
待办: {下一技能必须完成的动作、输入文件和验收条件}
注意:本 skill 通常是法律工作流的最后一站(格式化工序),后置接力场景较少。如用户的引注要素缺失且无法从已有材料提取,应先接力到 legal-research 补全来源信息。
诊断/格式化完成后,末尾提示用户:如需在原文书中直接修正脚注或引注,可让我用 docx-editing 打开编辑。
This skill is the citation brain. It diagnoses and generates citation text. It does not edit Word files directly; use docx-editing to apply corrections to existing .docx files.
Primary rule: never invent missing bibliographic facts. If a required element is absent and cannot be extracted from a provided source, mark it as [待补: 要素名] and tell the user where to find it.
Completeness rule: the simplified templates are not the whole handbook. Before giving a final answer for any non-trivial citation, consult a legally obtained copy of the citation handbook or a user-provided rule index. This public repository does not redistribute third-party handbook PDFs or OCR dumps. Use references/citation_rules.json only as a fast execution layer when the user or local installation provides it.
Weak-agent guardrail: if you are unsure, do not improvise. Follow references/operator_guardrails.md exactly. A citation answer is not complete unless it states the source type, missing elements, lookup path, suggested format, and relevant handbook rule numbers.
references/citation_rules.json for required elements, templates, and handbook anchors when that optional local reference file is available.references/source_lookup_guide.md.python scripts/self_test.py after changing this skill or before distributing it. Full handbook coverage tests require locally supplied reference data.For a first-pass report, run:
python scripts/citation_diagnose.py --text "王名扬《美国行政法》第18页"
Or:
python scripts/citation_diagnose.py footnotes.txt --json
The script provides classification, present/missing elements, source lookup hints, and a template-based corrected citation when local rule data is available. Treat script output as a first pass; verify edge cases against the handbook.
When responding to a user, use this shape unless they ask for something else:
识别结果:
- 类型:
- 置信度:
已有要素:
- ...
缺失/需确认要素:
- ...
去哪里找:
- ...
建议格式:
...
注意:
- ...
作者:《书名》,出版社年份年版,第X页。作者:《论文名》,载《期刊名》年份年第X期,第X页。作者:《文章名》,载编者主编:《书名》,出版社年份年版,第X页。《法律名称》第X条第Y款第Z项。 Use Arabic numerals.《文件名称》(发文字号)。同前注〔X〕,第Y页。 or a clear short form. Do not use 前引, 同上, supra, or Ibid. for Chinese sources.参见; indirect or conceptual borrowing: use 参见; secondary citation: use 转引自.The rule index is mandatory for careful work:
For any source type not explicitly represented in citation_rules.json, search the user-provided handbook or rule index by rule number or source name. If the OCR text is unclear, verify against the legally obtained original document.
references/README_REFERENCE_DATA.md: explains which optional reference files are intentionally not redistributed.references/handbook_rule_index.json: optional local full machine-readable index of rules 1-150. Load/search this for exhaustive coverage if the user provides it.references/handbook_rule_index.md: optional readable version of the full index with raw line ranges.references/citation_rules.json: optional machine-readable rule source. Load this first if present.references/missing_elements_matrix.md: quick checklist for missing elements by type.references/source_lookup_guide.md: where to find missing bibliographic facts.references/operator_guardrails.md: strict operating rules for other AI agents.references/common_errors.md: common mistakes and course-specific footnote traps.references/citation_handbook_structured.md: structured digest of the handbook.assets/Law_Journal_Citation_Handbook_2025.pdf: original handbook PDF for final verification when needed.When a user has their own legally obtained copy of the handbook PDF, prefer the deterministic local builder before asking an AI to improvise the reference files:
python3 scripts/build_reference_data.py
The builder checks whether the PDF is searchable, extracts local Markdown, builds the 1-150 rule index, writes citation_rules.json, and then runs validation. If it refuses to overwrite existing files, rerun with --force only after the user confirms they want to rebuild local reference data. If extraction is incomplete, run:
python3 scripts/build_reference_data.py --print-ai-contract
Then use the printed contract to repair the local files.
When a user rebuilds handbook reference data from their own PDF/OCR, do not trust the generated files merely because they exist. If the builder was not used, run:
python3 scripts/audit_reference_data.py
python3 scripts/self_test.py
audit_reference_data.py checks required files, rule count, continuous rule numbers, category ranges, raw line ranges, common rule hints, OCR/mojibake markers, and citation_rules.json shape. If it reports ERRORS, stop and repair the reference data before claiming full handbook coverage.
docx-editing: use after this skill has produced or approved footnote text; handles DOCX insertion with formatting preservation.legal-research: use when missing law or case source facts require legal database retrieval (元典 MCP).如已搭建个人知识库,可引用方法论页对齐既有经验;无知识库时直接使用本 skill,不影响核心功能。