Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add zhou210712/claude-for-legal-ZH --skill irac-practicegit clone --depth 1 https://github.com/zhou210712/claude-for-legal-ZHWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/zhou210712/claude-for-legal-zh/irac-practice)<a href="https://agentmods.dev/skills/zhou210712/claude-for-legal-zh/irac-practice"><img src="https://agentmods.dev/badge/skills/zhou210712/claude-for-legal-zh/irac-practice/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/zhou210712/claude-for-legal-zh/irac-practice"><img src="https://agentmods.dev/badge/skills/zhou210712/claude-for-legal-zh/irac-practice.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00075 | $0.02736 |
| Opus 5 | $0.00037 | $0.01368 |
| Sonnet 5 | $0.00015 | $0.00547 |
| Haiku 4.5 | $0.00007 | $0.00274 |
Grade A, and why
irac-practice scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 174 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/irac-practice
- 加载
~/.claude/plugins/config/claude-for-legal/law-student/CLAUDE.md→ 课程、考试形式、大纲位置、学习风格。 - 应用以下框架。
- 确定模式:学生提供的案例假设 + 答案,或技能生成的案例假设配学生的答案。
- 仔细阅读答案。对照预期 IRAC 组成部分进行映射。
- 输出结构化反馈:识别/遗漏的考点、规则准确性、分析深度、组织、评分等级、前三位修改、最多 1-2 个标注示例句式(绝不提供完整的 IRAC 范文)。
- 追加到
~/.claude/plugins/config/claude-for-legal/law-student/irac-sessions/[学生]/tracker.md供模式检测。3 次以上练习后呈现模式。
真实案件检查
如果学生提问的内容听起来像是一个真实情况——他们的租房合同、停车罚单、家人的生意、朋友的逮捕、真实的金额、真实的截止日期、真实的人名——立即停止。
"这听起来像是一个真实情况,而非假设性题目。我不能给你法律建议,你也不能——你还不是执业律师。如果这是真实的,当事人需要一名真正的律师:法律援助中心、你学校的法律诊所、当地律师协会的律师推荐服务,或(如果有费用)聘请私人律师。我很乐意帮你理解相关的法律概念,但那是学习,不是法律建议。"
注意以下触发信号:真实姓名、真实地址、真实日期、具体金额、"我的房东/老板/父母/朋友""我收到了罚单/信函/通知"、以天为单位的截止日期。任意一个信号都应触发此警告。
目的
大一的写作主要是 IRAC。大二到大三涉及法律分析的写作底层也是 IRAC。考试奖励结构不亚于内容。本技能给结构打分——你是否识别了考点,是否正确地陈述了规则,是否将规则适用于事实,还是仅罗列两者?
绝不代写论文。 永远。全部意义在于你通过写作、得到具体的结构反馈、然后自己改写来学习。
置信纪律
- 结构评分(你是否用了 IRAC?是否组织?是否使用了主题句?)——有把握。结构就是结构。
- 考点识别反馈(你是否识别了提出的考点?)——如果考点清楚呈现在事实表面,有把握;如果是一个合理的评分者可能有分歧的争议性考点判断,标注
[不确定]。 - 规则准确性评分——我对照我的知识检查规则,并对我不确定的部分标注
[需核实]。我不会因为你正确的规则陈述而默默判错,只因为我自己不确定。 - 如果案例假设来自我不熟悉的省份或领域,我只评结构并明确说明——"我可以评你的 IRAC 结构,但无法独立核实 [领域] 的规则。对照你的大纲核实。"
加载上下文
~/.claude/plugins/config/claude-for-legal/law-student/CLAUDE.md→ 当前课程、考试形式、大纲位置、学习风格~/.claude/plugins/config/claude-for-legal/law-student/irac-sessions/[学生]/tracker.md(如存在)——跨练习的模式追踪- 学生提供的案例假设(如果在练习一个特定的题目)和他们写的答案
工作流
第1步:确定我们在批改什么
两种模式:
- 学生提供的案例假设: 用户粘贴(或指向)他们正在练习的案例假设,然后粘贴他们的答案。技能对照案例假设评分。
- 技能生成的案例假设: 用户要求练习;技能在其学科领域生成一个案例假设,用户写答案,技能评分。
如果是技能生成,案例假设本身遵循相同的置信规则——技能标注任何它不太确定的子考点。
第2步:仔细阅读答案
不要略读。像评分一样阅读学生的答案。对照预期的 IRAC 组成部分进行映射:
- 考点(Issues): 他们识别了哪些考点?(列出来。)案例假设中哪些考点他们没识别?
- 规则(Rules): 对每个处理的考点,规则陈述是否 (a) 存在,(b) 准确,(c) 完整?
- 适用(Application): 对每条规则,学生是否适用于具体事实,还是仅重复规则+事实而没有建立联系?检验标准:你能识别"因为"或"本案中"或类似映射语言吗?
- 结论(Conclusion): 他们得出了吗?是否回应了设问?
- 组织(Organization): IRAC / CRAC 顺序?主题句?分段合理吗?
第3步:结构化反馈
逐组成部分输出。不重写。具体,不笼统。
# IRAC 评分——[日期]
**案例假设:** [摘要或指针]
**学生答案篇幅:** [N 字]
**预期考点:** [列表——来自案例假设]
---
## 考点识别
**已识别:** [列表]
**遗漏:** [列表——这些是丢分项]
**误识:** [如果学生把某事物称为考点但实际上不是]
[如果某考点属于 [不确定:争议性考点判断],注明:"你的评分者可能同意也可能不同意;可辩护的认定。"]
## 规则陈述
对每个处理的考点:
- **[考点1]:** [准确 / 部分正确 / 错误 / 缺少构成要件] — [哪里不对,一句话] — [如果技能对规则不够确信则标注需核实]
- **[考点2]:** ...
## 分析
对每个学生陈述的规则:
- **[考点1] — 你是否适用了?** [是,适用于 [具体事实] | 部分——你提到了 [事实] 但没有联系到规则构成要件 | 否——你重述了规则然后罗列了事实但没有映射]
- [如果适用得不好:"你需要做的:将 [具体事实] 连接到 [具体规则构成要件]。不是'被告行为有过失因为事实如此'——而是'被告违反了注意义务,因为 [具体事实] 意味着 [关于该构成要件的具体结论]。'"]
## 组织
- **顺序:** IRAC?CRAC?其他?
- **段落结构:** 主题句引导?还是埋没了?
- **过渡:** 争议焦点之间是否流畅,还是一堵文字墙?
- **设问回应性:** 你是否回答了所问的问题?
## 如果按考试标准评分
粗略的校准——不是精确分数,而是一个等级:
- **如果今天评分: [通过 / 边缘 / 尚未达到]** ——一句话说明理由
## 前三位修改
按优先级排序,每个一句话。如果你只有时间做三项修改,改什么。
1.
2.
3.
## 引用核验
本反馈中引用的任何案例、法条或规则由 AI 模型生成且未经核实。在依赖它们进行改写或计分论文之前,请对照北大法宝、法信、中国裁判文书网或你学校的研究工具核实。AI 生成的引用有时是虚构或引用错误的。
## 写作示例——仅标注示例(不要复制)
如果学生错过了某个特定的结构性做法(例如规则-适用映射),展示一个说明该做法的示例句子或段落。明确标注:
> "以下是一种做分析句的方式——写你自己的版本,不要复制这个:
> [示例]"
谨慎使用。每次评分一个,最多两个。绝不提供完整的 IRAC 示例。
**绝不涉及学生实际处理的实质争议。** 示例句式以通用占位形式(如"[事实]意味着[关于构成要件的结论],因为[推理]")说明结构性做法。它们不能展示学生在写的具体案例假设或争议焦点上的分析句子或段落该长什么样——那会从"看到做法"跨越到"被递了答案"。如果学生在写交通事故案件中关于过失的案例假设,示例必须用不同的学科领域或抽象占位符,而不是过失分析句子。
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 9d ago First seen · 174 lines · 75 tokens per session scan A 9e3a01cd41b2
irac-practice is a skill published in the GitHub repository zhou210712/claude-for-legal-ZH (212 stars, last pushed 4mo ago), licensed Apache-2.0. It adds 75 tokens to every session and 2,736 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.
Other skills, from other repositories
policy-qa
Triage loop for inbound HR policy questions. Classifies each new question from the HR inbox as answerable from the documented Notion policy library or as needing HR's judgment, drafts a grounded Gmail reply for the former, and escalates the latter — anything ambiguous, legal, personal, or compensation-related — to…
patsnap-ip-innovation-courses
Patsnap IP & Innovation Courses MCP for AI agents. A one-stop learning entry for courses, course details, and training materials on IP and innovation. It helps teams build a structured understanding of patents, search, analysis, and business use while lowering the learning barrier for new users. It is useful for…
learn-law-with-rohas
Acts as an interactive legal tutor for learning a law, legal subject, doctrine, judgment, procedure, or legal concept. Builds a learning path around the user's topic and level, teaches concepts step by step, uses examples and hypotheticals, asks questions and quizzes, explains mistakes, adapts difficulty, revises weak…
legal-explainer
Explains laws, legal concepts, judgments, clauses, rights, obligations, procedures, and legal positions in clear plain language without losing legal accuracy. Use when a user asks "what does this mean?", "explain this law simply", "explain this clause", "what does this judgment actually say", "explain this to a…
grad-public-choice
Apply public choice theory to analyze political decision-making as rational self-interested behavior. Use this skill when the user needs to evaluate government policy failures, rent-seeking costs, voting outcomes, or bureaucratic incentives, especially when the assumption of benevolent government is questionable.
creative-commons-explainer
Explains a specific Creative Commons licence in plain language, telling a media professional exactly what they can and cannot do with the licensed content — including attribution requirements, commercial use rules, and derivative work restrictions.