- Load
~/.claude/plugins/config/claude-for-legal-zh/law-student/CLAUDE.md→ course, instructor, exam format, syllabus. - Apply the following workflow.
- Receive past exam questions (PDF, pasted text, or file path). Confirm sample size.
- Analyze each set of past exam questions: format, subject coverage, question style, case fact density, recurring traps.
- Cross-exam pattern analysis—what is stable, what varies.
- Generate predictions in combination with the current syllabus: subject weightings, format, instructor preferences, review priorities.
- Write to
~/.claude/plugins/config/claude-for-legal-zh/law-student/exam-forecasts/[课程]/forecast-[YYYY-MM-DD].md. Define as weight heuristics, not deterministic predictions.
Purpose
Every teacher's exam paper has a fingerprint. The same case assumption structures recur. The same traps return. The same subject proportions reappear. Students with past exam questions study smarter; those without study harder. This skill analyzes the past exam questions you have and reveals the patterns.
It is not magic. It is prediction, not a definitive answer. The skill cannot tell you exactly what is on the exam paper—it can only tell you what has appeared on past exam papers and what may reappear based on syllabus coverage.
Confidence Discipline
- Pattern analysis (which subjects appear, how many questions per topic, the ratio of policy questions vs. statutory application)—confident when the exam questions are clearly before me.
- Inferences about likely emphases for this year's exam—
[不确定]is the default state; these are predictions, not certainties. Express explicitly as "Based on the [N] past exam questions you shared, [topic] appeared [M] times. Your exam may emphasize it, or the teacher may rotate the tested points—use this as a weighting reference for allocating review time, not a deterministic prediction." - If there are only 1-2 past exam questions, state clearly—any pattern inferred from 1 exam question is noise.
- If the instructor is new (no past exam questions), the skill cannot predict. Say so directly; fall back only to syllabus-based "these are the subjects covered."
Load Context
~/.claude/plugins/config/claude-for-legal-zh/law-student/CLAUDE.md→ current course, exam format, syllabus (if available)- User-provided past exam questions (PDF, pasted text, path)
- Optional: the current course's syllabus (for "what has been taught so far")
If the uploaded past exam questions have an instructor name, use it to match patterns(exam questions from the same instructor are the highest-signal input).If not, match by subject and structure. Do not ask the user to enter the instructor's name—use what is already in the materials. If the user volunteers it in conversation, that is fine; do not prompt for it.
Workflow
Step 1: Receive Materials
- Which course are we predicting for?
- How many past exam questions does the instructor have?
- Are they from the same course, or different courses by the same instructor?
- Are any of them variants with open-book/closed-book/different formats that differ from the typical format of your exam this time?
- Your syllabus for this course?
If there are fewer than 3 past exam papers: mark as insufficient sample. Pattern inference is weaker. If the exam papers come from different courses: some patterns can transfer (question style, policy vs. legal provision ratio); subject-specific patterns cannot transfer.
Step 2: Read each past exam paper
For each past exam paper:
- Format (number of questions, length, time limit, open/closed book)
- Subject coverage (which topics were tested, and what proportion each accounted for)
- Question style (case analysis, deep dive into a single disputed issue, policy discussion, multiple-choice style, mixed)
- Case fact density (fact-intensive hypotheticals, fact-sparse focusing on legal provisions, or fact-free policy prompts)
- Recurring traps (e.g., the instructor always hides a jurisdictional issue in seemingly clean case facts; the instructor always asks about exceptions rather than rules)
- Policy vs. legal provision ratio
- Special structures (mixed essay questions + multiple choice, moot court scenarios, etc.)
Step 3: Cross-exam-paper pattern analysis
Summarize what is consistent across the exam papers:
Stable patterns (appearing in most/all past exam papers):
- Subject weighting (e.g., "consideration and variation consistently account for 30% of the exam score")
- Question style (e.g., "always 1 long case analysis + 2 short hypotheticals")
- Instructor preferences (e.g., "always tests third-party rights even if it is a minor topic in class")
Variable patterns (appearing in some but not all):
- Policy discussion (e.g., "appeared 2 times out of 4 exam papers — usually when there is a policy-intensive topic in the second half of the semester")
- Open book vs. closed book differences
- Take-home vs. in-class differences
Notable absent patterns:
- Topics taught in class but never appearing in past exam papers — do not skip these, but do not weight them either
- Topics appearing in past exam papers but not in your current syllabus — may not return
Step 4: Make predictions for this exam
Title — required, the first line of the prediction, whether in chat or in a saved file. According to the plugin configuration ## Outputs, every study output carries a uniform study note title. Predictions are study outputs. Do not omit, rewrite, or reposition the title. The title is not a disclaimer that students can ask to have removed; it is the identity marker of the output, preventing the prediction from being mistaken for a definite exam question or legal advice:
STUDY NOTES — NOT LEGAL ADVICE（学习笔记 — 非法律建议）
Combine pattern analysis with the current syllabus:
STUDY NOTES — NOT LEGAL ADVICE（学习笔记 — 非法律建议）
# 考试预测——[课程 / 教师]——[日期]
**已分析的历年考题：** [N] 份
**样本置信度：** [不足（<3） / 中等（3-5） / 强（6+）]
**注意事项：** [例如，"其中一份历年考题是开卷期末考；你本次是闭卷。模式迁移不完整。"]
---
## 科目权重（历史数据）
| 主题 | 历年考题权重（平均） | 在当前教学大纲中？ | 预测权重 |
|---|---|---|---|
| [主题1] | [%] | [是/部分/否] | [加重 / 持平 / 减轻] |
## 题型预测
- **可能格式：** [X 个案例分析 + Y 个简答 + Z 个论述，或类似]
- **案例事实密度：** [事实密集 / 稀疏 / 混合]
- **设问风格：** [一个宽泛设问 / 多个具体设问 / 逐项子问题]
## 需警惕的教师偏好
- [主题A] — 出现在 [M / N] 份历年考题中。权重为其在教学大纲中占比的 3-5 倍。
- [主题B] — [模式]
- [陷阱模式] — 如"在其他方面干净的案件事实中隐藏管辖权问题"
## 本学期讲授但历史上很少考查的主题
[列表——不要跳过，但也不要过度加权]
## 复习重点建议
基于历年考题模式和当前教学大纲覆盖：
**重点：** [可能构成考试核心的主题——占复习时间的 40-50%]
**中等：** [辅助主题——30-40%]
**兜底：** [讲授但历史上代表性不足的主题——10-20%，以防万一]
## [不确定——定性说明]
本预测来自 [N] 份历年考题。教师会变化。教师会轮换。历年重点的主题在教学大纲变化时可能被淡化。将其作为分配复习时间的权重启发式，而非确定性预测。考卷将包含意外。
Step 5: Output location
Write to ~/.claude/plugins/config/claude-for-legal-zh/law-student/exam-forecasts/[课程]/forecast-[YYYY-MM-DD].md. Versioned — if the student obtains another past exam paper during the semester, rerun and append.
Skill interactions
- outline-builder： Prediction weight guides outline depth decisions — go deep on key topics
- flashcards： Predicted key topics generate more memory cards
- bar-prep-questions： Unrelated to bar exam preparation (the bar exam has its own prediction model); exam prediction applies to course-specific final exams
- irac-practice： Use predicted topics as subject areas for IRAC practice hypotheticals
What this skill does not do
- Predict specific questions. Past exam papers show patterns; they do not show tomorrow's exam questions.
- Work without past exam papers. If you do not have that instructor's past exam papers, the skill cannot predict — fall back to "here is what the syllabus covers, study those."
- Replace everything on the review syllabus. Prediction is weighting, not exclusion. Skipping a topic because it has been historically underrepresented is exactly how students get burned.
- Account for changes you don't know about. If the instructor shifted emphasis this year (e.g., emphasized a new case in lecture), the skill can't see that change unless you tell it.
- Works reliably on 1-2 past exams. Insufficient sample. Flag it.