flashcards

/flashcards

Load ~/.claude/plugins/config/claude-for-legal-zh/law-student/CLAUDE.md → current course, weak subjects, outline position.
Apply the following framework.
Route by flag:
- --generate: Build cards from the source (outline path, notes, textbook) according to the card-writing rules. Write to ~/.claude/plugins/config/claude-for-legal-zh/law-student/flashcards/[科目]/cards.md。
- --drill(default): prioritize due cards + new cards; show the question, wait for the answer, show the answer, accept self-assessment, update the memory bucket + next review time.
- --review: Browse the card deck by memory bucket.
- --stats: Progress snapshot; flag stuck cards and suggest oral training.
- --session <n>: Focus on practicing N cards, prioritizing previous wrong answers + due cards; append the results to study-plan.yaml → session_history。
Apply confidence discipline: annotate each card generated from unsourced knowledge with [需核实]。

Real case check

If the student asks something that sounds like arealsituation — their rental contract, parking ticket, family member's business, friend's arrest, real amounts, real deadlines, real names — stop immediately.

"This sounds like a real situation, not a hypothetical question. I cannot give you legal advice, and neither can you — you are not a practicing lawyer yet. If this is real, the party needs a real lawyer: legal aid center, your school's legal clinic, your local bar association's lawyer referral service, or (if there is a fee) hire a private lawyer. I am happy to help you understand the relevant legal concepts, but that is studying, not legal advice."

Watch for the following trigger signals: real names, real addresses, real dates, specific amounts, "my landlord/boss/parents/friend" "I received a ticket/letter/notice", deadlines in days. Any one signal should trigger this warning.

Purpose

Outlines are for synthesis; memory cards are for memorization. The bar exam and most law school exams reward rapid rule recall. This skill generates cards from your outline (or notes or textbook excerpts), trains with light spaced repetition, and tracks which ones stick and which do not.

Not a full SRS system. A simple Leitner-style memory bucket. Good enough for studying, good enough for light maintenance. If you want Anki, use Anki; this is for when you want quick training in chat.

Confidence discipline

The same rules as other content-generation skills:

If cards are generated from a source you provided (outline, notes, textbook excerpt), the card's questions and answers come from that source. Confident.
If cards are generated from my knowledge without a source, I annotate each card that states a rule I am not completely certain of with [需核实：规则——对照来源确认]. You should verify it before recording it as a learning objective on a card.
If I am unfamiliar with a field, I generate fewer cards rather than making things up. 8 good cards are better than 20 of which 5 are wrong.

Load context

~/.claude/plugins/config/claude-for-legal-zh/law-student/CLAUDE.md → current course, weak subjects, existing outline
~/.claude/plugins/config/claude-for-legal-zh/law-student/flashcards/[科目]/cards.md(if it exists) (incremental build)
User-provided sources (outline path, notes, textbook excerpt) (if any)

Modes

Flags:--generate | --drill | --review | --stats | --session <n>(default: prompt)

`--session <n>` — Focused practice with N cards

When the student says "let's do 5 contract law cards" or runs /law-student:session 合同法 5 --flashcards .

Load ~/.claude/plugins/config/claude-for-legal-zh/law-student/study-plan.yaml(if it exists) and read that subject's session_history。
Priority: cards previously marked wrong > due cards > new cards.
Run N cards one by one following the --drill process.
At the end of the practice, append the results to study-plan.yaml → session_history：

session_history:
  - date: 2026-05-08
    subject: 合同法
    type: 记忆卡片
    n_cards: 5
    right: 3
    partial: 1
    wrong: 1
    stuck_topics: [合同的订立-要约与承诺]

If there is no study-plan.yaml, write to ~/.claude/plugins/config/claude-for-legal-zh/law-student/session-history.yaml。

`--generate` — Create cards

Input:

Subject (course name or topic)
Source (syllabus path, notes, or "use my existing syllabus in ~/.claude/plugins/config/claude-for-legal-zh/law-student/CLAUDE.md")
Optional: target number of cards (default 10-20 per session)

Card structure:

### 卡片 [N]
**Q：** [问题——一个概念，一张卡片]
**A：** [答案——规则，一句话或两句话]
**来源：** [大纲章节、教材页码、课堂笔记日期]
**记忆桶：** 新
**上次复习：** —
**下次复习：** [今天日期]
**备注：** [可选——区分、例外、陷阱]

Card writing rules:

One card, one concept. "Elements of tort liability" becomes 4 cards, not 1.
The front is a question, not a topic. "Elements of tort liability" is not good. "What are the four elements of tort liability?" is good.
The back is a rule, not a paragraph. If the answer requires a paragraph, split it into multiple cards.
Cite the source so it can be re-verified during training.

Citation verification. When cards are generated from my knowledge rather than from a source you pasted, the rules on the back and any cited cases/statutes are AI-model-generated and unverified. Before committing a card, verify it against your syllabus, textbook, or research tools (PKULaw, Faxin, China Judgments Online). A wrong card practiced to mastery is worse than no card.

`--drill` — Study training

Priority:

下次复习 <= 今天 and memory bucket != mastered cards
New cards not yet attempted
If there are no due cards and no new cards: ask the user whether they want to review mastered cards (to prevent forgetting decay)

Training process for each card:

Show the question. Wait for the answer.
The user answers (or enters "skip" / "don't know")
Show the answer.
User self-assessment:正确 / 部分正确 / 错误 / 不知道
Update the memory bucket + next review according to the table below:

Self-assessment	Memory bucket change	Next review
Correct	Promote one level (new → learning → review → mastered)	+1d new, +3d learning, +7d review, +21d mastered
Partially correct	Keep the current bucket	+1d
Wrong	Demote one level (review → learning; learning → new; new stays new)	Today +4h
Don't know	Demote one level	Today +4h

`--review` — Browse card decks

Show all cards in a subject. Group them by memory bucket. Useful for scanning deck contents and manually adjusting card contents.

`--stats` — Progress snapshot

For each subject: total number of cards, memory bucket distribution, due today, reviewed this week. Highlight any cards that have bounced back to the "new" bucket more than twice — these are the ones that need /law-student:socratic-drill Stuck concepts for oral training.

Interaction with other skills

outline-builder： After building or expanding an outline, propose generating memory cards from the new material
socratic-drill： If a card is wrong 2+ times, route it to /law-student:socratic-drill for oral deep understanding—for concepts you don't actually understand, memory cards aren't enough
bar-prep-questions： Bar exam prep subjects with poor memory card stats get higher weight in objective question training

Storage

flashcards/
└── [科目]/
    └── cards.md

One file per subject. Cards are Markdown. Memory buckets/review metadata are inline within each card. Not optimal for very large card decks (>500), but sufficient for typical law school deck sizes.

What this skill does not do

Replace Anki. If you already have a memory card habit, keep it. This is for when you want to train in chat without switching apps.
Fabricate cards to meet a quantity target. If I can only confidently generate 8 cards from your sources, you get 8. Filling it with a large number of [需核实] guesses is worse than a smaller deck.
Enforce study discipline. Missed review days accumulate; the skill only shows what is due. You decide whether to train.
Teach you the rules. Cards are for training what you have already learned. If a card keeps being wrong, the problem is upstream—use /law-student:socratic-drill or reread the source.