universal-exam-cram-coach

name	universal-exam-cram-coach
description	帮助学生在临考前进行结构化极速复习：解析课程资料/大纲/重点，按章节生成 wiki 知识库与标准题库，组织针对性刷题与判分，并记录复习进度和错题。当用户即将考试、需要快速复习计划、练习题、错题复盘或考前小抄时使用（关键词：期末/备考/复习/刷题/划重点/错题；exam, cram, study plan, quiz, review）。不适用于长期学习规划、与考试无关的写作或编程任务。
license	MIT
metadata	{"version":"4.2","author":"ZeKaiNie"}

Universal Exam Cram Coach — Root Router

This skill turns the agent into a last-minute exam-cram coach whose core memory carrier is a chaptered LLM wiki: it ingests the student's materials into a wiki + standard question bank, teaches chapter by chapter, quizzes and grades from the bank only, replays mistakes, and compiles a pre-exam cheat sheet — with persistent source, readiness, and progress gates. This root file is a language-neutral ROUTER, not the manual: it only dispatches to the language packs and the control layer below.

Language dispatch

Read the canonical study_state.json.language value and load the matching compatibility entry plus its per-skill wording pack BEFORE emitting any student-visible output:

• 中文 → locales/zh/SKILL.md plus the selected sub-skill's zh wording pack under locales/zh/skills/
• English → locales/en/SKILL.md plus the selected sub-skill's en wording pack under locales/en/skills/
• 双语 → compose the zh and en wording block by block, with zh first and a > EN: mirror for each block (composition rules in docs/language-policy.md)

zh, en, and bilingual are accepted input aliases only; update_progress.py set --language normalizes them and they MUST NOT be treated as persisted state values. Unset language → this is the first conversation: the merged first-ask (mode × time budget × language, persisted in one update_progress.py set call) decides it; default English unless the student opened in Chinese. set --language can switch mid-session; from the next turn on, load the new pack.

Control layer (behavior)

Behavior lives in the modular skill collection, not in this file. Main skill / orchestrator: skills/exam-cram/SKILL.md. The 9 sub-skills:

Sub-skill	Role
exam-ingest	Build the workspace from materials (wiki + quiz bank + progress)
exam-tutor	Lazy per-chapter teaching (seven-step walkthroughs, visual-first gate)
exam-study-guide	Compile persisted chapter sources into formula-readable, self-contained HTML/PDF study material
exam-quiz	Select & grade from the bank only (fail-closed on invisible figures)
exam-review	Replay mistakes & confusions
exam-cheatsheet	Pre-exam cheat sheet
exam-audit	Read-only workspace health check
exam-help	Quick reference
confusion-tracker	Concept-confusion tracking

Generic agents that will not read the full rules: AGENTS.md (one-screen condensed contract).

Install & run essentials

• Engine scripts live under scripts/: ingest_course.py is the normal one-command materials-to-validated-workspace entry; exit 10 means compilation succeeded but content readiness is blocked and must enter the typed ingest_review.py workflow. ingest.py is the lower-level compiler for an already-built raw payload. update_progress.py owns study_state.json (the single source of truth — study_progress.md is a generated view, never hand-edit it); select_questions.py / select_hard_questions.py are the official question selectors.
• Workspace file contract (wiki / quiz bank / state / asset metadata): docs/file-format.md.
• Language policy (single-language purity, EN canonical vocabulary, persisted canonical values): docs/language-policy.md.
• Install in Claude Code at ~/.claude/skills/universal-exam-cram-coach/ or project-local .claude/skills/universal-exam-cram-coach/; load matrix for other hosts: docs/agent-portability.md.
• PDF capabilities differ by host; use the audited, no-silent-download routing table in docs/pdf-capability-adapters.md.
• Artifact output is user-controlled: missing/legacy artifact_mode defaults to chat (normal teaching + state/notebook, no automatic HTML/PDF); only an explicit standing visual choice or a one-shot chapter handout/chapter PDF request invokes exam-study-guide. A cheatsheet PDF routes to exam-cheatsheet; if “make a PDF” does not identify the artifact, ask which artifact once. Agents never infer a subscription tier. Persist the standing choice with update_progress.py set --artifact-mode chat|visual.