ai-exam-coach

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

AI Exam Coach

Overview

概述

Create Vietnamese AI practice exams for AI Thực Chiến-style review, then grade submissions and update the learner knowledge base. This skill handles study repository setup, question generation, answer keys, rubrics, feedback, and proficiency tracking; it does not fabricate learner scores, expose hidden source paths, or change unrelated project files.

创建适用于AI Thực Chiến式复习的越南语AI练习考试，然后评分提交内容并更新学习者知识库。该技能可处理学习仓库搭建、题目生成、答案密钥、评分标准、反馈和熟练度跟踪；不会伪造学习者分数、暴露隐藏源路径或修改无关项目文件。

Default Behavior

默认行为

If the user asks to create a practice exam, generate 20 questions by default.
If the user supplies parameters, follow them: question count, topic, track, difficulty, duration, output path, answer-key visibility.
Use Vietnamese with full diacritics unless the user explicitly requests another language or no-diacritic text.

Never generate Vietnamese exam content as ASCII/no-diacritic text. Use

Đề ôn tập

Mã đề

Câu hỏi

Đáp án

Giải thích

, not

De on tap

Ma de

Cau hoi

Dap an

Giai thich

If the user only says "tạo đề", "tạo đề ôn tập", "mock exam", or similar without a topic, check KB status first with
```
scripts/study_repo.py kb-status
```
.
If the KB is empty, ask exactly one onboarding question: diagnostic tổng hợp first or luyện từng kỹ năng. Do not create a personalized exam from an empty KB.
If the KB has prior evidence, create a mixed 20-question exam from
```
references/exam-blueprint.md
```
without asking for Day folders.
Use the topic taxonomy in
```
references/topic-map.md
```
.
Enforce item-writing rules in
```
references/question-quality.md
```
.
Before generating or grading, ensure a study repository exists using
```
scripts/study_repo.py init
```
.
Save exams and answer keys as Markdown files using
```
scripts/study_repo.py new-exam
```
; do not only print questions in chat.
Save or update knowledge base artifacts using
```
references/knowledge-base-workflow.md
```
.

如果用户要求创建练习考试，默认生成20道题目。
如果用户提供参数，则遵循参数要求：题目数量、主题、跟踪项、难度、时长、输出路径、答案密钥可见性。
默认使用带完整变音符号的越南语，除非用户明确要求其他语言或无变音符号文本。
切勿生成无变音符号的ASCII格式越南语考试内容。请使用
```
Đề ôn tập
```
、
```
Mã đề
```
、
```
Câu hỏi
```
、
```
Đáp án
```
、
```
Giải thích
```
，而非
```
De on tap
```
、
```
Ma de
```
、
```
Cau hoi
```
、
```
Dap an
```
、
```
Giai thich
```
。
如果用户仅输入
```
tạo đề
```
、
```
tạo đề ôn tập
```
、
```
mock exam
```
或类似表述但未指定主题，先通过
```
scripts/study_repo.py kb-status
```
检查知识库状态。
如果知识库为空，仅询问一个引导问题：
```
diagnostic tổng hợp
```
（综合诊断）优先还是
```
luyện từng kỹ năng
```
（分技能练习）。不得从空知识库生成个性化考试。
如果知识库已有历史记录，从
```
references/exam-blueprint.md
```
生成一套包含20道题的混合题型考试，无需询问Day文件夹相关内容。
使用
```
references/topic-map.md
```
中的主题分类体系。
严格执行
```
references/question-quality.md
```
中的题目编写规则。
在生成或评分前，确保通过
```
scripts/study_repo.py init
```
创建学习仓库。
使用
```
scripts/study_repo.py new-exam
```
将考试和答案密钥保存为Markdown文件；不得仅在聊天窗口中打印题目。
根据
```
references/knowledge-base-workflow.md
```
保存或更新知识库工件。

Workflow Decision Tree

工作流决策树

Initialize study repo: Run

python scripts/study_repo.py init --root <path>

→ creates

docs/user-knowledge-base.md

exams/

, answer folders, and optional git repo.

Empty KB onboarding: Run
```
python scripts/study_repo.py kb-status --root <path>
```
→ if empty, ask diagnostic vs skill drill before generating.
Generate exam: Ensure repo → read learner KB → select topics → run
```
new-exam
```
→ write exam file → write answer key → append a generation note to KB → return file paths.
Grade exam: Read exam + learner answers + answer key/rubric → score → explain misses → append grading report → update KB weaknesses/proficiency.
Review weak areas: Read KB → identify low-confidence topics → produce targeted 20-question drill or short theory review.
Refresh KB: Normalize topic names, merge duplicate weakness notes, add new topic coverage without deleting prior history.

初始化学习仓库：运行
```
python scripts/study_repo.py init --root <path>
```
→ 创建
```
docs/user-knowledge-base.md
```
、
```
exams/
```
、答案文件夹及可选的git仓库。
空知识库引导：运行
```
python scripts/study_repo.py kb-status --root <path>
```
→ 如果为空，在生成前询问用户选择综合诊断还是分技能练习。
生成考试：确保仓库存在 → 读取学习者知识库 → 选择主题 → 运行
```
new-exam
```
→ 写入考试文件 → 写入答案密钥 → 向知识库追加生成记录 → 返回文件路径。
评分考试：读取考试内容+学习者答案+答案密钥/评分标准 → 打分 → 解释错误原因 → 追加评分报告 → 更新知识库中的薄弱点/熟练度。
复习薄弱领域：读取知识库 → 识别低熟练度主题 → 生成针对性的20道题练习或简短理论复习内容。
刷新知识库：规范化主题名称、合并重复的薄弱点记录、添加新主题覆盖内容，同时保留历史记录。

Initialize Study Repository

初始化学习仓库

Run the bundled script whenever the user starts a new study workspace, asks to "init repo", or asks to create an exam but the expected structure is missing:

bash

python <skill-dir>/scripts/study_repo.py init --root <study-repo-path> --title "AI Practice Study"

Use

--no-git

only if the user does not want a git repository.

The script is idempotent: it creates missing folders/files and keeps existing KB/exam files intact.

当用户启动新学习工作区、要求"init repo"，或要求创建考试但预期结构缺失时，运行捆绑脚本：

bash

python <skill-dir>/scripts/study_repo.py init --root <study-repo-path> --title "AI Practice Study"

仅当用户明确不需要git仓库时，使用

--no-git

参数。

该脚本具有幂等性：仅创建缺失的文件夹/文件，保留现有知识库/考试文件的完整性。

Generate Exam

生成考试

Parse user parameters:
- ```
count
```
  : default 20.
- ```
scope
```
  : common, business, infrastructure, app-build, mixed, or named topics.
- ```
difficulty
```
  : default 30% easy, 50% medium, 20% hard unless a real exam format says otherwise.
- ```
formats
```
  : default MCQ + multi-select + fill-in-blank + short scenario.
Ensure the study repository exists; if not, run
```
scripts/study_repo.py init
```
.

Check KB status:

bash

python <skill-dir>/scripts/study_repo.py kb-status --root <study-repo-path>

If
```
is_empty
```
is true and the user did not explicitly request a specific topic/skill, ask: "Knowledge base hiện chưa có dữ liệu năng lực. Bạn muốn làm bài diagnostic tổng hợp trước hay luyện từng kỹ năng?" Offer only:
- Diagnostic tổng hợp 20 câu (Recommended): đo baseline ban đầu.
- Luyện từng kỹ năng: người dùng chọn RAG, Agent, Prompt Engineering, RAGAS, AI Product, Model Serving, hoặc chủ đề khác.
If the user chooses diagnostic, create a
```
diagnostic
```
/
```
mixed
```
exam using the diagnostic blueprint.
If the user chooses skill drill or already specified a topic, create a scoped drill and mark it as not-yet-personalized until graded.
If
```
is_empty
```
is false, read the learner knowledge base before selecting topics.
Create the exam/answer files before writing content:
bash
```
python <skill-dir>/scripts/study_repo.py new-exam --root <study-repo-path> --scope mixed --count 20
```
Use the script output paths for all generated content.
Prioritize weak, stale, or under-tested topics only when KB has evidence; otherwise use diagnostic/scoped coverage.
Generate questions only from available course/KB context and stable domain knowledge.
Write the student-facing exam to the exam file and the key/rubric to the answer file.
Include metadata: title, code, timestamp, scope, count, estimated time, scoring.
Update KB after generation with exam code, topics covered, intended difficulty, and pending status.
In chat, return only the created file paths and brief next step; do not duplicate the full exam unless requested.

解析用户参数：
- ```
count
```
  ：默认20。
- ```
scope
```
  ：通用、业务、基础设施、应用构建、混合或指定主题。
- ```
difficulty
```
  ：默认30%简单、50%中等、20%困难，除非真实考试格式另有规定。
- ```
formats
```
  ：默认单项选择题+多项选择题+填空题+短情景题。
确保学习仓库存在；如果不存在，运行
```
scripts/study_repo.py init
```
。

检查知识库状态：

bash

python <skill-dir>/scripts/study_repo.py kb-status --root <study-repo-path>

如果
```
is_empty
```
为true且用户未明确指定主题/技能，询问： "Knowledge base hiện chưa có dữ liệu năng lực. Bạn muốn làm bài diagnostic tổng hợp trước hay luyện từng kỹ năng?" 仅提供以下选项：
- Diagnostic tổng hợp 20 câu (推荐)：测量初始基线水平。
- Luyện từng kỹ năng：用户选择RAG、Agent、Prompt Engineering、RAGAS、AI Product、Model Serving或其他主题。
如果用户选择综合诊断，使用诊断蓝图创建
```
diagnostic
```
/
```
mixed
```
类型考试。
如果用户选择分技能练习或已指定主题，创建限定范围的练习，并标记为“未个性化”直至完成评分。
如果
```
is_empty
```
为false，在选择主题前读取学习者知识库。

在写入内容前创建考试/答案文件：

bash

python <skill-dir>/scripts/study_repo.py new-exam --root <study-repo-path> --scope mixed --count 20

将脚本输出路径用于所有生成内容。

仅当知识库有记录时，优先选择薄弱、过时或测试不足的主题；否则使用诊断/限定范围的覆盖内容。
仅从可用课程/知识库上下文和稳定领域知识生成题目。
将面向学生的考试内容写入考试文件，将密钥/评分标准写入答案文件。
包含元数据：标题、编号、时间戳、范围、题目数量、预计时长、评分规则。
生成后更新知识库，记录考试编号、覆盖主题、预期难度和待完成状态。
在聊天窗口中仅返回创建的文件路径和简要下一步操作；除非用户要求，否则不得重复完整考试内容。

Clarification Policy

澄清规则

Do not ask the user to choose among discovered Day folders when the request is broad.
Do not turn source folder names such as
```
day-22
```
,
```
day-24
```
, or
```
day-26
```
into user-facing exam options unless the user explicitly asks for a day-specific drill.
Use discovered course notes only as grounding material mapped into the blueprint sections: Common, Business, Infrastructure, App Build.
Ask a clarification only when a required output location cannot be inferred or when the user requests multiple incompatible scopes.
For broad generation, always proceed with the default mixed distribution: Common 10, Business 3, Infrastructure 3, App Build 4.

当用户请求宽泛时，不得要求用户从已发现的Day文件夹中选择。
不得将
```
day-22
```
、
```
day-24
```
或
```
day-26
```
等源文件夹名称转换为面向用户的考试选项，除非用户明确要求特定日期的练习。
仅将已发现的课程笔记作为基础材料，映射到蓝图章节：通用、业务、基础设施、应用构建。
仅当无法推断所需输出位置，或用户请求多个不兼容范围时，才进行澄清询问。
对于宽泛的生成请求，始终遵循默认混合分布：通用10题、业务3题、基础设施3题、应用构建4题。

Grade Exam

评分考试

Collect learner answers from chat or file.
Use exact matching for single-answer MCQ/fill-in-blank; use rubric for multi-select, short answer, scenario debug, case study, and code.
Award partial credit only when the rubric permits it.
Explain every incorrect or partial answer with the smallest useful correction.
Identify root-cause weaknesses by topic, not just by question number.
Update KB with score, topic performance, fixed misconceptions, new weaknesses, and next drill recommendation.
Do not overwrite prior history; append or revise the relevant matrix row.

从聊天窗口或文件收集学习者答案。
单项选择题/填空题使用精确匹配评分；多项选择题、简答题、情景调试、案例分析和代码题使用评分标准打分。
仅在评分标准允许的情况下给予部分分数。
对每个错误或部分正确的答案，提供最简洁有效的修正说明。
按主题而非仅按题目编号识别根本原因薄弱点。
更新知识库，记录分数、主题表现、纠正的误解、新的薄弱点和下一次练习建议。
不得覆盖历史记录；追加或修改相关矩阵行。

Output Contracts

输出规范

Exam file: questions only, student-facing.
Answer key: correct answers, rationale, rubric, topic tags, difficulty.
Grading report: total score, section scores, missed concepts, next actions.
KB update: factual record of generated/graded exams and proficiency deltas.

Use

references/output-formats.md

for concrete Markdown templates.

考试文件：仅包含题目，面向学生。
答案密钥：正确答案、理由、评分标准、主题标签、难度。
评分报告：总分、章节分数、遗漏的概念、下一步行动。
知识库更新：生成/评分考试的事实记录和熟练度变化。

使用

references/output-formats.md

中的具体Markdown模板。

Security And Privacy

安全与隐私

Never include private paths, hidden source locations, credentials, phone numbers, or personal data in generated exams or reports unless the user explicitly asks to edit a personal document that already contains them.
Do not reveal internal source locations used to build context.
Treat learner performance records as personal study data. Keep updates scoped to the KB files the user is working with.
Refuse requests to generate answer leaks for an active real exam; offer practice questions instead.
Do not invent official policy, grading, or legal claims. Mark uncertain legal/regulatory details as needing verification.

除非用户明确要求编辑已包含此类信息的个人文档，否则不得在生成的考试或报告中包含私有路径、隐藏源位置、凭据、电话号码或个人数据。
不得透露用于构建上下文的内部源位置。
将学习者表现记录视为个人学习数据。仅在用户正在使用的知识库文件中进行更新。
拒绝为当前正式考试生成答案泄露的请求；改为提供练习题目。
不得编造官方政策、评分规则或法律声明。将不确定的法律/监管细节标记为需要验证。

References

参考资料

```
references/exam-blueprint.md
```
: exam structure, scoring, default distributions.
```
references/topic-map.md
```
: AI topic taxonomy by common section and three tracks.
```
references/question-quality.md
```
: MCQ, multi-select, fill-in-blank, short-answer rules.
```
references/knowledge-base-workflow.md
```
: generation/grading KB update rules.
```
references/output-formats.md
```
: Markdown templates.
```
references/research-notes.md
```
: condensed research rationale and citations.
```
scripts/study_repo.py
```
: initialize study repo and create exam/answer Markdown files.

```
references/exam-blueprint.md
```
：考试结构、评分规则、默认分布。
```
references/topic-map.md
```
：按通用章节和三个跟踪项划分的AI主题分类体系。
```
references/question-quality.md
```
：单项选择题、多项选择题、填空题、简答题的编写规则。
```
references/knowledge-base-workflow.md
```
：生成/评分的知识库更新规则。
```
references/output-formats.md
```
：Markdown模板。
```
references/research-notes.md
```
：浓缩的研究依据和引用。
```
scripts/study_repo.py
```
：初始化学习仓库并创建考试/答案Markdown文件。