Training Quiz
Overview
This skill turns the product knowledge base into an interactive learning system for staff.
It adapts difficulty, tracks progress, and celebrates improvement — making training
feel less like a chore and more like a game.
Depends on: products[] + policy_entries[] + faqs[] in knowledge base.
Quiz Modes
| Mode | Trigger | Format | Best For |
|---|
| Flashcard | "考我产品知识" | Q → reveal A | Quick daily review |
| Multiple Choice |
"选择题模式" | Q + 4 options | Structured testing |
| Scenario | "情景练习" | Role-play customer scenario | Sales skill practice |
| Policy Drill | "考我政策" | Policy rule questions | Compliance training |
| New Product | "考我新品" | Focus on recently added items | New arrival onboarding |
Default mode: Flashcard (lowest friction).
Session Flow
Start
- 1. Greet the learner by name (if known from staff config)
- Ask or confirm: mode, topic focus, number of questions (default 10)
- Begin immediately — don't over-explain
During Quiz
- - Ask one question at a time
- Wait for answer before revealing correct response
- On correct: brief positive reinforcement ("✅ 答对了！") + optional fun fact
- On incorrect: show correct answer + brief explanation (2 sentences max)
- Track: correct / total, running accuracy %
End of Session
CODEBLOCK0
Reference: question-bank.md — question templates by type.
Question Generation
Questions are auto-generated from the knowledge base. No manual authoring needed.
From products:
- - "这款[产品名]的[属性]是什么？" → answer from
description/ INLINECODE4 - "下面哪个是[产品名]的正确价格？" → MCQ using real + nearby prices as distractors
- "[顾客描述] → 你会推荐哪款产品？" → scenario from INLINECODE5
From policies:
- - "退货政策中，[条件]，顾客可以享受什么？"
- "以下哪种情况不在退货政策范围内？" → MCQ with real exceptions as options
From FAQs:
- - Use
question field directly - Shuffle real FAQ answers as MCQ distractors
Script: scripts/generate_questions.py — generates a quiz set from the KB.
Progress Tracking
Store per-employee progress in agent memory under training_progress.<staff_id>:
CODEBLOCK1
Report progress to manager on request:
"张三本月完成 5 次练习，平均正确率 82%，政策类题目需加强。"
Gamification
Keep it motivating:
| Achievement | Trigger |
|---|
| 🌟 首次完成 | First quiz session |
| 🔥 连续挑战 |
3+ consecutive days |
| 📚 政策达人 | 5 policy quizzes with ≥90% |
| 🏆 产品专家 | Overall accuracy ≥90% over 10+ sessions |
| ⚡ 闪电手 | 10 consecutive correct answers |
Announce badges immediately when earned. Keep it brief and genuine.
培训测验
概述
该技能将产品知识库转化为面向员工的互动学习系统。它能自适应难度、追踪进度并庆祝进步——让培训不再像苦差事，更像一场游戏。
依赖项： 知识库中的 products[] + policy_entries[] + faqs[]。
测验模式
| 模式 | 触发词 | 形式 | 最佳用途 |
|---|
| 闪卡 | 考我产品知识 | 提问 → 揭示答案 | 快速日常复习 |
| 选择题 |
选择题模式 | 提问 + 4个选项 | 结构化测试 |
| 情景 | 情景练习 | 角色扮演客户场景 | 销售技巧练习 |
| 政策 | 考我政策 | 政策规则问答 | 合规培训 |
| 新品 | 考我新品 | 聚焦最新添加内容 | 新品入职培训 |
默认模式：闪卡（最低门槛）。
会话流程
开始
- 1. 以姓名问候学员（如从员工配置中获取）
- 询问或确认：模式、主题重点、题目数量（默认10题）
- 立即开始——不要过度解释
测验中
- - 一次只问一个问题
- 等待回答后再揭示正确答案
- 答对时：简短正面强化（✅ 答对了！）+ 可选趣味知识
- 答错时：显示正确答案 + 简短解释（最多两句话）
- 追踪：正确数/总数，实时正确率%
会话结束
🎓 本次练习结束！
结果：[正确数]/[总数] — [得分]%
[评级: 优秀 ≥90% | 良好 70-89% | 需加强 <70%]
[如果得分 < 70%]: 建议重点复习：[列出薄弱类别]
[如果得分 ≥ 90%]: 太棒了！你已经达到优秀水平 🏆
下次想练习什么？
参考： question-bank.md — 各类题型模板。
题目生成
题目从知识库自动生成，无需手动编写。
来自产品：
- - 这款[产品名]的[属性]是什么？ → 答案来自 description/variants
- 下面哪个是[产品名]的正确价格？ → 选择题，使用真实价格及相近价格作为干扰项
- [顾客描述] → 你会推荐哪款产品？ → 场景题，来自 suitable_for
来自政策：
- - 退货政策中，[条件]，顾客可以享受什么？
- 以下哪种情况不在退货政策范围内？ → 选择题，使用真实例外作为选项
来自常见问题：
- - 直接使用 question 字段
- 打乱真实常见问题答案作为选择题干扰项
脚本： scripts/generate_questions.py — 从知识库生成测验集。
进度追踪
在智能体记忆中以 training_progress.<员工ID> 存储每位员工的进度：
json
{
staffid: zhangsan,
sessions: [
{
date: 2024-07-15,
mode: flashcard,
score: 8,
total: 10,
accuracy: 80,
weak_categories: [policy, pricing]
}
],
cumulative_accuracy: 82,
badges: [firstquiz, 7daystreak, policy_master]
}
应要求向经理汇报进度：
张三本月完成 5 次练习，平均正确率 82%，政策类题目需加强。
游戏化
保持激励性：
| 成就 | 触发条件 |
|---|
| 🌟 首次完成 | 第一次测验会话 |
| 🔥 连续挑战 |
连续3天以上 |
| 📚 政策达人 | 完成5次政策测验且正确率≥90% |
| 🏆 产品专家 | 10次以上会话总体正确率≥90% |
| ⚡ 闪电手 | 连续答对10题 |
获得成就时立即宣布。保持简短和真诚。
