基于多阶段协作的智能面试系统,模拟真实面试场景,对用户进行技术面试评估。

v1.0.0

AI 模拟面试助手 — 基于多阶段协作的智能面试系统。用户说'我要开始模拟面试'即可启动,自动进行简历画像管理、岗位匹配分析、5道题面试循环、实时评分和总结报告。70%关键词匹配+30%思路评分,60分合格。支持匿名模式。

0· 135·0 current·0 all-time

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for paradise123-bot/ai-interview-assistant.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "基于多阶段协作的智能面试系统,模拟真实面试场景,对用户进行技术面试评估。" (paradise123-bot/ai-interview-assistant) from ClawHub.
Skill page: https://clawhub.ai/paradise123-bot/ai-interview-assistant
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install ai-interview-assistant

ClawHub CLI

Package manager switcher

npx clawhub@latest install ai-interview-assistant
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Benign
high confidence
Purpose & Capability
Name/description describe an interview assistant and the code + SKILL.md implement that: resume management, JD matching, question retrieval, scoring, and report generation. The scripts fetch interview material from the declared GitHub repo (wdndev/llm_interview_note) and call platform tools (tool_scorer, tool_resume_db, web_fetch) consistent with the stated purpose. There are no declared env vars, binaries, or config paths that are unrelated to the task.
Instruction Scope
SKILL.md instructs the agent to use L3 tools (tool_kb_search/tool_web_fetch/tool_scorer/etc.) and follow a bounded, signal-driven interview flow. The included Python scripts perform network fetches (curl to raw.githubusercontent.com via subprocess) and spawn the helper script (python3 fetch.py). Those network/subprocess actions are expected for fetching question content, but they are the main external behaviors to review.
Install Mechanism
No install spec is provided (instruction-only plus included scripts). No remote archive downloads or obscure URLs in an installer. The skill ships two Python scripts that are executed via subprocess; this is low-to-moderate risk but consistent with the skill's function.
Credentials
The skill declares no required environment variables or credentials and the SKILL.md doesn't ask for secrets. The code does not read environment variables or request unrelated credentials. Note: it depends on external helper tools (tool_resume_db, web_fetch, tool_scorer) that may themselves require credentials or storage—those are not declared here and should be trusted/reviewed separately.
Persistence & Privilege
always:false and no special platform privileges. The SKILL.md states 'local storage' for user profiles and the flow updates the resume/profile via tool_resume_db.update_feedback — persistence is expected for user profiles, but the actual storage location and retention are controlled by the underlying tool_resume_db implementation (not included). Verify where and how profile data is stored before enabling.
Assessment
This skill appears to do what it says, but review a few things before installing: 1) The scripts fetch content from a public GitHub repo (wdndev/llm_interview_note) via curl/subprocess — ensure you trust that repo because its contents drive question generation. 2) The skill calls external platform tools (tool_resume_db, tool_scorer, web_fetch, jd_matcher). Those tools may store or transmit data (or require credentials) even though this skill doesn't declare them — inspect/confirm their behavior and storage location. 3) Confirm the 'local storage' claim by checking the implementation of tool_resume_db (where profiles are saved, retention, deletion API) if you care about privacy. 4) Because the scripts run subprocesses and perform network fetches, consider running in a sandboxed environment or reviewing the fetch.py output to ensure no unexpected network endpoints are contacted. If you need higher assurance, ask the author for the resume DB implementation or run the skill offline with curated question sets.

Like a lobster shell, security has layers — review code before you run it.

latestvk977fcgzp8g8axy1edja274am18423ph
135downloads
0stars
1versions
Updated 3w ago
v1.0.0
MIT-0

AI 模拟面试助手

基于多阶段协作的智能面试系统,模拟真实面试场景,对用户进行技术面试评估。

启动方式

用户只需说:"我要开始模拟面试""开始面试练习"

系统架构

┌─────────────────────────────────────────────────┐
│  L1 总调度官                                      │
│  协调各阶段执行,管理面试状态流转                  │
└─────────────────────────────────────────────────┘
                          ↓
    ┌─────────────────────────────────────────┐
    │  L2 阶段执行导演                          │
    │  简历管理 → JD匹配 → 面试循环 → 总结报告  │
    └─────────────────────────────────────────┘
                          ↓
    ┌─────────────────────────────────────────┐
    │  L3 原子工具                             │
    │  出题/评分/匹配/画像/报告                 │
    └─────────────────────────────────────────┘

业务流程 (SOP)

阶段 1:简历/画像管理

职责:建立用户技术画像

流程

  1. 询问用户姓名(或使用匿名模式)
  2. 调用 tool_resume_db.query_name 查询是否存在画像
  3. 老用户:确认身份,复述其技术栈
  4. 新用户:引导输入技术栈标签
  5. 输出信号:SUCCESS: 画像已就绪

引导话术

"欢迎使用 AI 模拟面试助手! 请告诉我你的姓名(或输入'匿名'使用匿名模式), 我将为你建立技术画像,后续面试将针对性出题。"


阶段 2:岗位匹配(可选)

职责:分析用户与目标岗位的匹配度

触发条件:用户选择"是"并提供 JD

流程

  1. 获取用户画像
  2. 获取 JD 文本
  3. 调用 tool_jd_matcher.match_jd()
  4. 展示匹配分、核心竞争力、待提升项
  5. 输出信号:SUCCESS: 匹配分析完成

引导话术

"您的画像已就绪。是否需要先进行【岗位匹配分析】? 输入'是'提供 JD,输入'否'直接开始面试。"


阶段 3:面试实战循环

职责:执行 5 道题的核心面试

流程

For i in [1, 5]:
    ├─ Step A: 出题
    │   ├─ 基础题 → tool_kb_search(tags, difficulty, count=1)
    │   └─ 前沿题 → tool_web_fetch_question(tags)
    │
    ├─ Step B: 提问
    │
    ├─ Step C: 评分
    │   └─ 每轮回答 → tool_scorer(key_points, answer)
    │
    ├─ Step D: 追问 (最多3轮)
    │   ├─ 得分<60 → 追问或给提示
    │   └─ 得分≥60 或 3轮满 → 结算本题
    │
    └─ Step E: 进入下一题

面试话术模板

{'='*50}
🎯 第 {current}/5 题
{'='*50}

【题目类型】{type}
【题目】{question}

请作答:

评分反馈

📊 本轮评分:
关键词得分: {keyword_score}/70
思路得分: {logic_score}/30
总分: {total_score}/100

约束

  • 同一问题最多追问 3 次
  • 每题必须有明确得分记录
  • 用户说"跳过"可跳过当前题(记 0 分)

阶段 4:总结归档

职责:生成报告,持久化反馈

流程

  1. 汇总 5 题得分数据
  2. 调用 tool_summary_generator 生成报告
  3. 调用 tool_resume_db.update_feedback 更新画像
  4. 输出信号:TERMINATE: 任务圆满完成

报告格式

{'='*50}
📋 模拟面试报告
{'='*50}
总分: {total_score}/500
平均分: {avg_score}/100
评级: {grade}

💪 优势领域: {strong_points}
📚 待加强领域: {weak_points}

评分标准

tool_scorer 评分维度

维度权重说明
关键词匹配70%回答是否覆盖考察要点中的核心关键词
整体思路30%逻辑清晰度、深度、表述准确性

合格线:总分 ≥ 60 分

评级标准

评级分数范围说明
S90-100优秀
A80-89良好
B70-79中等
C60-69及格
D<60需努力

状态流转图

                    ┌──────────────────┐
                    │  START (启动)    │
                    └────────┬─────────┘
                             ↓
              ┌──────────────────────────┐
              │  阶段1: 简历/画像管理      │
              └──────────────┬───────────────┘
                             ↓
              ┌──────────────────────────┐
              │  是否需要 JD 匹配?       │
              │  (用户选择)               │
              └──────────────┬───────────────┘
                    ┌────────┴────────┐
                   是                 否
                    ↓                  ↓
     ┌──────────────────┐   ┌──────────────────┐
     │ 阶段2: JD匹配    │   │ 跳过             │
     └────────┬─────────┘   └────────┬─────────┘
              ↓                       ↓
     ┌──────────────────┐   ┌──────────────────┐
     │ 阶段3: 面试循环   │←─┤ (直接进入)       │
     └────────┬─────────┘   └──────────────────┘
              ↓
     ┌──────────────────┐
     │ 阶段4: 总结归档  │
     └────────┬─────────┘
              ↓
     ┌──────────────────┐
     │  TERMINATE       │
     └──────────────────┘

依赖工具清单

L3 原子工具

工具名功能输入输出
tool_kb_searchGitHub知识库检索出题(wdndev/llm_interview_note)tags, difficulty, count题目列表(含content_preview、url)
tool_web_fetch_question获取前沿技术题tags, topic最新面试题
tool_scorer实时打分key_points, answer评分结果
tool_jd_matcherJD 匹配分析resume, jd匹配报告
tool_resume_db用户画像管理-CRUD 操作
tool_summary_generator生成面试报告scores, profile结构化报告

L2 阶段剧本

剧本名职责输出信号
phase_resume_manager简历/画像SUCCESS: 画像已就绪
phase_match_managerJD 匹配SUCCESS: 匹配分析完成
phase_interview_engine面试循环SUCCESS: 5轮面试结束
phase_report_manager总结归档TERMINATE: 任务圆满完成

技术方向支持

支持的面试方向

方向包含技术GitHub分类
LLM基础Word2Vec、CNN/RNN/TF、Decoder-only01.大语言模型基础
LLM架构Attention、位置编码、Transformer、BERT02.大语言模型架构
分布式训练数据并行、流水线并行、张量并行、DeepSpeed04.分布式训练
LLM微调LoRA、Adapter、Prompting05.有监督微调
推理优化vLLM、TensorRT-LLM、量化06.推理
强化学习PPO、RLHF、DPO07.强化学习
RAG/Agent检索增强、Agent技术08.检索增强RAG
模型评估评测、幻觉问题09.大语言模型评估
前沿应用CoT、LangChain10.大语言模型应用

题目类型

类型占比说明
八股题50%基础概念、原理性问题
实战题30%项目经验、应用场景
前沿题20%最新技术、趋势问题

隐私保护策略

  • 匿名模式:用户可选择不提供姓名
  • 最小采集:仅收集技术标签和面试表现数据
  • 本地存储:画像数据存储在本地
  • 可清除:用户可随时请求删除画像

约束规则

  1. 严格评分:不因用户态度、表现友好而加分
  2. 状态锁定:当前阶段未完成前,不开启下一阶段
  3. 信号驱动:通过握手信号控制流程流转
  4. 上下文传递:确保各阶段间数据连贯

异常处理

情况处理方式
用户拒绝提供信息使用默认配置,允许匿名面试
工具调用失败降级为随机题目生成
用户说"跳过"当前题记 0 分,进入下一题
用户说"结束"提前结束面试,生成当前报告

Comments

Loading comments...