A personal interview training officer for tech roles — runs as a Claude Code plugin. One
/interviewcommand orchestrates mock interviews, post-interview debriefs, progressive skill drilling, a self-updating capability profile, pipeline tracking, and resume tailoring.一套技术岗 面试训练官（Claude Code 插件）。一个
/interview命令打通模拟面试、面试复盘、 渐进式技能闯关、自动更新的能力画像、面试管线追踪和简历定制。
Interview Trainer is a Claude Code plugin distilled from a real interview-prep system that was battle-tested across a full job-hunting season (18+ real interviews). It is not a static question bank. It is a closed-loop coaching system:
- Mock interview engine — Claude role-plays a hiring-bar interviewer, asks one question at a time, adaptively drills deeper, and never leaks the answer mid-interview. Scores you afterward.
- Debrief — after a real interview, Claude walks you through recall, diagnoses each answer against a target level, and writes a structured record.
- Progressive drilling (L1→L5) — a graded ladder per skill (analogy → principle → project → resisting follow-ups → cross-domain transfer) with a cross-session mastery profile. Low scores flow back as targeted remediation next time.
- Capability profile — a living matrix of topics × levels × weak-spot tracking × readiness, updated automatically by both mock and real interviews.
- Pipeline + analytics — track every company through the funnel; get cross-interview analysis.
- Resume tailoring — JD-driven match scoring and ATS optimization with a strict no-fabrication rule.
- Calibration engine — tracks two drifts: your self-assessment vs coach scores, and mock scores vs real interview outcomes. When they diverge, scoring gets recalibrated. Mastery also goes stale: topics passed more than 30 days ago are flagged for retest.
- STAR storybank management — per-story strength ratings and automatic usage write-back after real interviews; prep allocates stories to predicted questions with fit scoring, so one story never has to carry two answers in the same interview.
Everything is connected by feedback loops: a weak spot exposed in a real debrief automatically becomes a remediation level in your next drill; mock results update your capability profile; the smart navigator reads all of it to tell you what to do next.
The plugin ships the engine. You own your profile and knowledge base. Your interview state is auto-maintained.
ENGINE (this repo, shared) skills/ — 9 skills behind one /interview router
│
▼ reads / writes
WORKSPACE (yours, private)
├── profile.md ← who you are, your target, your locked metrics, resume path
├── knowledge-base/ ← your topics, analogies, STAR stories, company styles, methodologies
└── data/ ← auto-maintained: pipeline, capability profile, records, mock logs…
Your workspace is just a directory you run Claude Code from. /interview setup creates it for you.
/plugin marketplace add Dora0512/interview-trainer
/plugin install interview-trainer
# 1. Create a directory to be your interview workspace, and run Claude Code from inside it.
mkdir my-interview-prep && cd my-interview-prep
# 2. In Claude Code, bootstrap your profile + knowledge base (interactive):
/interview setup
# 3. From then on, just ask the navigator what to do next:
/interview
| Command | What it does | Writes files? |
|---|---|---|
/interview |
Smart navigation — analyzes your state, recommends next action | No |
/interview setup |
One-time onboarding: builds profile + knowledge-base + data skeleton | Yes |
/interview status |
Pipeline dashboard + capability profile + weak-spot tracker | No |
/interview analytics |
Cross-interview analysis report | Yes |
/interview apply <company> <JD> |
Tailor resume + add to pipeline | Yes |
/interview prep <company> |
Targeted prep plan for an upcoming round | Yes |
/interview mock [topic] [--company X] |
Interactive mock interview, then scoring | Yes |
/interview debrief |
Post-interview debrief (interactive) | Yes |
/interview review <topic> |
One-shot review card for a skill | Usually no |
/interview review-deep <topic> |
Progressive L1-L5 drilling, cross-session | Yes |
/interview coach <question> |
Generate a high-quality answer | No |
Each sub-skill is also callable directly: /mock-interview, /interview-debrief, /deep-review,
/review-skill, /interview-coach, /interview-prep, /resume-tailor.
Interview Trainer is role-agnostic within tech. Backend, frontend, mobile, data, infra, PM-ish
hybrid roles — you define your own topic taxonomy and company styles in knowledge-base/. See
docs/customization.md.
- Architecture — the three layers and how the feedback loops fit together.
- Customization — define your own topics, company styles, methodologies.
- Example workspaces — fictional filled-in workspaces for Android, Backend, Frontend, and Data Engineering roles. They show the engine is role-agnostic: only the config layer changes. Copy the shape, not the content.
Your profile.md, resume, STAR stories, and real interview records are personal. Keep your
workspace in a private directory, not committed to this public repo (see .gitignore). The
example workspace uses an entirely fictional candidate and company.
Interview Trainer 是一个 Claude Code 插件，源自一套经过完整求职季（18+ 场真实面试）打磨的面试备战系统。 它不是静态题库，而是一个闭环教练系统：
- 模拟面试引擎 —— Claude 扮演卡线面试官，一次只问一个问题，自适应深挖，面试中绝不透露答案，结束后评分。
- 面试复盘 —— 真实面试后，Claude 引导你回忆、按目标层级诊断每道题、生成结构化记录。
- 渐进式闯关（L1→L5） —— 每个技能一条评分阶梯（类比 → 原理 → 项目落地 → 抗追问 → 跨场景迁移）， 跨会话维护掌握度档案，低分自动倒灌成下次的针对性补强关。
- 能力画像 —— 话题 × 层级 × 薄弱点追踪 × 准备度 的活矩阵，模拟和真实面试都会自动更新它。
- 管线 + 分析 —— 追踪每家公司在漏斗中的位置，生成跨面试分析。
- 简历定制 —— 按 JD 做匹配评分和 ATS 优化，严守不编造红线。
- 校准引擎 —— 追踪两条漂移：「自评 vs 教练评分」和「模拟分 vs 真实结果」，发散时自动校准评分口径。 掌握度也会过期：达标超过 30 天未复测的话题自动标记复测。
- STAR 故事组合管理 —— 每个故事有强度评级，真实面试后自动回写使用记录；准备阶段按预测问题做 故事适配评分，同一场面试不让一个故事硬撑两道题。
所有模块由反馈回路串联：真实复盘暴露的薄弱点，自动变成下次闯关的补强关；模拟结果更新能力画像； 智能导航读取这一切，告诉你下一步该做什么。
插件发布的是引擎。你拥有自己的画像和知识库。你的面试状态自动维护。
引擎层（本仓库，共享） skills/ —— 一个 /interview 路由器 + 9 个 skill
│
▼ 读 / 写
工作区（你的，私有）
├── profile.md ← 你是谁、目标、锁定的量化数据、简历路径
├── knowledge-base/ ← 你的话题体系、类比、STAR 故事、公司风格、方法论
└── data/ ← 自动维护：管线、能力画像、面试记录、模拟记录……
工作区就是你启动 Claude Code 的那个目录。/interview setup 会帮你建好。
/plugin marketplace add Dora0512/interview-trainer
/plugin install interview-trainer
# 1. 建一个目录作为你的面试工作区，在里面启动 Claude Code
mkdir my-interview-prep && cd my-interview-prep
# 2. 在 Claude Code 里交互式初始化画像 + 知识库
/interview setup
# 3. 之后只管问导航该做什么
/interview
| 命令 | 作用 | 写文件？ |
|---|---|---|
/interview |
智能导航 —— 分析状态，推荐下一步 | 否 |
/interview setup |
一次性初始化：建画像 + 知识库 + 数据骨架 | 是 |
/interview status |
管线仪表盘 + 能力画像 + 薄弱点追踪 | 否 |
/interview analytics |
跨面试分析报告 | 是 |
/interview apply <公司> <JD> |
定制简历 + 加入管线 | 是 |
/interview prep <公司> |
针对下一轮的准备计划 | 是 |
/interview mock [话题] [--company X] |
交互式模拟面试 + 评分 | 是 |
/interview debrief |
面试后交互式复盘 | 是 |
/interview review <话题> |
技能点一站式复习卡片 | 通常否 |
/interview review-deep <话题> |
渐进式 L1-L5 闯关，跨会话 | 是 |
/interview coach <问题> |
生成高质量面试回答 | 否 |
每个子 skill 也可独立调用：/mock-interview、/interview-debrief、/deep-review、/review-skill、
/interview-coach、/interview-prep、/resume-tailor。
Interview Trainer 在技术岗内领域无关。后端、前端、移动、数据、基础架构、偏产品的混合岗都行——
你在 knowledge-base/ 里定义自己的话题体系和公司风格。见 docs/customization.md。
- 架构 —— 三层结构和反馈回路如何拼合。
- 定制 —— 定义你自己的话题、公司风格、方法论。
- 示例工作区 —— Android / 后端 / 前端 / 数据工程 四个岗位的虚构工作区， 证明引擎领域无关：只有配置层不同。学结构，别照搬内容。
你的 profile.md、简历、STAR 故事、真实面试记录都是个人信息。把工作区放在私有目录里，
不要提交到这个公开仓库（见 .gitignore）。示例工作区使用完全虚构的候选人和公司。
Built from a real, battle-tested interview system. Engine is shared; your data stays yours. 源自一套真实打磨的面试系统。引擎共享，数据归你。