interview-debrief
Skill1 starsChanged 36 days agoMITView on GitHub
Turn an interview transcript into a structured debrief. Two modes - self mode (own interview, scored, with career-track judgement) and consultant mode (paid diagnosis for someone else, delivered as a Word report). Triggers on "面试复盘", "逐字稿", "面试录音", "帮我复盘", "这场面试怎么样", "帮我看看他面试有什么问题". Not for live interview coaching or writing answers before an interview happens.
5 other projects use a version of this.
Adopted by
Most recently taken up by
- ZoolCoder/scoutjobstheir own version, first seen Nov 1, 2026
- osamabinjannat/remote-hiring-skillstheir own version, first seen Oct 25, 2026
- ficilcom/otame4-work-skillstheir own version, first seen Aug 28, 2026
- palashjain95/jobhuntertheir own version, first seen Mar 28, 2026
- jackhendon/pm-skillstheir own version, first seen Mar 1, 2026
Used with
Projects with it also use interview-prep.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
AnthropicSkillinterview-prep
Create structured interview plans with competency-based questions and scorecards. Trigger with "interview plan for", "interview questions for", "how should we interview", "scorecard for", or when the user is preparing to interview candidates.
Also in 3 projects that use this
Other versions5 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 5
- ZoolCoder/scoutjobs★ 4same name
- ficilcom/otame4-work-skills★ 3same name
- palashjain95/jobhunter★ 2same name
- osamabinjannat/remote-hiring-skills★ 1same name
- jackhendon/pm-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository30 other files it gives its agents.
- anna-app-publishSkill
- campus-jobs-radarSkill
- career-pivot-researchSkill
- cf-pages-deploySkill
- company-risk-checkSkill
- github-repo-launch-polishSkill
- github-repo-repair-auditSkill
- headless-webapp-screenshotsSkill
- historical-remotion-filmSkill
- job-application-screenerSkill
- job-application-trackerSkill
- llm-output-eval-gateSkill
- memory-sync-abcSkill
- offer-decision-frameworkSkill
- online-assessment-solverSkill
- personal-secretary-memorySkill
- personal-tone-writingSkill
- portfolio-github-workflowSkill
- recruiting-source-huntingSkill
- remotion-delivery-qaSkill
- resume-jd-matcherSkill
- resume-jd-tailorSkill
- resume-polish-craftSkill
- session-checkpointSkill
- session-log-archaeologySkill
- single-file-html-toolSkill
- static-site-deploy-verifySkill
- vibecoding-resume-storySkill
- video-portfolio-performanceSkill
- workflow-distillerSkill
The file
We split this skill into the rules it gives an agent
1 rule. 0 are common, 1 is distinctive, and 0 are very specific to this project.
0
Common
Most of the 4 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 10
Not to judge — to find what worked, what didn't, and exactly how to fix it.
“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Interview Debrief (面试复盘)
Turn a raw interview transcript into a structured debrief. Not to judge — to find what worked, what didn't, and exactly how to fix it.
Step 0: Decide the mode first
This decides everything downstream. Ask if it is not obvious.
| 自我模式 (self) | 顾问模式 (consultant) | |
| 对象 | 用户本人 | 别人(付费/帮忙) |
| 语气库 | 用他自己的说话方式改写 | ❌ 不用,他不念这些稿 |
| 赛道判断 | 给(他要不要换方向) | ❌ 不给,那是别人的赛道 |
| 评分 | 给(能力/表达/契合/流程) | ❌ 不给,客户要的是问题清单不是分数 |
| 交付物 | Markdown | Word(要发给客户) |
| 措辞硬度 | 可以直接 | 一样要硬,但去掉人身评价 |
顾问模式的唯一目标:让客户看完就知道「我下一场面试具体改哪三句话」。不是安慰,不是打分。
Getting the transcript
| 来源 | 正确做法 |
| 飞书妙记/智能纪要 | ⚠️ WebFetch 会在第一个大表格处截断,只能拿到前半段。必须 lark-cli note +transcript --note-id <id> --output x.md --overwrite |
| 飞书会议 | ⚠️ lark-cli meeting 命令不存在,域名是 note |
| 音频文件 | AudioTranscribe 工具 |
| 截图 | ⚠️ Houston 准则:看不到就第一句声明,不许猜内容 |
Consultant mode: diagnosis
Layered severity (固定四层,不要自创)
| 层 | 定义 | 判定标准 |
| 致命 | 单独出现就足以让这场失败 | 面试官已经明确给出方向而候选人没切进去 |
| 严重 | 不判死,但持续拉低专业度 | 准备了话术,没准备事实 |
| 表达 | 不影响判断对错,影响像不像同行 | 自造词、叫法不统一、填充词 |
| 收尾 | 最后两分钟丢的分 | 没表态、没感谢、反问全是索取型 |
致命层的三类高频信号(比"立场冲突"更常见)
- 答的不是问题 —— 面试官连问同一个问题 ≥2 次。这是最严重的信号:它同时说明候选人拿不出东西并且没察觉自己没答到点上。 例:连问三次「数据情况怎么样」,三次回答都落在创意逻辑上,全程零个数字。
- 把"讲方法"当成"讲结果" —— 问「结果怎么样」,答「我怎么优化」。这是认知问题不是表达问题,换任何一场都会栽在同一处。
- 简历上写明的东西被面试官反问 —— 「现在是大三吧?」「你们公司是做电商的吧?」。说明连自我介绍和公司功课都没做。
一条不合格的诊断长什么样
❌ 「表达能力有待提升」 —— 客户没法执行。 ✅ 「06:47/08:04/08:58 连问三次数据,三次都没给数字;改法:每个案例准备三个数字,面试前读三遍。」
每条必须四件套
1) 现场原话(带时间戳) —— 证据,不许概括
2) 为什么这条扣分 —— 机制,不是"这样不好"
3) 改法:可直接念的一句话 —— 客户复制就能用
4) 归到四层中的哪一层 —— 决定改的优先级改法必须写成能直接念出来的完整句子,不能写「注意要有逻辑」这类原则。
收尾四条(几乎每场都丢,成本最低)
- 没有明确说「我可以接受」薪资 → 不表态等于没谈定
- 收尾口吻像挂电话(「那就先这样」)
- 全程没有说过一次谢谢
- 提的问题全是索取型(创业公司最怕「来学东西的」)→ 至少准备一个价值型问题
必带的三张表(顾问模式)
- 总体判断 —— 一句话核心矛盾 + 三个关键事实(都带时间戳)
- 优先修正顺序 —— 只列 3 条,每条写「为什么它排第一」
- 面试前信息清单 —— 姓名/上家/项目数据/目标公司/薪资表态/反问,逐项标「本次是否做到」
Consultant mode: delivering the Word file
客户要的是能转发的文件,不是 Markdown。
S1 doc-writer → stage1/final_draft.md
S2 doc-formatter → stage2/formatted.html 门禁 html-review ≥ 95
S3 doc-converter → 成品 .docx convert_doc.py --require-page-furniture- 文档必须有封面和页码 → S3 禁止
--no-preflight - ⚠️ Word COM 可能损坏(
TYPE_E_CANTLOADLIBRARY,Office/WPS 混装时)。要 PDF 就让用户手动另存,别耗在自动化上;本机没 LibreOffice 时也别去找。 - 转换完必须读回验证(python-docx 按 body 顺序 dump 段落+表格),确认章节齐全、页码域在正文节而不在封面节。脚本返回 success 不等于内容对。
脱敏红线(公开场合 / 可转发的产物)
客户姓名、学校、公司名、产品名、面试官姓名 —— 一律泛化。
❌「XX 师范学院计算机专业,面的是 XX 情感社交 APP」(真人真公司) ✅「某高校计算机专业,面的是一家情感社交产品」 逐字稿原文是客户私有的,报告里只留必要原话,不留身份信息。
Self mode (保留 v1 流程)
问题分级
| 级别 | 检测信号 |
| 🔴 红灯 | 立场冲突 / 硬条件暴露且无补救 / 时间线矛盾 / 自曝短板无转折 |
| 🟡 黄灯 | 口语碎片 > 15% / 问 A 答 B / 反问环节浪费 / 过度谦虚或过度自信 |
| 🟢 绿灯 | 可精炼、可换例子(文字逐字稿看不到的不判) |
| ⚪ 观察项 | 面试官反应、是否有笔试、面试形式 —— 逐字稿里没有就标"未知" |
评分
能力匹配度 X/10 · 表达清晰度 X/10 · 岗位契合度 X/10 · 时间/流程 X/10
核心矛盾:一句话总结为什么是这个评分输出结构
一、基本面(时长/形式/面试官数 + 总评 + 核心矛盾)
二、亮点(类型 + 原文引用 + 为什么好)
三、问题(级别 + 原文引用 + 为什么 + 怎么改)
四、改写稿(符合他本人语气,可直接念)
五、下轮行动清单(3个必讲故事 / 3个反问 / 避坑表)
六、赛道判断(这个方向对不对 / 同公司其他选择)Red lines
- 🔴 Houston 准则:读不到的内容,第一句必须是「我看不到这个,请转成文字」。在此之前不许输出任何关于内容的断言。
- 🔴 不编造面试官反应:逐字稿里没有的非语言信息一律标"未知"。
- 🔴 不给虚假希望:确实差就直说。安慰的话只放在长期视角里(自我模式),顾问模式里不放。
- 🔴 每条论断必须能追到原话位置,没有时间戳的断言不许写进报告。
- 🔴 顾问模式不做赛道判断,那是别人的职业选择,不是这份报告该管的。
- 🔴 自我模式不用顾问式的打分清单:给自己的复盘要含"接下来怎么办",纯问题清单对他没用。
Example
自我模式:
用户发来飞书妙记导出的逐字稿(12分37秒)→ 六步判断框架(亮点) + 立场冲突(红灯) + 时间线矛盾(黄灯) → 结论 🟡黄灯偏红灯,能力过了卡在时间 → 赛道判断:内容风控比泛运营更适配
顾问模式:
客户发来 27 分钟面试录音,30 元求职梳理单 → 取逐字稿 → 诊断出 20 条,核心是「面试官连问三次数据、三次没给一个数字」和「把讲数据理解成讲投放策略」→ 七章 Word(致命3 / 严重8 / 表达5 / 收尾4 + 优先修正顺序 + 信息清单),学校公司名全部泛化
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

