XINJIE ZHANG 求职面试 AI Skill / 工作流AI Skill and Workflow for Job Applications and Interviews
把「看懂岗位 → 匹配我的经历 → 判断是否申请 → 准备自我介绍和定向简历 → 预测面试题 → 面试后复盘归档」这一整条求职链路，做成一套可以对不同岗位反复运行的方法。它由两个相互衔接的 Skill 组成：「求职副驾」负责整理和判断，「面试官视角」负责独立评估。 A method that runs the whole job-search chain — read the role, match my evidence, decide whether to apply, prepare an introduction and a role-specific résumé, predict interview questions, then debrief and archive — repeatably across different roles. It is built from two connected Skills: a job-application co-pilot that organises and judges, and an independent interviewer layer that assesses.
为什么做Why I made it
求职信息很分散，判断步骤又多——看 JD、对经历、改简历、准备面试、面试后复盘，每一步都容易出错，也很容易把 AI 的推测当成事实写进简历。我想要一套既能省时间、又能守住「每句话都有出处」的流程，而且换一个岗位还能重新跑。 Job-search information is scattered and the process has many judgement calls — reading the JD, matching experience, editing the résumé, preparing for interviews, debriefing afterwards. Each step is easy to get wrong, and it is easy to let AI inference slip into the résumé as fact. I wanted a process that saved time while keeping every sentence traceable to a source, and that I could re-run for a different role.
我做了什么What I did
我把整条链路拆成明确的步骤，定下「用户确认 > 原始材料 > 证据笔记 > AI 推断」的取信顺序，规定母简历只作事实源、不被覆盖，每份定向简历单独出文件。我建了简历事实核验流程，又加了「面试官视角」这一层——它可以独立建岗位评分卡、出题、逐题追问、给候选人反馈。最后把这些方法沉淀成两个可复用的 Skill，并接上面试后复盘和知识库归档。 I broke the chain into explicit steps and fixed a source hierarchy — user-confirmed facts > original artifacts > evidence notes > AI inference. The master résumé is a fact source that is never overwritten; every targeted résumé is a separate file. I built a résumé fact-check pass and added an interviewer layer that can independently build a role scorecard, write questions, probe answer by answer, and give candidate feedback. Both were turned into reusable Skills, connected to a post-interview debrief and an archive step in my knowledge base.
AI 帮了什么How AI helped
AI 负责按流程执行：拆解 JD、在材料里检索证据、把要求和经历做成对照表、生成定向简历草稿和自我介绍、预测面试题、在模拟面试里扮演面试官。 AI runs the steps: breaking down the JD, searching my materials for evidence, building a requirement-to-evidence table, drafting the targeted résumé and self-introduction, predicting interview questions, and playing the interviewer in a mock interview.
我负责的判断What I decided
我定义每一步要产出什么、取信顺序怎么排；确认哪些是事实、哪些是过度推断；决定哪些经历能用、简历最终怎么写、要不要申请这个岗位。流程能提速，判断和最终结果由我负责。 I define what each step must produce and how the source hierarchy is ordered; I decide what is fact and what is over-inference; I decide which experiences are usable, how the résumé is finally worded, and whether to apply. The process speeds things up; the judgement and the final result are mine.
工作流的样子How the workflow runs
给一份真实的 JD，「求职副驾」会依次跑完下面八步，产出一份可追溯的求职准备文档；真实笔试或面试阶段再调用「面试官视角」。Given a real job description, the co-pilot runs these eight steps into one traceable preparation document; the interviewer layer is called for the real written-test or interview stage.
-
看懂岗位Read the role
把 JD 拆成「显性事实」和「需要推断的部分」：这个岗位到底解决什么问题、最重要的三项能力、必须满足的硬条件、面试大概会考什么。每条结论都贴一小段 JD 原文，再写出从原文到结论的推理；没有原文支撑的就标成推断或未知。Split the JD into explicit facts and interpretation: what problem the role solves, its three most important capabilities, the must-have conditions, and what the interview will likely test. Every conclusion carries a short verbatim JD excerpt and the reasoning from it; anything unsupported is labelled inference or unknown.
-
匹配我的经历Match my evidence
在母简历和个人材料里找具体证据，对每条岗位要求标「强 / 部分 / 弱 / 缺失」，并对每段经历分清是我做的、队友做的、还是 AI 做的。不把性格标签当证据。Search the master résumé and personal materials for concrete evidence, rate each requirement strong / partial / weak / missing, and separate my contribution from teammates' and AI's for every experience. Personality labels don't count as proof.
-
判断优势与缺口Advantages and gaps
把能力分成三类：有真实经历能证明的、自己以为有但证据不足的、目前明显缺的。每个缺口都配三条：投递前怎么补、面试前准备什么证据、补不齐时怎么诚实说。Sort abilities into three groups — proven by real experience, believed-but-under-evidenced, and clear current gaps. Each gap gets three lines: how to strengthen it before applying, what evidence to prepare before the interview, and how to state it honestly if it can't be closed in time.
-
判断是否值得申请Decide whether to apply
做双向匹配：招聘方为什么可能选我 / 为什么可能不选，这个岗位对我的价值 / 风险和代价。给出建议和适用条件，但不用一个「匹配分数」当结论，也不替我做决定。A two-way fit check: why the employer might pick me and why they might not; the role's value to me and its risks and costs. It gives a recommendation with conditions, but never reduces the decision to a match score and never decides for me.
-
准备自我介绍Prepare the introduction
按场景和听众，用 FAB（特征 / 优势 / 对听众的价值）挑 2–3 个有证据的点，做 1 分钟、2 分钟、3 分钟三个版本，并说明每个版本侧重什么、比上一版多讲了什么。Using FAB (Feature / Advantage / Benefit) and picking two or three evidence-backed points for the audience, it drafts one-, two- and three-minute versions and explains what each emphasises and what the longer one adds.
-
生成定向简历Generate the targeted résumé
从母简历里已核实的事实中挑选、重排，生成一份针对这个岗位的简历（母简历不被改写，定向简历单独成文件）。每条 bullet 追溯到证据来源，不编造指标、头衔、日期、技术或归属；有公开链接的作品用可点击链接。附一份「改动说明」：留了什么、删了什么、为什么。It selects and reorders verified facts from the master résumé into a role-specific résumé (the master is never rewritten; the targeted one is its own file). Every bullet traces to a source; no invented metrics, titles, dates, technologies or ownership; works with a public link get a clickable link. A change log records what was kept, cut and why.
-
预测面试题Predict interview questions
从排过序的要求和证据弱点里预测十个高概率问题，不套题库。每题标：考察什么、我可以用哪段经历、可能的追问、目前的准备缺口。From the ranked requirements and evidence weaknesses it predicts ten likely questions — not from a generic bank. Each is tagged with what it assesses, which experience I can use, the likely follow-up, and my current preparation gap.
-
练习与复盘Practise and debrief
两种模式：教练式逐题演练（一次一题，答完给反馈和更好的版本）；真实模拟（交给「面试官视角」那一层，过程中不指导、不给标准答案，结束后统一评估）。真实投递或面试之后做结构化复盘，把新确认的事实加进个人证据库，只有真正能改进以后决策的经验才写回方法本身，最后整个周期归档进知识库。Two modes: a coached one-question-at-a-time drill (feedback and a stronger version after each answer), and a realistic simulation handed to the interviewer layer (no coaching, no model answers during it, one consolidated assessment at the end). After a real application or interview, a structured debrief adds newly confirmed facts to the evidence bank; only lessons that improve future decisions are written back into the method, and the whole cycle is archived in my knowledge base.