八个拒绝虚构事实的求职Claude代码技能：简历优化器、STAR故事库、360Brew LinkedIn计划本、求职信、模拟面试、求职、git证据......
概览
通用的AI职业建议让人被拒绝：光鲜的空洞、虚构的指标、招聘人员三秒钟就能识别的“以结果为导向的专业”模板。这些技能的构建方式正好相反。它们的定义性行为是，他们，并且强化了写作技巧（动词、数字、自然发言），这种技巧既能经受ATS解析器，也能通过人类阅读。 在本地运行 Claude Code 或 Claude Desktop，在你自己的订阅中运行，数据存储在你的机器上。 这些技能还会组成每个任务的端到端流程： 审核员是一个独立的子代理，会审核草稿中的每一个声明，对照你的真实资料和故事进行审核;任何无来源的内容都会被标记删除，而不是“改进”。流程结束后，/mock-interview 会排练评估的真实不足，/upskill 则将 2+ 申请中反复出现的空白转化为学习计划。完整流程：docs/WORKFLOW.md。按技能合同：docs/SKILLS.md。
概述（AI 翻译）
JobMentis 背后的提示架构，开源为可安装的 Claude 技能。
通用的AI职业建议让人被拒绝：空洞无力、虚构指标、招聘人员三秒钟就能识别的“以结果为导向的专业”套路。这些技能的构建方式恰恰相反。它们的定义性行为是问你问题，而不是捏造关于你的事实，并且强化写作技巧（动词简单、数字、自然说出来）能通过ATS解析器和人类阅读器都能存活下来。
在本地运行 Claude Code 或 Claude Desktop，在你自己的订阅中运行，数据存储在你的机器上。
八大技能
| 技巧 | 它的作用 | 让它与众不同的规则 |
|---|---|---|
| 无情简历优化器 | 根据具体职位描述重写你的简历 | 两阶段流程：它诊断并询问可疑数字和空项目符号，等待你的回答，然后重写。空白处会诚实报告，绝不掩盖。 |
| 星辰故事提取器 | 把一个混乱的脑力倾倒变成犀利的STAR采访故事，存档到你的故事库 | 缺少指标：它只问你一个问题，而不是估算。 |
本页由 AI 翻译，技术细节请以原文为准。
README
以下为原文 README 节选，完整内容请查看原文。
Open Career Skills
The prompt architecture behind JobMentis, open-sourced as installable Claude skills.
Generic AI career advice gets people rejected: polished emptiness, invented metrics, “results-driven professional” boilerplate that recruiters spot in three seconds. These skills are built the opposite way. Their defining behavior is that they ask you a question rather than invent a fact about you, and they enforce the writing craft (plain verbs, real numbers, say-it-out-loud naturalness) that survives both ATS parsers and human readers.
Run them locally in Claude Code or Claude Desktop, on your own subscription, with your data on your machine.
The eight skills
| Skill | What it does | The rule that makes it different |
|---|---|---|
| ruthless-cv-optimizer | Rewrites your CV against a specific job description | Two-phase flow: it diagnoses and ASKS about suspect numbers and empty bullets, waits for your answers, then rewrites. Gaps are reported honestly, never papered over. |
| star-story-extractor | Turns a messy braindump into a sharp STAR interview story, filed in your story bank | Missing metric: it asks you ONE question instead of estimating. Also matches your bank against any pasted JD. |
| linkedin-360brew-planner | Plans and drafts LinkedIn posts under the 2026 ranking rules | Built on LinkedIn’s 360Brew decoder model (the reason hashtags and pods died). Saves over likes, analytical first line, pillar discipline. |
| cover-letter-writer | A 300-400 word letter grounded in your real stories | Two or three deep JD connections instead of six shallow ones; banned-cliche list enforced. |
| mock-interviewer | A pressure-tested rehearsal, one question at a time | Stays in character, interrupts rambling, drills “what did YOU do?”, then debriefs against structure / specificity / ownership / quantification. |
| job-search | Finds current postings matching your profile | Zero dependencies; only reports jobs it actually fetched or you pasted. Blocked boards get you ready-made search URLs, never invented listings. |
| evidence-miner | Harvests real achievements from your git history into the story bank | Groundedness rule: no citable artifact (commits, PRs, dates), no proposed achievement. Business impact numbers come from you, never inferred from code. |
推荐工具
换一个关键词，或者移除筛选条件。
安装
npx skillfish add squerne/open-career-skills