Skip to content
Back to search

llm-intern-skill

Use when polishing, diagnosing, tailoring, or exporting resumes for LLM, RAG, Agent, Agentic RL, post-training, pretraining, AIGC, search/ranking, multimodal, AI backend, or LLM algorithm internships from raw resume text, a materials folder, and/or a target job description. Audits evidence, maps JD fit, enforces truth boundaries, writes polished and targeted resumes, generates interviewer-style grilling questions, answer cards, evidence-upgrade plans, and optional open-source project recommendations without fabricating experience.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/wanyichen06-llminternskill/badge.svg)](/listings/wanyichen06-llminternskill)

HTML

<a href="/listings/wanyichen06-llminternskill"><img src="/api/listings/wanyichen06-llminternskill/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 138 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. ---
  2. name: llm-intern-skill
  3. description: Use when polishing, diagnosing, tailoring, or exporting resumes for LLM, RAG, Agent, Agentic RL, post-training, pretraining, AIGC, search/ranking, multimodal, AI backend, or LLM algorithm internships from raw resume text, a materials folder, and/or a target job description. Audits evidence, maps JD fit, enforces truth boundaries, writes polished and targeted resumes, generates interviewer-style grilling questions, answer cards, evidence-upgrade plans, and optional open-source project recommendations without fabricating experience.
  4. ---
  5. # LLMInternSkill
  6. Use this Skill when the user wants resume polish, resume diagnosis, JD tailoring, project packaging, interview preparation, or final resume export for LLM-related internship applications.
  7. Core rule:
  8. ```text
  9. Do not fabricate. Diagnose first, polish second.
  10. ```
  11. ## Inputs
  12. Preferred input folder:
  13. ```text
  14. materials/
  15. ├── target_jd.txt
  16. ├── resume.md / resume.pdf
  17. ├── projects/
  18. ├── code/
  19. ├── notes/
  20. ├── papers/
  21. ├── awards/
  22. └── other/
  23. ```
  24. If the user only provides a JD and no materials, ask the intake questions from `templates/intake.md`.
  25. If the user only asks for resume polish, run a lightweight version:
  26. ```text
  27. raw resume line -> claim extraction -> evidence/risk check -> polished wording -> interview risk
  28. ```
  29. ## Main Workflow
  30. 1. **Decide the mode**
  31. - Resume polish only: use `skill-references/resume-polish.md`.
  32. - JD tailoring: use `skill-references/jd-analysis.md` and `skill-references/resume-tailoring.md`.
  33. - Full materials folder: run the complete workflow below.
  34. - Interview prep only: use `skill-references/interview-grilling.md` and `skill-references/answer-cards.md`.
  35. - Project Scout only: use `skill-references/project-scout.md`.
  36. 2. **Read the target JD when present**
  37. - Use `skill-references/jd-analysis.md`.
  38. - Detect role type: RAG, Agent, Agentic RL, post-training, pretraining, LLM app, LLM algorithm, search/ranking, AIGC, multimodal, backend AI, infra, or mixed.
  39. - Load the matching role file under `skill-references/roles/` when relevant.
  40. 3. **Audit the materials folder when present**
  41. - Use `skill-references/materials-audit.md`.
  42. - Extract projects, claims, evidence, missing evidence, and unclear ownership.
  43. 4. **Set truth boundaries**
  44. - Use `skill-references/truth-boundary.md`.
  45. - Classify content as `可以写`, `谨慎写`, `补证据后写`, `不能写`, or `无法判断`.
  46. 5. **Build the evidence contract**
  47. - Use `skill-references/evidence-contract.md`.
  48. - Every strong claim needs evidence, risk, safe wording, and interview proof.
  49. 6. **Generate polished / targeted resume**
  50. - Use `skill-references/resume-polish.md` for line-level polish.
  51. - Use `skill-references/resume-tailoring.md`.
  52. - Produce conservative, standard, and stronger-after-evidence bullets.
  53. - Generate a targeted full resume draft when enough information exists.
  54. - If the user wants a PDF-ready resume, use `templates/resume-latex/bill-ryan-elegant-zh_CN/resume-zh_CN.tex` as the LaTeX base.
  55. 7. **Generate interview grilling**
  56. - Use `skill-references/interview-grilling.md`.
  57. - Ask interviewer-style questions based on JD gaps and resume claims.
  58. 8. **Generate answer cards**
  59. - Use `skill-references/answer-cards.md`.
  60. - For high-risk questions, produce dangerous / passable / strong answers.
  61. 9. **Create upgrade plan**
  62. - Use `skill-references/upgrade-plan.md`.
  63. - Split into half-day, 1-day, 3-day, and 1-week evidence upgrades.
  64. 10. **Optional Project Scout**
  65. - Use `skill-references/project-scout.md` when the user's evidence is weak or they ask for projects to learn.
  66. - Recommend projects only as learning/reproduction/modification opportunities, not as fake experience.
  67. 11. **Assemble final pack**
  68. - Use `templates/final-pack.md`.
  69. ## Output Files
  70. When writing files, prefer this structure:
  71. ```text
  72. output/
  73. ├── 01_jd_analysis.md
  74. ├── 02_materials_audit.md
  75. ├── 03_truth_boundary.md
  76. ├── 04_evidence_contract.md
  77. ├── 05_resume_polish.md
  78. ├── 06_targeted_resume.md
  79. ├── 07_interview_grilling.md
  80. ├── 08_answer_cards.md
  81. ├── 09_upgrade_plan.md
  82. ├── 10_project_scout.md
  83. └── 11_final_pack.md
  84. ```
  85. If the user wants only an answer in chat, still follow the same section order.
  86. ## Fit Verdict
  87. Always give one:
  88. ```text
  89. strong fit
  90. weak fit
  91. risky fit
  92. not recommended
  93. ```
  94. Explain the verdict with:
  95. - JD must-haves.
  96. - User evidence.
  97. - Gaps.
  98. - Highest interview risk.
  99. - Fastest useful upgrade.
  100. ## Non-Negotiables
  101. - Never invent internships, production status, metrics, user scale, model training, ranking gains, or ownership.
  102. - Do not write "主导" when evidence only supports "参与".
  103. - Do not write "上线" when evidence only supports demo, local run, or internal trial.
  104. - Do not write open-source learning as work experience unless the user actually reproduced, modified, and documented it.
  105. - If materials are insufficient, ask questions or produce a conservative report instead of polished fiction.
README.md · README · 407 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. <div align="center">
  2. # LLMInternSkill / LLM Intern Skill
  3. [English](README_EN.md) | **简体中文**
  4. **大模型实习简历与求职工具箱:LLM internship resume polish, JD tailoring, RAG / Agent / Post-training 项目包装、面试拷打、开源项目补强。**
  5. <sub>一个面向大模型实习、AI 求职和 Codex Skill 工作流的 evidence-bound resume toolkit:不只是把简历写好听,而是让每一行都能扛住面试追问。</sub>
  6. <sub>Keywords: LLM internship resume, 大模型实习简历, Codex Skill, resume polish, JD tailoring, interview prep, RAG, Agent, post-training, pretraining, search ranking, AIGC, multimodal.</sub>
  7. <br>
  8. <br>
  9. ![Type](https://img.shields.io/badge/type-Codex%20Skill-111827)
  10. ![License](https://img.shields.io/badge/license-MIT-blue)
  11. ![Focus](https://img.shields.io/badge/focus-LLM%20Internship-4f46e5)
  12. ![Method](https://img.shields.io/badge/method-evidence--bound-f59e0b)
  13. ![Language](https://img.shields.io/badge/language-ZH%20%2B%20EN-10b981)
  14. ![Status](https://img.shields.io/badge/status-final%20MVP-ec4899)
  15. ![GitHub Repo stars](https://img.shields.io/github/stars/couragec/llm-intern-skill?style=flat&label=stars)
  16. [30 秒看懂](#30-秒看懂) · [快速开始](#快速开始) · [Seed 示例](#旗舰示例豆包-seed-搜索排序) · [功能地图](#功能地图) · [Star History](#star-history) · [参考](#参考)
  17. </div>
  18. ---
  19. ## 30 秒看懂
  20. | 你给它什么 | 它帮你做什么 |
  21. | --- | --- |
  22. | 一份原始简历 | 📝 润色成更清楚、更技术、更像实习候选人的表达 |
  23. | 一个目标 JD | 🎯 判断匹配度,重排经历,生成定制版简历 |
  24. | 一个材料文件夹 | 🔍 审计代码、项目素材、论文笔记、日志、截图和奖项 |
  25. | 一堆单薄项目 | 🧪 包装成真实、可讲、能补证据的项目故事 |
  26. | 一份担心被问穿的简历 | 🎤 生成面试官式追问、危险回答、及格回答、强回答 |
  27. | 证据不足 | 🌱 推荐开源项目学习、复现、改造,形成新证据 |
  28. | 最终要投递 | 📄 用 Bill Ryan LaTeX 模板生成 PDF-ready 简历草稿 |
  29. 核心判断:
  30. ```text
  31. 可以写 / 谨慎写 / 补证据后写 / 不能写 / 无法判断
  32. ```
  33. 核心产出:
  34. ```text
  35. raw resume + materials/ + target_jd.txt
  36. -> 简历润色
  37. -> JD 匹配
  38. -> 真实性边界
  39. -> Evidence Contract
  40. -> 定制简历
  41. -> 面试拷打
  42. -> 回答卡
  43. -> 补证据计划
  44. -> Project Scout
  45. -> LaTeX 简历草稿
  46. ```
  47. ---
  48. ## 一眼 Before / After
  49. <table>
  50. <tr>
  51. <th width="28%">原始简历</th>
  52. <th width="36%">安全润色版</th>
  53. <th width="36%">为什么这样写</th>
  54. </tr>
  55. <tr>
  56. <td>熟悉大模型与 RAG,优化搜索效果。</td>
  57. <td>围绕搜索相关性场景整理 query-doc 样例与长尾查询 bad case,分析歧义查询、时效性不足和低权威 DOC 对检索结果的影响。</td>
  58. <td>没有指标时不写“优化效果”;先把真实证据写成问题分析能力。</td>
  59. </tr>
  60. <tr>
  61. <td>完成企业级知识库问答系统。</td>
  62. <td>完成企业文档问答 demo 的文档切分、向量检索和 Prompt 模板配置,记录召回错误、引用错位和无依据回答等问题。</td>
  63. <td>没有上线、权限、监控和用户记录时,不写“企业级系统”。</td>
  64. </tr>
  65. <tr>
  66. <td>训练大模型并提升生成质量。</td>
  67. <td>复现小参数 LLM 的训练与推理流程,记录配置、日志、样例输出和 bad case,形成端到端理解笔记。</td>
  68. <td>没有数据、配置、训练日志和 checkpoint 时,不写“训练大模型”。</td>
  69. </tr>
  70. </table>
  71. 一句话原则:
  72. ```text
  73. 润色可以让真实经历更清楚、更有技术含量;
  74. 但不能把没有证据的经历润色成事实。
  75. ```
  76. ---
  77. ## 快速开始
  78. ### 方式一:只想润色简历
  79. ```text
  80. 使用 LLMInternSkill。
  81. 请帮我做简历润色,但不要编造经历。
  82. 目标岗位:
  83. [贴 JD 或写岗位方向]
  84. 我的原始简历:
  85. [贴简历内容]
  86. 请输出:
  87. 1. Before / After
  88. 2. 哪些行可以写、哪些需要降级
  89. 3. 技术表达增强版
  90. 4. JD 定制版
  91. 5. 面试可能追问的问题
  92. ```
  93. ### 方式二:完整材料文件夹
  94. 准备:
  95. ```text
  96. materials/
  97. ├── target_jd.txt
  98. ├── resume.md
  99. ├── projects/
  100. ├── code/
  101. ├── notes/
  102. ├── papers/
  103. ├── awards/
  104. └── other/
  105. ```
  106. 然后说:
  107. ```text
  108. Use LLMInternSkill on ./materials.
  109. Generate resume polish, JD fit verdict, targeted resume, interview grilling,
  110. answer cards, evidence upgrade plan, and project scout recommendations.
  111. ```
  112. ### 方式三:安装成 Codex Skill
  113. ```bash
  114. mkdir -p ~/.codex/skills
  115. git clone https://github.com/couragec/llm-intern-skill.git ~/.codex/skills/llm-intern-skill
  116. ```
  117. 重启 Codex 或开启新会话后,直接说:
  118. ```text
  119. 使用 LLMInternSkill,读取我的 materials/ 文件夹。
  120. ```
  121. ---
  122. ## 旗舰示例:豆包 Seed 搜索排序
  123. 这个示例故意选了一个强 JD:豆包大模型团队 Seed 搜索 / Ranking / DOC 理解研究实习。
  124. 它展示了这个项目最核心的能力:
  125. ```text
  126. 强 JD + 弱/中等材料
  127. -> 不硬吹 strong fit
  128. -> 找到可写证据
  129. -> 降级危险表达
  130. -> 生成能投递的保守简历
  131. -> 生成面试官会追问的问题
  132. -> 给出 1 天 / 3 天 / 1 周补证据计划
  133. ```
  134. 直接看三份文件:
  135. - [目标 JD](examples/doubao-seed-search-ranking-jd.md)
  136. - [候选人材料输入](examples/doubao-seed-materials-input.md)
  137. - [最终输出包](examples/doubao-seed-final-pack.md)
  138. - [模拟 LaTeX 简历 PDF](examples/seed-resume-latex/seed-topseed-resume.pdf)
  139. 关键输出片段:
  140. ```text
  141. Verdict: risky fit
  142. Why: 有 RAG 和小搜索 demo,但缺 ranking metrics、强算法实习证据和 DOC 理解实验。
  143. Fastest upgrade: 把 mini search demo 补成 BM25 vs embedding vs rerank 对比,加入 NDCG@10 / MRR 和 bad-case 分析。
  144. ```
  145. ---
  146. ## 功能地图
  147. | Icon | 模块 | 输出 |
  148. | --- | --- | --- |
  149. | 📝 | Resume Polish | Before/After、技术表达增强、CN/EN 版本、ATS 关键词 |
  150. | 🎯 | JD Tailoring | JD match table、定制版简历、岗位关键词重排 |
  151. | 🔍 | Resume Diagnosis | 弱表达、风险 claim、缺证据、被问穿点 |
  152. | 🧪 | Project Packaging | 项目 bullet、架构叙事、2-3 分钟项目故事 |
  153. | 🛡 | Evidence Guard | Truth Boundary、Evidence Contract、危险表达降级 |
  154. | 🎤 | Interview Drill | 逐行拷打、追问链、危险/及格/强回答 |
  155. | 🌱 | Project Scout | 开源项目推荐、最小运行路径、改造点、可写证据 |
  156. | 📄 | LaTeX Export | Bill Ryan 中文 LaTeX 模板、PDF-ready 简历草稿 |
  157. ---
  158. ## 输出长什么样
  159. 完整模式会生成:
  160. ```text
  161. output/
  162. ├── 01_jd_analysis.md
  163. ├── 02_materials_audit.md
  164. ├── 03_truth_boundary.md
  165. ├── 04_evidence_contract.md
  166. ├── 05_resume_polish.md
  167. ├── 06_targeted_resume.md
  168. ├── 07_interview_grilling.md
  169. ├── 08_answer_cards.md
  170. ├── 09_upgrade_plan.md
  171. ├── 10_project_scout.md
  172. └── 11_final_pack.md
  173. ```
  174. 简历润色模式会优先输出:
  175. ```text
  176. 1. Resume Polish Verdict
  177. 2. Before / After
  178. 3. Line-by-line diagnosis
  179. 4. Final polished resume section
  180. 5. Optional English version
  181. 6. Evidence upgrade list
  182. ```
  183. ---
  184. ## 支持岗位
  185. | 方向 | 会重点检查什么 |
  186. | --- | --- |
  187. | RAG / 知识库问答 | chunk、retrieval、rerank、citation、eval、拒答、权限 |
  188. | Agent / Tool Use | tool schema、参数校验、trace、重试、状态、人审 |
  189. | Agentic RL | trajectory、environment、reward/verifier、GRPO/PPO、任务成功率 |
  190. | Post-training / Alignment | SFT、DPO、RLHF/RLAIF、reward model、偏好数据、评估 |
  191. | Pretraining / Mid-training | 数据引擎、tokenizer、训练配置、loss、分布式、污染检查 |
  192. | LLM 应用工程 | structured output、prompt 版本、bad case、成本、延迟 |
  193. | LLM 算法 | Transformer、SFT/LoRA/DPO、训练配置、日志、评测 |
  194. | 搜索 / 推荐 / 排序 | query-doc、召回、粗排、rerank、NDCG/MRR、DOC 质量 |
  195. | AIGC | 生成链路、质量评估、安全、人工复核、成本 |
  196. | 多模态 | OCR/layout、VLM、跨模态检索、视觉 bad case |
  197. | AI 后端 | API、队列、重试、观测、权限、部署、成本 |
  198. ---
  199. ## 为什么不只是简历润色器
  200. 普通简历润色器经常只做这件事:
  201. ```text
  202. 把一句普通话改成一句漂亮话。
  203. ```
  204. LLMInternSkill 要多走三步:
  205. ```text
  206. 这句话有没有证据?
  207. 这句话会被面试官怎么追?
  208. 如果现在不能写,补什么证据后能写?
  209. ```
  210. 所以它不会把:
  211. ```text
  212. 做过 RAG demo
  213. ```
  214. 硬包装成:
  215. ```text
  216. 主导企业级智能知识库系统上线,显著提升业务效率。
  217. ```
  218. 它会更倾向于输出:
  219. ```text
  220. 完成企业文档问答 demo 的文档切分、向量检索和 Prompt 模板配置,
  221. 记录召回错误、引用错位和无依据回答等问题,
  222. 为后续 query-answer-evidence 评估集建设沉淀样例。
  223. ```
  224. 这就是它的差异化:**好听,但不虚。**
  225. ---
  226. ## Project Scout
  227. 当当前材料太弱时,LLMInternSkill 会建议你补项目,而不是硬写。
  228. 示例:
  229. - MiniMind-style from-scratch LLM 项目
  230. - RAG evaluation / citation accuracy 项目
  231. - Search / rerank baseline 项目
  232. - Agent tool-calling workflow 项目
  233. - LLM inference / serving / quantization 项目
  234. 每个推荐都必须包含:
  235. ```text
  236. why this fits
  237. why not / risk
  238. minimum run path
  239. what to modify
  240. what evidence to collect
  241. resume-safe claim after completion
  242. interview grilling questions
  243. ```
  244. 边界:
  245. ```text
  246. 学习开源项目不能直接包装成工作经历。
  247. 只有真实复现、理解、改造、记录证据后,才能写成个人项目或开源实践。
  248. ```
  249. 示例:[MiniMind Project Scout](examples/project-scout-minimind.md)
  250. 更多例子:
  251. - [Resume Polish Before / After](examples/resume-polish-before-after.md)
  252. - [Prompt-only Work Safely Downgraded](examples/prompt-only-downgrade.md)
  253. - [Search / Rerank Project Scout](examples/project-scout-search-rerank.md)
  254. - [LoRA Experiment With Weak Metrics](examples/lora-weak-metrics.md)
  255. - [Post-training / Agentic RL Case](examples/posttraining-agentic-rl-case.md)
  256. - [Seed Resume LaTeX Demo](examples/seed-resume-latex/README.md)
  257. ---
  258. ## LaTeX 简历模板
  259. 默认使用 Bill Ryan's elegant LaTeX resume 的中文模板:
  260. ```text
  261. templates/resume-latex/bill-ryan-elegant-zh_CN/
  262. ```
  263. 来源:
  264. - Overleaf: <https://www.overleaf.com/latex/templates/bill-ryans-elegant-latex-resume/xcqmhktmzmsw>
  265. - Upstream: <https://github.com/billryan/resume>
  266. 说明:
  267. - 中文入口:`resume-zh_CN.tex`
  268. - 编译方式:XeLaTeX
  269. - 保留 upstream attribution 和 license
  270. - 大号 CJK 字体本地保留但不进 Git
  271. ---
  272. ## Star History
  273. [![Star History Chart](https://api.star-history.com/svg?repos=couragec/llm-intern-skill&type=Date)](https://www.star-history.com/#couragec/llm-intern-skill&Date)
  274. ---
  275. ## 参考
  276. 本项目参考和学习了以下开源项目,感谢这些项目的启发:
  277. - [`resumify`](https://github.com/breaker505/resumify)
  278. - [`shushu-internship-resume-optimizer`](https://github.com/Sunanzhe2004/shushu-internship-resume-optimizer)
  279. - [`shushu-internship-tool`](https://github.com/LiuMengxuan04/shushu-internship-tool)
  280. - [`shushu-ProjectProof`](https://github.com/YingaoWang-casia/shushu-ProjectProof)
  281. - [`vibe-resume`](https://github.com/LiuMengxuan04/vibe-resume)
  282. - [`resuml`](https://github.com/phoinixi/resuml)
  283. - [`Auto-claude-code-research-in-sleep`](https://github.com/wanshuiyin/Auto-claude-code-research-in-sleep)
  284. ---
  285. ## 仓库结构
  286. ```text
  287. .
  288. ├── SKILL.md
  289. ├── skill-references/
  290. │ ├── resume-polish.md
  291. │ ├── resume-tailoring.md
  292. │ └── roles/
  293. ├── templates/
  294. │ ├── resume-polish-report.md
  295. │ └── resume-latex/
  296. ├── examples/
  297. ├── evals/
  298. ├── docs/
  299. └── agents/
  300. ```
  301. 欢迎通过 issue / PR 补充更多 JD、案例、角色参考和 eval。
  302. ## License
  303. MIT. The bundled Bill Ryan resume template keeps its upstream license and attribution; see `templates/resume-latex/README.md`.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.