← Back to Skills Marketplace
soulzhong

Recruiting Resume Screening

by SoulZhong · GitHub ↗ · v0.1.1 · MIT-0
cross-platform ⚠ suspicious
48
Downloads
0
Stars
1
Active Installs
2
Versions
Install in OpenClaw
/install recruiting-resume-screening
Description
Use when reviewing one or more Chinese-language resumes against a JD, batch-ranking candidates, identifying red flags and follow-up interview questions, or r...
README (SKILL.md)

简历筛选

Overview

按岗位 JD(或岗位标准包)对候选人做多维度评估,识别疑点与矛盾,给出排序与面试问题。

Core principle: 看证据,不看关键词。

简历出现关键词 ≠ 真的做过 ≠ 做深;必须看项目证据和个人贡献。

When to Use

  • 用户发送一份或多份简历 PDF 要求筛选。
  • 用户要求"根据 JD 评估这个候选人"。
  • 用户要求"对比这几个候选人"并给排序。
  • 用户要求"找出简历里的疑点 / 风险点"。
  • 用户要求"生成面试必问题"。
  • 用户要求"重新评估之前那批简历"。

如果任务覆盖 JD + 筛选 + 面评的完整流程,先加载 [[recruiting-skillset]]。

Prerequisites

  • 已有岗位 JD 或岗位标准包;没有则先从 JD 抽取核心要求。
  • 简历文件已在本地或可获取。
  • 本机可用 PDF 提取后端(任一即可,越多越好):fitz (PyMuPDF) / pdfplumber / pdfminer.six
  • 已了解招聘上下文:团队现状、业务阶段、用人风格。

Workflow

Step 1 — 获取并提取简历文本

不要默认单跑 pdfplumber。中文简历 PDF 常有页边跟踪串、竖排装饰字、隐藏文本层,单一后端不可靠。

推荐命令(脚本见 scripts/resume_pdf_extract.py):

# 单份 PDF
python3 scripts/resume_pdf_extract.py /path/to/resume.pdf --output-dir /tmp/resume_extract_one

# 批量 PDF / 一个目录
python3 scripts/resume_pdf_extract.py /path/to/resume_dir --output-dir /tmp/resume_extract_batch

脚本会:

  1. 分别尝试可用的 PDF 后端(fitz / pdfplumber / pdfminer.six)。
  2. 做 Unicode 归一化、空白折叠、跟踪串 / 碎片行清洗。
  3. 计算 score / chars / lines / keyword_hits / quality
  4. 选择分数最高的文本作为正式筛选输入。
  5. 输出 summary.json、每份 PDF 对应的最佳 .txt、Markdown 索引表。

详见 pdf-extraction.md

Step 2 — 批量简历必须先做质量分层

用户一次性发送多份简历时不要直接逐份硬读。先做一轮 triage,把能直接评估和需要补核的简历分开。详见 batch-screening.md

质量分层建议:

  • highchars >= 2600lines >= 55score >= 65
  • mediumchars >= 1600lines >= 35score >= 35
  • low:低于阈值,标记"待补核",不阻塞整批交付

Step 3 — 五维评估 + 疑点识别

每位候选人,从五个维度逐一评估,每维度 ★~★★★★★:

  1. 专业匹配度(Skill Fit)
  2. 履历稳定性(Career Stability)
  3. 潜在风险(Risk Assessment)
  4. 抗压与自驱(Resilience & Drive)
  5. 团队适配(Team Fit)

完整维度定义与打分依据见 five-dimension-rubric.md

逐条审读简历,主动发现疑点。完整疑点清单(时间类 / 内容类 / 逻辑类 / 行为类)见 red-flags-catalog.md。每个疑点标注严重程度:

  • 🔴 需面试必问
  • 🟡 需关注
  • 🟢 可能正常

Step 4 — 综合评级

评级 含义 条件
⭐⭐⭐ 强推 直通面试 专业匹配度 ≥ 4★ 且无重大风险,总得分 ≥ 20/25
⭐⭐ 推荐 进入面试 专业匹配度 ≥ 3★,总得分 ≥ 15/25
待定 需电话筛选 有亮点但存在明显短板或风险
不推荐 不进入面试 专业匹配度 \x3C 3★ 或存在不可接受风险

Step 5 — 输出评估报告

每人一份个人评估卡 + 整批汇总排序表。模板见 output-templates.md

Step 6 — 面试反馈复核(可选)

用户补充技术一面 / HR 二面 / 用人经理初评后,做一次"去偏见复核":

  • 把面试反馈与简历原文逐项对照,区分"有证据支撑"和"主观印象"。
  • "能力更强 / 更弱 / 成长性更好"这类结论,若没有具体案例支撑,标记为证据不足。
  • 自动化、性能、AI 提效等容易被泛化的能力,默认按"接触过"处理,除非有明确独立负责证据。
  • 优先检查时间线、工作年限、离职原因、薪资诉求等是否前后矛盾。

必须输出:总体结论(含排序调整)、逐人可疑点(🔴🟡🟢)、对已有面评的修正建议、终面追问(每人 ≥ 5 个 + 横向必问)、可选《终面准备说明》。

Red Lines(一票否决)

以下情况直接 ❌ 不推荐,无论其他维度多强:

  1. 简历造假或严重夸大(有明确证据)。
  2. 方向完全不相关且无转型意愿。
  3. 平均在职时间 \x3C 8 个月且无合理解释。
  4. 薪酬预期超出岗位预算 50% 以上。

Anti-Patterns(严格禁止)

反模式 后果 正确做法
凭记忆输出评估结论 凭印象漂移 必须先提取文本 → 五维评估 → 评级
复用旧会话结论 标准漂移、错过更新 重新提取、重新评估
省略个人评估卡 用户无法追溯结论 每人必须有完整评估卡
无文件评估 凭空编造 找不到简历必须告知用户并请求
"看起来像做了",3 行总结 失真 完整执行五维 + 疑点 + 面试题
把内部过程混入正式报告 干扰决策 文本提取索引、清洗日志只作内部材料
全员都给推荐 排序失去意义 敢下不推荐 / 待定

Self-Check

输出前逐项确认:

  • 每位候选人都提取了简历文本?
  • 每人都有五维 ★ 评分?
  • 每人都有疑点清单(🔴🟡🟢)?
  • 每人都有 3 个面试必问题?
  • 输出了汇总排序表?
  • 评级引用了简历中具体内容作为证据?
  • 不同结论的候选人在排序中区分开?
  • 报告里没有混入提取脚本明细、清洗日志等过程性内容?

Output Discipline

  • Markdown 格式,适合飞书 / Notion 粘贴。
  • 中文输出。
  • 最终面向用户的报告只保留与筛选决策直接相关的内容:个人评估卡、汇总排序、亮点、风险、疑点、面试问题、必要的待补核说明。
  • 不要输出与筛选无关的过程性版块:文本提取质量索引、backend/score/chars/lines 统计、清洗日志等只作内部工作材料,除非用户明确要求。
  • 评估结论要有理有据,引用简历中具体内容作为证据。
  • 不回避矛盾——候选人某方面强但另一方面有风险,两面都写。

Feedback Loop

完成筛选后主动询问用户:

  1. 评级是否准确?哪些人评高了 / 评低了?
  2. 是否遗漏了重要评估维度?
  3. 面试后的实际表现与评估是否一致?

用户反馈 → 更新本 Skill 的评估标准。新发现的风险模式 → 补充到 red-flags-catalog.md。重大变更 → 写入 EVOLUTION.md

See Also

Usage Guidance
Use this only with resumes and JDs you are authorized to process. Provide exact files or a dedicated folder, do not allow broad automatic searching of Downloads or attachment caches unless you have reviewed the scope, delete extracted text outputs after use, and require explicit approval before the agent modifies any skill standards or catalog files.
Capability Assessment
Purpose & Capability
The stated purpose, rubric, PDF extraction helper, and output templates are coherent for Chinese-language resume screening, but the workflow necessarily handles sensitive candidate data and produces consequential hiring recommendations.
Instruction Scope
Most steps are user-directed, but the re-screening workflow expands scope by directing the agent to search local notes, session history, Downloads, and attachment caches for prior materials instead of asking the user to provide or approve specific files.
Install Mechanism
There is no automatic install spec; the docs recommend user-run installation of public PDF libraries and running the included local Python script. This is purpose-aligned but unpinned.
Credentials
Reading explicitly provided resume PDFs is proportionate, but broad local cache/Downloads recovery and local persistence of extracted resume text need stricter scoping, user approval, and cleanup guidance.
Persistence & Privilege
The skill writes reusable extraction outputs and explicitly instructs updating the skill’s standards/catalog files from user feedback, without clear review, diff, or approval controls.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install recruiting-resume-screening
  3. After installation, invoke the skill by name or use /recruiting-resume-screening
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v0.1.1
Add homepage link to GitHub source repository (https://github.com/SoulZhong/recruiting-skills).
v0.1.0
Initial release of recruiting-resume-screening skill. - Provides structured workflow for evaluating Chinese-language resumes against job descriptions. - Emphasizes evidence-based, multi-dimensional assessment rather than keyword matching. - Includes PDF extraction guidelines using multiple backends for robust text retrieval. - Outlines step-by-step process: extraction, triage, five-dimension evaluation, red flag identification, and output discipline. - Enforces reporting standards and anti-patterns to ensure consistency and prevent bias. - Supplies self-checklists and feedback mechanisms for continuous improvement.
Metadata
Slug recruiting-resume-screening
Version 0.1.1
License MIT-0
All-time Installs 1
Active Installs 1
Total Versions 2
Frequently Asked Questions

What is Recruiting Resume Screening?

Use when reviewing one or more Chinese-language resumes against a JD, batch-ranking candidates, identifying red flags and follow-up interview questions, or r... It is an AI Agent Skill for Claude Code / OpenClaw, with 48 downloads so far.

How do I install Recruiting Resume Screening?

Run "/install recruiting-resume-screening" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Recruiting Resume Screening free?

Yes, Recruiting Resume Screening is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Recruiting Resume Screening support?

Recruiting Resume Screening is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Recruiting Resume Screening?

It is built and maintained by SoulZhong (@soulzhong); the current version is v0.1.1.

💬 Comments