Job Search System is a reusable Codex Skill for turning verified experience into defensible career positioning and resumes. It starts with candidate evidence; it never infers a candidate's capabilities from the wording of a job description or copies JD language as if it were personal experience.
It captures and reconciles experience, analyzes a target JD when tailoring is requested, develops an HRBP-level positioning strategy, drafts a resume only after approval, and audits the exact draft before delivery. It supports both general resumes and JD-tailored resumes while preserving traceability from every public claim back to its source.
The Job Search Orchestrator routes one valid next action and protects workflow gates. Five specialist roles have separate write scopes:
| Role | Responsibility |
|---|---|
| Experience Curator | Captures and reconciles candidate facts |
| JD Analyst | Records and analyzes a target JD without changing candidate facts |
| HRBP Strategist | Develops career positioning and capability mapping |
| Resume Writer | Produces the approved resume draft |
| Fact & Quality Auditor | Audits the exact draft against canonical evidence |
- Facts use four explicit states:
confirmed,estimated,inferred, andmissing. Evidence support is recorded as a separate dimension, not collapsed into fact state. - Attribution is explicit:
personal,team,company,client,supplier, orplatform. Team or company outcomes are never silently presented as individual achievements. - Metrics retain their period, denominator, inclusion scope, and exclusions. Career totals must be de-duplicated before use.
- Resume copy is blocked until the user approves the exact HRBP strategy ID and version. Delivery is blocked until the Fact & Quality Auditor returns
PASSfor the exact draft version.
Clone this repository, then copy the nested job-search-system directory into your personal Codex Skills directory:
git clone https://github.com/JessieLu615/job-search-system.git
copy job-search-system/job-search-system to $CODEX_HOME/skills/job-search-system
On Windows, the destination is commonly C:\Users\<username>\.codex\skills\job-search-system. <username> is a placeholder, not personal data. Restart or reload Codex if needed so the Skill is discovered.
Invoke the Skill as $job-search-system. Example prompts:
$job-search-system Initialize a job-search workspace in D:\career-workspace.
$job-search-system Analyze this target JD and identify evidence gaps without inventing candidate capabilities.
$job-search-system Continue from the current workspace status and perform the one next valid action.
The paths and prompts above are generic examples; use your own local workspace and JD source.
A standard workspace separates profile/, experience/, jobs/, strategy/, outputs/resumes/, and system/. The workflow uses seven valid stages. Tailored workflows include jd-analysis; general-resume workflows may route from experience-capture directly to hrbp-strategy. The valid stage vocabulary is:
experience-capturejd-analysishrbp-strategyawaiting-strategy-approvalresume-generationresume-auditcomplete
Each stage records one next valid action. A change to facts, analysis, strategy, or copy reopens the earliest affected gate rather than bypassing it.
A general resume requires verified experience and an explicitly approved HRBP strategy. A tailored resume requires those same prerequisites plus a completed analysis of the target JD. JD analysis may identify relevance and gaps, but it cannot create candidate facts.
Sequential execution is the default fallback. Subagents are optional only when they are available, the user permits them, and the tasks are independent. Two agents must never write concurrently to the same scope or artifact.
The published Skill package has verified evidence of 16/16 tests passing, official validation output Skill is valid!, workflow checks W1-W8 all PASS, an exact 19-file Skill inventory, and zero known author-PII or cache findings.
Candidate profiles, experience records, JDs, strategies, resumes, and audit artifacts remain local to the chosen workspace. Local workflow approval does not authorize publication, email, application submission, or any other external write; the user must separately authorize each external action and destination.
This repository is available under the MIT License. See LICENSE.
求职系统是一个可复用的 Codex Skill，用于把经过核实的经历转化为经得起追溯的职业定位与简历。系统从候选人的证据出发；绝不会根据职位描述（JD）的措辞反推候选人能力，也不会把 JD 文案当作个人经历照搬。
它能够采集并核对经历，在需要定制时分析目标 JD，形成 HRBP 视角的定位策略，在获得批准后起草简历，并在交付前审计该确切版本。系统同时支持通用简历和 JD 定制简历，并确保每一项公开表述都能追溯到来源。
求职编排器（Job Search Orchestrator）负责路由唯一有效的下一步，并保护流程门禁。五个专家角色拥有相互分离的写入范围：
| 角色 | 职责 |
|---|---|
| 经历策展人（Experience Curator） | 采集并核对候选人事实 |
| JD 分析师（JD Analyst） | 记录和分析目标 JD，但不修改候选人事实 |
| HRBP 策略师（HRBP Strategist） | 制定职业定位与能力映射 |
| 简历撰写者（Resume Writer） | 生成已批准的简历草稿 |
| 事实与质量审计员（Fact & Quality Auditor） | 依据规范证据审计确切草稿版本 |
- 事实使用四种明确状态：
confirmed（已确认）、estimated（估算）、inferred（推断）和missing（缺失）。证据支持度是独立维度，不能与事实状态混为一谈。 - 归因必须明确为
personal（个人）、team（团队）、company（公司）、client（客户）、supplier（供应商）或platform（平台）。团队或公司成果不得被悄然表述为个人成就。 - 指标必须保留统计周期、分母、纳入范围和排除项；职业生涯累计值在使用前必须去重。
- 在用户批准确切的 HRBP 策略 ID 与版本之前，不得生成简历文案；事实与质量审计员对确切草稿版本返回
PASS之前，不得交付。
克隆本仓库，然后把内层的 job-search-system 目录复制到个人 Codex Skills 目录：
git clone https://github.com/JessieLu615/job-search-system.git
将 job-search-system/job-search-system 复制到 $CODEX_HOME/skills/job-search-system
在 Windows 上，目标位置通常为 C:\Users\<username>\.codex\skills\job-search-system。<username> 是占位符，不是个人数据。如有需要，请重启或重新加载 Codex，使其发现该 Skill。
使用 $job-search-system 调用该 Skill。示例提示词：
$job-search-system 在 D:\career-workspace 初始化求职工作区。
$job-search-system 分析这份目标 JD，并识别证据缺口；不要虚构候选人能力。
$job-search-system 根据当前工作区状态继续，并执行唯一有效的下一步。
以上路径和提示词均为通用示例，请替换为你自己的本地工作区与 JD 来源。
标准工作区分设 profile/、experience/、jobs/、strategy/、outputs/resumes/ 和 system/。工作流使用七个合法阶段；JD 定制流程包含 jd-analysis；通用简历流程可以从 experience-capture 直接进入 hrbp-strategy。合法阶段如下：
experience-capturejd-analysishrbp-strategyawaiting-strategy-approvalresume-generationresume-auditcomplete
每个阶段只记录一个有效的下一步。事实、分析、策略或文案一旦变化，流程会重新打开最早受影响的门禁，而不会绕过它。
通用简历需要经过核实的经历和被明确批准的 HRBP 策略。定制简历除上述前提外，还需要完成目标 JD 分析。JD 分析可以识别相关性与缺口，但不能创造候选人事实。
顺序执行是默认回退模式。仅当子 Agent 可用、用户允许且任务彼此独立时，才可选择子 Agent。两个 Agent 绝不能并发写入同一范围或同一产物。
已发布 Skill 包的验证证据包括：16/16 项测试通过、官方验证输出 Skill is valid!、工作流检查 W1-W8 全部 PASS、精确的 19 文件 Skill 清单，以及已知作者个人身份信息（PII）和缓存发现均为零。
候选人档案、经历记录、JD、策略、简历和审计产物都保留在所选的本地工作区。对本地流程的批准不等于授权发布、发送邮件、提交申请或执行其他外部写入；每次外部操作及其目标都必须由用户另行授权。
本仓库采用 MIT License，详见 LICENSE。