get-job.skill: an Agent Skills workflow that rewrites a resume and builds round-by-round interview prep

The problem get-job.skill targets: real experience that reads as unrelated

The README opens with a specific diagnosis. The hard part of a job search is not a lack of experience but that the experience looks unrelated to the target role, or that a rewritten resume collapses the moment an interviewer asks a follow-up question. The project is aimed at candidates whose background is genuine but whose language does not match the posting: the README gives the examples of an education graduate applying to finance, a psychology graduate moving into AI, and a humanities student going into internet operations.

The claim in the README is that these moves work through translation rather than invention. That framing sets the scope of the tool. It is not a resume keyword generator and it is not an interview answer bank. It is a three-stage pipeline where each stage feeds the next, and the README states the dependency plainly: if the direction is wrong, a beautifully rewritten resume is useless. That is why job research comes first and produces a file before any rewriting happens.

Three stages, four generated artifacts, and a gate before delivery

The pipeline is defined in SKILL.md, which the README describes as the main flow that the skill runtime reads. Stage one researches the target role using WebSearch plus source grading, selecting local job boards and interview-report sources based on the target market, and writes 岗位调研.md with core capability keywords, hidden thresholds and recent developments. Stage two reverse-engineers the target role from that research and rewrites the resume, producing 改后简历.docx through scripts/generate_resume.py. Stage three discovers the actual interview rounds and produces a 面试准备/ folder containing an overview, a per-bullet deep-dive file, a self-introduction and delivery file, one file per confirmed or high-confidence round, and a post-interview review question bank.

The mechanism that distinguishes this from a prompt is the input check that runs before any generation. The README states that the skill first audits which inputs are complete (target role, target market, JD, resume, application status, interview rounds), which can be auto-filled, and which can only be degraded at low confidence. If the JD is missing it looks for an official careers page or a comparable JD. If the rounds are unknown it produces only an overview and bullet deep-dives. If the resume is incomplete it outputs a list of material gaps instead of inventing content.

A second mechanism sits at the other end. Before delivery, a quality gate checks source coverage, JD-to-resume matching, resume-bullet-to-interview-prep coverage, the evidence level of each claim, the basis for each round, and tag leakage in public files. The README says failing a red line blocks delivery, that anything degradable is degraded, and that anything unverifiable is marked low confidence. The gate logic lives in references/quality-gates.md and is scored P0/P1/P2, while references/regression-evals.md holds the test scenarios: cross-market applications, missing JD, already-submitted resumes, and inflated claims.

Installing get-job.skill and running a first job search

The README gives three installation routes. The recommended one is a single command, which is the fastest way to confirm the skill resolves in your runtime:

npx skills add agentenatalie/get-job.skill

If you prefer to control the location, the manual route clones the repository into your runtime's skills directory. The README shows the Claude Code path explicitly and notes that Codex and other runtimes use their own skills directory:

git clone https://github.com/agentenatalie/get-job.skill ~/.claude/skills/get-job

There is also a no-install route. The README states that you can hand the playbooks under references/ to any AI and ask it to follow the same method for resume rewriting and interview preparation.

Once installed, the interaction is a single message. The README's example is to paste your resume and the target role, then say: "这是我的简历 [贴上简历内容 / 拖入 PDF/Word]，我想申请 [公司] 的 [岗位]，JD 如下 [贴上 JD]，帮我准备。" The skill then runs the three stages and writes files. If your resume is already submitted, the README says to tell it to prepare the interview from the current resume, which skips stage two.

Expect confirmation points rather than a single dump. The README lists them: before the resume is finalized, at the per-bullet deep-dive table, at the self-introduction and delivery state, and before the round-by-round scripts. Each is a checkpoint where you review before it continues.

The honesty rails are constraints, not guarantees

The README devotes a section to what it calls the honesty floor, and it is worth reading as a limitation list rather than a feature list. Four points stand out. First, translation is not fabrication: restating something you actually did from a different angle is translation, while writing something you never did is fabrication that collapses under a follow-up question. Second, every claimed transferable skill must map to a real event. Third, the skill will ask for evidence and flag risk, but the README states directly that it cannot prove your input was honest; a claim with no supporting detail cannot be written as a strong fact. Fourth, background checks are a hard line: degree, employment dates and job titles must not change by a single character.

The practical consequence is that the quality gate is only as good as the inputs it is given. A candidate who supplies a vague resume gets a low-confidence output, not a polished one, because the degradation path is the designed response to missing evidence. The README also draws a line around AI-assisted work, arguing that stating it honestly beats pretending to be a senior engineer and being exposed under questioning. That is a stance about how to present your history, and it will not suit anyone looking for the tool to close gaps in it.

Where a general-purpose chat assistant is the better choice

The obvious alternative is a general chat assistant with a long prompt, and the difference is not quality of prose. It is state. A chat session holds everything in context and forgets it when the window closes; get-job.skill writes 岗位调研.md, 改后简历.docx and a 面试准备/ directory to disk, and each stage reads the previous stage's file. The repository's examples/ folder shows the intended shape of that output: examples/文科生投运营/ contains a research file, a rewritten resume file, and an interview-preparation folder with numbered files from 00-总览.md through 99-面后复盘题库.md.

The second difference is the gate. A chat assistant will happily produce a confident resume bullet with nothing behind it. This project's README describes a delivery check that blocks on unsupported claims and marks them low confidence instead. If you only need a one-off rewrite and you are willing to judge the output yourself, a chat assistant is less setup. If you want the research, the rewrite and the round-by-round preparation to stay consistent with each other across a multi-week application, the file-based pipeline is the reason to install this. The scripts/README.md documents how to run the rendering script directly if you only want the docx output.

Licence, maintenance and what upgrading costs you

The licence is CC BY-NC-ND 4.0, which the README summarises as free to share with attribution but no commercial use and no distribution of modified versions. For an individual candidate running the skill locally this is unproblematic. For a university careers office, a bootcamp or a recruiting product that wants to embed the playbooks, the non-commercial and no-derivatives terms are the obstacle, and the README directs commercial licensing enquiries to the author. That is a description of the licence text, not legal advice; read LICENSE yourself before building anything on top of it.

On maintenance, the last push to the repository was on 2026-08-31. There are no retrieved releases, so upgrades happen by pulling the default branch rather than by pinning a version. That matters because the behaviour you depend on is spread across markdown playbooks and Python scripts rather than a versioned package: a change to references/quality-gates.md can alter what gets blocked at delivery without any version number changing. If you clone manually, record the commit you installed and diff the references/ directory before pulling. If you install via npx skills add, re-running the command is the update path, and you should re-read the playbook files afterwards. The repository also carries a GitHub Actions workflow at .github/workflows/quality.yml, so the project runs its own checks on push, but the README does not describe a release or changelog process.

Editorial conclusion

Adopt get-job.skill if you are a student or career changer whose experience is real but reads as unrelated to the target role, and you want the research, resume rewrite and interview preparation produced as files you can review. Do not adopt it if you need a commercially licensed component, if you intend to modify and redistribute the playbooks, or if you want a tool that fabricates experience, because the README states the opposite. Before relying on it, open references/quality-gates.md and references/regression-evals.md and confirm that the P0/P1/P2 thresholds and the degradation rules match what your own application can tolerate.

Frequently asked questions

Official sources

1. agentenatalie/get-job.skill on GitHub
2. Issues
3. README