Offer Toolkit Skill: a six-part Claude Code job-hunt pack that outputs HTML

What Offer Toolkit Skill is for, and who it is not for

The repository is a pack of six sub-skills under one top-level router, aimed at people running an agent that supports skills. The README lays out a pipeline: discover opportunities, decode a job description, tailor a resume, prepare behavioral answers, compare offers, negotiate. Each stage is a separate folder, and the README states that each sub-folder works on its own. That matters because most people will want one or two of them, not six.

The audience is narrow in a useful way. Someone who already has a resume and a shortlist of roles gets value from job-description-skill and offer-compare-skill. Someone starting cold gets more from job-hunt-skill, which the README says takes a resume, a target direction, a seed JD or existing job links as input and returns a searchable HTML job list. The README is explicit that it discovers and analyzes but never applies for you, so anyone hoping to automate applications is looking at the wrong project.

There is a second audience the README implies rather than names: bilingual candidates. The bq-skill description says stories are saved in English and Chinese, and the repository ships README.zh.md plus example/cn/ and example/en/ alongside the English README.

The router in SKILL.md and the six sub-skills behind it

The mechanism is file-based, not service-based. The top-level SKILL.md acts as a router over six directories: job-hunt-skill, job-description-skill, resume-skill, bq-skill, offer-compare-skill and salary-negotiation-skill. There is also a job-search-skill/ entry in the repository root listing, which does not appear as a numbered step in the README pipeline. That is a discrepancy worth checking before you assume the pipeline is exactly six folders deep.

Each sub-skill turns a conversation into a document. The README describes job-description-skill as taking a JD plus your resume and returning an HTML report covering fit, gaps, likely interview questions, salary reasonableness and a six-week action plan. offer-compare-skill compares two or more offers across four-year total compensation, growth, AI exposure, company and team risk, promotion, lifestyle, resume value and future mobility, and produces an HTML Offer Decision Report with an explicit recommendation. salary-negotiation-skill diagnoses Base, RSU, Sign-on and Bonus leverage, then writes phone and email scripts, simulates recruiter pushback, and defines sign, walk and stop-lines in an HTML Negotiation Playbook.

The shared constraint is stated as four rules, of which the first is the load-bearing one: never fabricate, with every experience, number and title coming from what the user actually said. The second rule, asking only what changes the answer, is the one most skill packs get wrong. It is a prompt-level constraint, not an enforced one, so its reliability depends on the model running it.

Installing Offer Toolkit Skill and running the resume sub-skill

The README gives no install command. It presents the repository as an agent skill pack and shows badges for Claude Code, Codex, OpenCode, OpenClaw and Hermes, without documenting a per-host installation path. The practical route is to clone the repository and point your agent at the folder, then let SKILL.md route. Confirm the layout first:

git clone https://github.com/yanliudesign/offer-toolkit-skill.git
cd offer-toolkit-skill
ls

You should see SKILL.md, the six skill directories, docs/, example/ and the two README files. If your agent supports a skills directory, the same clone is what you place there; the repository does not name that path, so check your host's own documentation rather than guessing.

A first real use is the resume sub-skill, because it has the most concrete output. The README says it polishes an existing resume, pulls one in from LinkedIn, or builds one from scratch by chatting, and that content then flows into eleven print-ready templates. Each render yields two files: one that prints straight to PDF, and one you can click in the browser to edit.

cd resume-skill
ls

What you should see is the template set the README names: Classic-ATS, Ledger, Tech Compact, Modern Sidebar, Pillar, Elegant Serif, Atelier, Timeline, Swiss, Executive and Color-block. The README does not document a CLI flag for choosing a template, so selection happens through the conversation with the agent. The README also does not document what happens to a render if you dislike the result, so keep your source resume outside the repository.

Where Offer Toolkit Skill breaks down

The intake cadence is the first failure mode. The README's second shared rule says to ask only what changes the answer and to follow each child skill's stated one-at-a-time or batched intake cadence, and warns against turning the conversation into a sprawling questionnaire. That rule is about the agent's behavior, not a validator. A model that ignores it will interrogate you for twenty turns before producing anything, and nothing in the repository structure prevents that.

The second issue is the honesty rule. Never fabricate is a good default, and the README pairs it with an instruction to verify numbers. But a resume builder that only rearranges what you told it cannot fix a resume whose underlying achievements are thin. If your experience is weak, the polished output is a better-formatted weak resume. That is a real boundary, and the README does not claim otherwise.

The third is scope creep in the opposite direction. job-hunt-skill searches public sources and separates verified facts from inference, per the README. Public job sources go stale, and the README does not describe a refresh or staleness policy for the generated HTML list. Treat the list as a snapshot taken when you ran it.

Finally, the repository has no releases. There is no version to pin and no changelog to read, so an upgrade means pulling the branch and re-reading the diff yourself.

Offer Toolkit Skill compared with generic prompting or a resume web app

The obvious alternative is prompting a general assistant directly, without a skill pack. The difference is artifact format and intake structure. A chat reply about a JD is prose you scroll through; job-description-skill returns an HTML report with defined sections (fit, gaps, interview predictions, compensation, action plan). Whether that is better depends on whether you revisit the output later. For a one-off question, plain prompting is faster and has no setup cost.

The closer alternative is a hosted resume builder. Those give you a template gallery and export, which is the same promise as resume-skill's eleven templates. The difference in approach: a hosted builder owns your content and its template set is fixed by the vendor, while this pack keeps content as files in your repository and generates templates from a conversation. That also means you maintain it. No vendor will migrate your data or fix a broken render.

Within the pack itself, the pieces overlap least with generic tools. bq-skill does not hand over canned answers; the README says it digs out your real past experiences, shapes each with STAR or CAR, tags them, and saves them in English and Chinese so one story answers different questions later. That reuse model is the part a single chat session does not give you. salary-negotiation-skill is the other end: it produces scripts and explicit stop-lines, which is a decision artifact rather than research.

Licence, maintenance and what an upgrade costs you

The licence is MIT, per the repository's LICENSE file and the README badge. MIT permits commercial use, modification and redistribution with the copyright notice and licence text retained. That is a permissive arrangement with few obligations, but it also means no warranty and no support commitment. Nothing here is legal advice; read LICENSE yourself if you plan to redistribute the templates.

The README carries a disclaimer badge reading Not professional advice, which is worth taking at face value for the negotiation and offer-comparison sub-skills. Those produce recommendations about compensation decisions, and the project explicitly frames itself as a tool rather than an advisor.

On maintenance: the last push to the default branch was on 2026-09-15, and the repository is not archived. There are no releases, so there is no versioned upgrade path. Upgrading means pulling main and diffing SKILL.md and the sub-skill folders against your copy, because the behavior you depend on lives in those instruction files, not in a compiled artifact. If you have customized any sub-skill, expect to re-apply those edits on every pull.

Editorial conclusion

Adopt it if you already run Claude Code or a compatible agent and want job-hunt output you can open in a browser rather than paste out of a chat window; the resume templates and the story bank are the two pieces with the least overlap with generic prompting. Skip it if you want an agent that applies to jobs, or if you need a versioned release you can pin, because the repository lists no releases and the README does not document rollback. Before relying on it, open resume-skill/ and count the templates yourself, and check whether your agent actually reads SKILL.md as a router or only loads one sub-folder.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. README
4. yanliudesign/offer-toolkit-skill on GitHub