ASu-skills: a Chinese job-search skill pack for Codex, Claude Code and Qoder

What ASu-skills is for, and who it is not for

The README describes ASu-skills as a Chinese job-search workflow plugin. It is not a resume editor with a web UI. It is a set of nine callable entry points that run inside an AI coding agent, each aimed at one stage of a campus recruiting pipeline. The table in the README lists them: /contributor for finding and shipping open source contributions, /evidence-recap for turning AI coding conversations into a verifiable evidence chain, /project-guide for generating source-reading courses and interview talking points, /great-resume for repositioning existing experience, /make-resume for producing an editable HTML resume and a PDF export, /job-match for comparing a job description against real experience, /job-apply for filling in application forms through a connected browser, /interview for question prediction and weak-spot drilling, and /offer for tracking applications, assessments, interviews, offers and rejections.

The intended user is a Chinese-speaking candidate, most likely a student or early-career developer, who already uses an agent host and wants the job-search work to happen where the code work happens. The README's first-use table routes by situation rather than by tool: if you lack verifiable project experience, start with /contributor; if you have AI coding records but no clear narrative, start with /evidence-recap; if you have a repository and need learning material, start with /project-guide.

If your job search is in English, or you want a hosted service that stores your applications, this is the wrong shape. The entry points, the generated documents and the HR opening-line feature all target Chinese-language hiring, and the README's own examples are in Chinese.

The registry is the architecture: one skills directory, many hosts

The repository is organized around a single shared skill set rather than per-platform forks. The README states that all entry points share the same skills/, assets/ and references/ directories, and that the entry list lives in skills.registry.json at the repository root as the single source of truth. That file is generated and reconciled by npm run sync:skills, and CI validates it with a --check flag.

That design decision has a visible consequence. The top-level directories include .claude-plugin/, .codex-plugin/, .opencode-plugin/, .qoder-plugin/, .trae-plugin/ and .workbuddy-plugin/, so each host gets a thin adapter while the substance stays in one place. Adding a tenth skill means editing the registry and regenerating, not editing six plugin manifests by hand. The package.json confirms the packaging story: name asu-skills, version 0.4.0, type module, with a dsh bundle pointing at cordis.patch.yml, and keywords that include dsh-plugin and deepseek-harness.

The Node engine constraint is explicit: ^22.19.0 or >=24.0.0. There is also a Python side. The scripts block includes validate:skills, which runs python3 scripts/validate_skills.py, and test:python, which runs python3 -m unittest discover -s tests -t . -v. So a full local check needs both a recent Node runtime and a Python 3 interpreter. That is a real setup cost for a plugin that a candidate might otherwise expect to install and forget.

Installing ASu-skills on Claude Code, Codex, TraeWork or Qoder

The README gives a per-platform install table rather than a single command. On Claude Code the two commands are a marketplace add followed by a plugin install, and the README says to run /reload-plugins afterwards. The names below are copied from that table.

/plugin marketplace add Hisn00w/ASu-skills
/plugin install asu-skills@asu
/reload-plugins

On Codex the flow is conversational: send the GitHub repository link to Codex and explain that you want to install it, then start a new conversation and pick the entry point from the / menu. TraeWork expects the repository copied to a versioned plugin directory, after which you restart the application.

~/.trae-cn/plugins/<publisher>/asu-skills/<version>/

OpenCode and WorkBuddy are described as lightweight bridge entries, with separate installation guides in .opencode-plugin/installation-guide.md and .workbuddy-plugin/install.md. Qoder has its own .qoder-plugin/install.md, and the README notes that uninstalling can be done with qodercli plugin uninstall or by deleting the plugin directory, depending on how it was installed.

For a first real use, the README's routing table is the starting point, not a sample prompt. If you have a job description and a resume and want to know whether to apply, the table sends you to /job-match. If you have a repository and need interview material, it sends you to /project-guide. A representative invocation from the README's own examples looks like this.

/great-resume

目标岗位：AI 应用工程师
请根据我下面的实习和项目经历，给出稳妥版和进取版定位，改写简历要点，并生成一段发给 HR 的开场白。

What you should see, per the README, is a conservative and an aggressive positioning, rewritten resume bullets, and a Chinese opening message for HR. If you supply too little information, the documented behavior is to produce a usable draft and mark gaps with the placeholder 【待补】 rather than inventing a title, company, stack or metric.

The fact-boundary rule is the most interesting design choice

Several of the skills are built around refusing to inflate. The README states that when information is insufficient, /great-resume gives a draft and marks missing pieces with 【待补】, and that it will not fabricate a title, company, technology stack or figure. /contributor has a similar rule expressed as a status ladder: an unmerged pull request is recorded as submitted or in collaboration, and the stronger claim of a completed contribution is only allowed once the GitHub page shows it merged and the project explicitly states the scope of adoption.

/evidence-recap goes further and is the clearest expression of the project's stance. It takes AI coding conversations and delivery records and organizes them into a nine-part evidence chain, separating individual actions, delivery stages and outcome evidence. The point is to distinguish what you actually did from what an assistant produced on your behalf, which is exactly the ambiguity that makes AI-assisted project bullets hard to defend in an interview.

This is a constraint, not a feature list. It means the tool will sometimes hand back a document with visible holes instead of a polished paragraph. Candidates who want the polished paragraph will find that annoying. The README's own position is that the holes are the point, and /interview is built to exploit them: it predicts questions, drills contract-style follow-ups, reviews evidence and re-practices weak areas. A resume claim with no evidence chain behind it is what that skill is designed to expose.

Where ASu-skills breaks down

The strongest limitation is stated in the README itself, in a section titled 事实边界, and it is structural rather than a bug. Every skill depends on material you supply and on a host agent that can read files and, for /job-apply, drive a browser. There is no standalone binary, no web interface and no server component documented. If your host does not support plugin installation, the bridge directories for OpenCode and WorkBuddy are the only documented path, and those are described as lightweight bridges with their own separate guides.

The second limitation is /job-apply. The README says it connects to a browser, fills in a job application automatically, and then stops before submission for you to check. That stop is deliberate, and it is also the boundary of what the tool will do. Anyone expecting end-to-end automated applications will be disappointed, and the README does not document what happens when a form uses a widget the automation cannot see. The README is silent on rollback for a partially filled form.

The third is language and market. The product description in package.json describes Chinese job-search workflows, and the README's generated artifacts have Chinese filenames such as 导学-{简称}.md and 面经-{简称}.md. Nothing here suggests an English output mode.

Finally, version 0.4.0 with no releases retrieved means you are installing from the default branch. The last push was on 2026-09-15, so the code is current, but there is no tagged release to pin against.

How it differs from a resume builder or a plain prompt library

The obvious comparison is a hosted resume builder, the kind that gives you a form, a template gallery and a PDF button. The difference is the starting point. A resume builder assumes you already know what to write and need formatting. ASu-skills assumes the opposite problem: the material exists as scattered AI coding sessions, half-finished repositories and internship fragments, and the hard part is deciding what is defensible. That is why /evidence-recap and /contributor sit at the front of the documented workflow and /make-resume sits near the end.

The second comparison is a folder of saved prompts. A prompt library is text you paste. ASu-skills is packaged as a plugin with a registry file, per-host adapters, a validation script and a test suite. The trade-off is real in both directions: the prompt library works anywhere and does nothing automatically, while ASu-skills can read your repository and write files like tutorial.md and practice.md, but only inside a supported host and only with Node 22.19 or newer plus Python 3 available for the validation scripts.

A third difference is the interview loop. /project-guide generates a source-reading path and STAR talking points with a source-code evidence index, and /interview then attacks those claims. Most resume tooling stops at the document. This one keeps going until the document has to survive questioning.

Licence, maintenance and upgrade cost

The repository is MIT licensed, and LICENSE sits at the top level. The package.json also declares "license": "MIT". For a skills pack that generates documents you may send to employers, MIT is permissive and imposes no obligation on the output. That is a practical point, not a legal one: the generated resume and the HR opening line are yours to use, and the licence text is the authority on what the licence actually permits.

Maintenance cost comes from two places. First, the host adapters. Six plugin directories exist, and a host that changes its plugin manifest format will require a change to its adapter. The README states that skills.registry.json is the single source of truth and that npm run sync:skills regenerates the catalog with CI checking it, which concentrates that maintenance in one script rather than six.

Second, the runtime floor. Node ^22.19.0 or >=24.0.0 is a narrow band at the time of writing, and the Python validation scripts add a second interpreter. Upgrading Node across that boundary is the upgrade event to watch for.

If you fork it, run the documented checks before trusting a checkout: npm run sync:skills:check for the catalog, npm run validate:skills for the skill definitions, and npm test for the Node test suite. The package version is 0.4.0, and no releases were retrieved, so there is no changelog to read between versions. That absence is the main upgrade risk: you will be diffing commits rather than reading release notes.

Editorial conclusion

Adopt ASu-skills if you are running a Chinese-language job search and already work inside Codex, Claude Code, TraeWork or Qoder, because the nine entry points assume that environment and its slash-command menu. Skip it if you need English-language output, a standalone web app, or a tool that works without an agent host. Before committing, verify two things yourself: that your host's plugin install path matches the table in the README, and that npm run sync:skills:check passes on the checkout you intend to use, since skills.registry.json is the single source of truth for the entry list.

Frequently asked questions

Official sources

1. Hisn00w/ASu-skills on GitHub
2. Issues
3. License: MIT
4. README