Industry Resume Toolkit counts its own commands three different ways

Five slash commands in the header, four files in the install step

The command count changes three times inside one file. The centred header advertises 5 one-click commands, and a later section is headed 5 Slash Command, listing /诊断, /改简历, /创建简历, /JD匹配 and /加规则. The manual Windows instructions still say to copy the 4 .md files that sit in commands/ into ~/.claude/commands/, and the v0.1 summary line counts four slash commands as well, naming diagnose, rewrite, JD match and add rule. The fifth command arrived with v0.2 and the manual step was never renumbered, so anyone who installs by hand on Windows receives four of the five. The same paragraph then asks for a full restart of the Claude desktop app and a verification prompt asking the assistant to read ~/.claude/skills/industry-resume-toolkit/SKILL.md and explain the skill, with the expected answer being the toolkit positioning plus a list of all five commands. A verification step that expects five cannot fail visibly on a four-file install.

The Mac clone line still carries the [your-username] placeholder

The only installer the project ships is install.sh, and Shell is the language the repository is written in. Its three documented commands are these:

git clone https://github.com/[your-username]/industry-resume-toolkit.git
cd industry-resume-toolkit
sh install.sh

The clone URL has never been filled in. The square bracket placeholder is still sitting in it, so the first command cannot succeed as printed unless the reader substitutes an account of their own, which is also the only reason the placeholder makes sense: the intended route is fork first, then clone the fork. That assumption is nowhere stated. Two smaller details travel with it. The script is invoked with sh rather than bash, so anything it does has to stay inside a POSIX shell, and the whole flow assumes a Mac, since the Windows route is described as manual file copying instead of script execution. A docs/ directory exists in the repository and INSTALL.md is referenced as the long version of these instructions, so the three lines above are the short form of something written down elsewhere.

v0.2 is described as a seven step flow and as a six step flow

The v0.2 announcement says the new /创建简历 command ships 4 sets of user specific questions plus a 7 step complete flow. The roadmap near the bottom of the same file describes that release with a 6 step complete flow and then lists the steps: type triage, diagnosis, experience mining, restructuring, drafting, bilingual output. Six names, six steps, seven steps, with nothing in the file marking which count is current. The rest of v0.2 is more precise. The command can take an uploaded resume, described as a LinkedIn export, an older version or an overseas one, reads it as a starting point, skips the questions already answered and asks only for the missing parts, and defaults to Chinese and English output at the same time. The audience is spelled out as people at a career turning point: fresh graduates with no experience, career changers, returnees from overseas, and anyone restarting after a gap. The two commands are separated by intent, with /创建简历 rebuilding the narrative and /改简历 only polishing wording, quantifying and adding keywords, and the file says to prefer the rebuilding one even when a resume already exists.

The header carries five links and none of them has text

Under the centred title sits a block of five markdown links whose visible text is empty: two pointing at LICENSE, one at https://docs.claude.com, one at #, and one at README.md. Rendered, that is a row of unlabelled targets, one of which is a dead anchor and one of which leaves the project for a third party product documentation site. Nothing in that block points at the docs/ directory or at CONTRIBUTING.md, even though both sit in the repository root listing alongside install.sh, README.md, README_EN.md, commands/ and skill/. The navigation line directly below does something the header cannot: it links the English README, an install anchor, a quick start anchor, the job pattern section and the roadmap. So the file carries two navigation systems, one with labels and no dead ends, one without labels and two. The same LICENSE target appearing twice and the total absence of any visible label are what a reader gets; whether that was the intent is not stated.

The index shows no license while the file names CC BY-NC-ND 4.0

The repository index records no license for this project, while the README states that it is released under Creative Commons Attribution NonCommercial NoDerivatives 4.0 and spells out three rules: free to learn, use, repost as is and share; attribution required, naming the author with a Xiaohongshu account number and a link back to the repository; commercial use forbidden, covering selling, paid courses and any monetised scenario; and modified versions forbidden from being published or redistributed. A LICENSE file sits at the root of the tree, so the licence is physically present, but the listing does not say which text it holds, and the two sources disagree about whether one is declared at all. The no derivatives clause also sits next to a Contributing section that invites pull requests and says forks are welcome as long as the author merges them into the original, so contributions travel inward rather than as redistributed copies. The same file then offers one on one personalised service by direct message for readers whose situation is unusual, which is exactly the commercial case the licence names.

The name carries two meanings inside a single description line

The one line description field is bilingual and mixes scripts: a Chinese phrase about a resume methodology for job seekers in China, followed by the English rendering, Industry resume methodology for Chinese job seekers. In the Chinese half, the English word industry stands in for the word for a job sector, while in the English half it reads as the category the project belongs to. The repository name makes that collision permanent, since the account shangsitongshizaitiantang is the romanised form of the same handle the licence section names and then asks readers to find on Xiaohongshu. The tagline under the title is a slang phrase about what suits Chinese readers, which sets a jokey tone for a project otherwise built out of very formal comparison tables. An English README sits in the root listing and the header links to it, so a non Chinese reader is not stranded, though the command list and the pattern tables are only reachable in Chinese.

Four employer types times twelve job families, with no precedence rule

The methodology lives as two directories of pattern files. One holds four employer type files under patterns/企业性质/, covering large internet and technology firms, state owned and centrally administered enterprises, banks and financial institutions, and foreign companies, each with a stated style: quantified, matched to the job description, agile and data driven for the first, steady, formal and long term for the second, rigorous and compliance minded with precise numbers for the third, and international, STAR framed, strongly worded and concise for the fourth. The other holds twelve job family files under patterns/岗位/, covering operations, product management, data analysis, marketing, finance, audit, HR, sales, development, legal, strategy and consulting, and design. The file insists that these two dimensions stack to hit the target scenario, which gives forty eight combinations. It never says which one wins when they conflict, and the styles pull in opposite directions. Choosing between them is left to the reader, and the diagnose command that could have encoded the choice only records the target as stated input.

Input .pdf is turned into Word before anything else happens

The input and output contract is the most concrete part of the project. Accepted inputs are .docx, .pdf, .md and plain text, and a PDF is converted into Word automatically so nobody has to convert it by hand. Output defaults to .docx, after which the tool asks whether you want another format such as Markdown, PDF, an English version or a bilingual one. The interesting branch is what happens to the file you handed over. If the input was .docx, the changes are written onto the original file and your original layout is preserved. If the input was a .pdf or plain text, the result is a .docx with generic professional formatting instead. A resume pasted in as plain text therefore comes back typeset from scratch, while a Word file comes back with whatever margins, fonts and header layout it already had. If you would rather use your own template, the documented route is to drop the template file into inputs/ and tell the assistant to use that one.

Editorial conclusion

This is a methodology package rather than a program: what ships is prompt text, pattern files and one shell script, so its value depends on how closely the pattern tables match the employer you are aiming at. Use it if you are targeting Chinese employers and can read the Chinese README, and start from the diagnose command rather than the rewrite one. Before relying on it, count the command files your own copy actually received, read the LICENSE file in the repository root instead of trusting the licence summary, and decide for yourself which of the two v0.2 step counts you are getting. Skip it if you need an academic CV, which the project excludes by name.

Frequently asked questions

Official sources

1. Issues
2. README
3. shangsitongshizaitiantang/industry-resume-toolkit on GitHub