Resume graph

You turn one person's resume into one data file (profile.json) and one web page (index.html) that draws it. The page is theirs to put on their own site.

ROOT is the folder that contains both scripts/profile.mjs and view/. Look in this skill's own folder first; if they are not there, it is the folder two levels up (the repository root, where the sibling folders read, company, skills, check and publish are one-step shortcuts into this same skill). Find it once and use it for every path below. Run every command from the folder that contains resume-graph/, the working folder.

Read PROFILE-FORMAT.md (next to this file) before writing any data.

Three rules that hold in every step

1. Nothing about the person is invented. Every sentence about what they did comes from their resume. You may shorten and reword. You may not add a fact, a number, a tool or a result the resume does not state. Numbers are copied exactly.
2. Facts about organisations come from a page you opened, and the address goes in org.source. If you cannot find or cannot tell which organisation it is, leave org out. Never describe a company from memory.
3. Private details stay out: phone number, street address, date and place of birth, marital status, nationality, religion, parents' names, licence and registration numbers, references. Email and other contact links go in only if the person says yes.

Run it in one go

Do all six steps without stopping to ask. The person asked for a graph, and a finished page they can look at is easier to judge than questions about a draft. Stop only if there is no resume to read. Things you would have asked go in the final message as "say if you want this changed".

Which file: the one the person gave. If they gave none, look in the current folder for a PDF, Word or text file whose name contains cv, resume, lebenslauf or curriculum. One match: use it, even if the name on it is not the name of the person you are talking to (people make graphs for others); say whose resume you used in the final message. Several that look like versions of the same person's resume: use them all. None: ask for the file. Look only in the current folder; do not go through the person's other folders for a resume.

Contact links, without asking: keep the public profile links printed on the resume (LinkedIn, GitHub, a personal site, a portfolio). Leave out the email address and everything else in rule 3.

Work in a folder named resume-graph/ in the current directory unless told otherwise. Each step has its own file under steps/; read it when you get there.

1. Read (steps/read.md). Get the plain text of every resume, save it as resume-graph/resume.txt, and write a first resume-graph/profile.json with the profile fields and one bare node per entry. No org, no sections, no skills yet.
2. Profile each entry (steps/company.md), in parallel. For every node that names an organisation or has more than a line of detail, start a company-profiler agent, all in one message, each given ROOT, the node, and the resume lines that belong to it. With more than twelve such nodes, give each agent two or three. Merge what they return into profile.json.
   • No company-profiler agent type: start ordinary subagents on a mid-size model with the same input and tell them to follow steps/company.md.
   • No subagents at all (Claude on the web): do the nodes yourself, one after another.
   • No web access: skip the org blocks and say so at the end.
3. Logos. node ROOT/scripts/profile.mjs logos resume-graph/profile.json. It takes each organisation's icon from its own website (through org.url) and stores it in the node, so the mark in the graph and the panel show the logo. Entries it finds nothing for keep the plain shape. Without network access, skip it.
4. Skills (steps/skills.md). With all nodes final, choose the skills and write the uses sentences.
5. Check (steps/check.md). Run the script and fix every error. Then start a checker agent with ROOT and the two file paths for the read-through (no such agent: an ordinary subagent that has not seen your drafts; no subagents: read it through yourself). Fix what it finds and rerun the script.
6. Publish (steps/publish.md). Build resume-graph/index.html and open it in the person's browser. If they then ask to put it online, the last part of that file says how.

What to say at the end

Keep it short and plain:

• where the page and the data file are, and that the page is open in their browser;
• how many entries and skills the graph has;
• what was left out and why: private details, the email address, skills with no evidence in the resume, organisations you could not identify;
• anything you decided for them that they may want changed (a date two resume versions disagreed on, an unusual heading you kept as its own group), each as one line they can answer;
• any warning the check still prints.

Then offer: "Tell me what to change and I'll rebuild it." Do not describe the graph as impressive; the person will look at it.