agentleFS
qbekaqbeka/jev-job-search/.claude/skills/setup/SKILL.md

setup

Skill368 starsChanged 5 days agoMITView on GitHub

First-run setup, start to finish. Use when the user says /setup, "set this up", "get me started", "help me use this", hands you this repository for the first time, or when data/profile.json does not exist. Installs what is missing, builds the profile from the resume and a short interview, has the user add their own JEV key (TypeSafe or OpenRouter), finds jobs, rehearses, and stops for a "go" before anything is sent.

  • Reads credentials

192 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use add-atomic-chat-tool, add-dashboard and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • MicrosoftSkill

    add-atomic-chat-tool

    Add Atomic Chat MCP server so the container agent can call local models served by the Atomic Chat desktop app via its OpenAI-compatible API.

    Also in 8 projects that use this

  • MicrosoftSkill

    add-dashboard

    Add a monitoring dashboard to NanoClaw. Installs @nanoco/nanoclaw-dashboard and a pusher that sends periodic JSON snapshots.

    Also in 8 projects that use this

  • +10 more from microsoft

The file

Show its rules and text

We split this skill into the rules it gives an agent

25 rules. 0 are common, 9 are distinctive, and 16 are very specific to this project.

  • 0

    Common

    Most of the 169 edited versions we read keep these.

  • 9

    Distinctive

    This author's own choices. Worth a look.

  • 16

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Run every command from the repo root.

    line 13
  2. Very specific:

    If node_modules is missing, run npm install.

    line 20
  3. Very specific:

    Run npx jev doctor. Fix what it names, in order:

    line 21
  4. Very specific:

    Look in documents/ first. Besides resume.pdf, a user may have put past material there for you to read: old resumes, a LinkedIn profile …

    line 31
  5. Distinctive:

    Ask for the resume as a PDF.

    line 42
  6. Distinctive:

    Read the PDF. It is your first source for the profile.

    line 45
  7. Distinctive:

    Ask the user what the resume does not state.

    line 54
  8. Distinctive:

    Ask in small groups, at most four questions at a time.

    line 57
  9. Very specific:

    Write summary (three or four sentences: who they are, what they want, where, when) and facts (one line per claim, every number exact, …

    line 59
  10. Very specific:

    Write answers, the standing answers, from part G of the questions.

    line 62
  11. Very specific:

    Run doctor again. A schema error names the field to fix.

    line 64
  12. Distinctive:

    Never invent or round a fact.

    line 68
  13. Distinctive:

    Never guess a demographic answer.

    line 69
  14. Very specific:

    Write data/bank.json in the shape of data/bank.example.json: one starting draft per common question, from the user's own facts.

    line 77
  15. Distinctive:

    Ask them to create a key and add a few dollars of credit, at either https://console.typesafe.ai/keys (TypeSafe, who make JEV) or …

    line 92
  16. Very specific:

    Run cp .env.example .env, then open -e .env so the file opens in a text editor.

    line 96
  17. Very specific:

    Ask them to paste a TypeSafe key after TYPESAFE_API_KEY=, or an OpenRouter key after OPENROUTER_API_KEY=, save, and tell you when it is …

    line 98
  18. Very specific:

    Run npx jev doctor --online.

    line 101
  19. Distinctive:

    Never ask the user to paste the key into the chat, and never print it.

    line 104
  20. Very specific:

    Run npx jev discover. It takes about a minute.

    line 110
  21. Very specific:

    If the user wants other locations or kinds of roles than the ranking prefers, the settings are in src/config.ts (LOCATION_MULTIPLIER, …

    line 115
  22. Very specific:

    Run npx jev apply --dry --count 3.

    line 121
  23. Distinctive:

    Do not list commands. Say only:

    line 149
  24. Very specific:

    If many of the jobs they would want are on Workday (npx jev status lists "workday" among the skip reasons), say that Workday employers each …

    line 159
  25. Very specific:

    If they asked for it to run by itself, /daily.

    line 162

“Common” is measured: kept by at least 6 in 10 of the 169 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/setup: from a fresh clone to the first applications

The user should need to do four things: give you their resume, answer your questions, paste one key into a file, and say "go". You do everything else.

Run every command from the repo root. After each step run npx jev doctor --json. It lists what is in place and names the next thing to do, so you never have to guess where you are. If the user stops halfway and comes back, start from what doctor says.

1. Install

  1. If node_modules is missing, run npm install.
  2. Run npx jev doctor. Fix what it names, in order:
    • Node.js older than 22, Google Chrome missing, or Claude Code missing: tell the user what to install and wait. You cannot install these. (You are running inside Claude Code, so it is present; a user who prefers a Claude API key can add ANTHROPIC_API_KEY to .env and the commands use the API instead.)
    • Everything else is covered by the steps below.

2. The resume, and anything else in documents/

Look in documents/ first. Besides resume.pdf, a user may have put past material there for you to read: old resumes, a LinkedIn profile export as PDF, transcripts, past applications, a portfolio page saved as PDF. Read every PDF and text file there (the Read tool opens PDFs). Each is a source for the profile in step 3: a bullet, a date, a project, a skill that the current resume leaves out. Nothing personal in documents/ is ever committed; the folder is git-ignored apart from its README. Re-running /setup after more files land there is safe: it adds, it does not overwrite what the user confirmed.

  1. Ask for the resume as a PDF. The user can drag the file into the chat or give its path.
  2. Copy it to documents/resume.pdf.
  3. Read the PDF. It is your first source for the profile.

3. The profile

Build data/profile.json in the shape of data/profile.example.json. The schema is src/profile/schema.ts.

  1. Fill in everything the resume states: name, contact details, links, education, each job and project with its bullets, skills.
  2. Ask the user what the resume does not state. The full list is in questions.md. Follow it top to bottom, and skip a question only when the resume or the user has already answered it. Ask in small groups, at most four questions at a time. Use AskUserQuestion when a question has a fixed set of answers.
  3. Write summary (three or four sentences: who they are, what they want, where, when) and facts (one line per claim, every number exact, copied from the resume or said by the user).
  4. Write answers, the standing answers, from part G of the questions.
  5. Show the user a short summary of the profile and correct what they flag.
  6. Run doctor again. A schema error names the field to fix.

Rules for this step:

  • Never invent or round a fact. If you are not sure, ask.
  • Never guess a demographic answer. Every one of them can be declined.
  • resume.path is the absolute path of documents/resume.pdf.
  • These files are git-ignored. Never commit them and never paste their contents into a file that is tracked.

4. Their voice and their drafts

  1. Copy data/voice.md to data/voice.local.md.
  2. Write data/bank.json in the shape of data/bank.example.json: one starting draft per common question, from the user's own facts. Keep the text in braces where an answer depends on the company.
  3. Draft one sample answer to "Why do you want to work here?" for a made-up company and ask whether it sounds like them. Adjust data/voice.local.md and the drafts until it does. One or two rounds is enough.

5. The key

JEV is the only part of this tool the user is billed for. Tell them the numbers before they pay: about 6 cents each time the tool finds jobs, and about 1 cent for every 10 applications. Five dollars of credit lasts for months.

  1. Ask them to create a key and add a few dollars of credit, at either https://console.typesafe.ai/keys (TypeSafe, who make JEV) or https://openrouter.ai/keys (OpenRouter, who resell it). The model and the price are the same.
  2. Run cp .env.example .env, then open -e .env so the file opens in a text editor.
  3. Ask them to paste a TypeSafe key after TYPESAFE_API_KEY=, or an OpenRouter key after OPENROUTER_API_KEY=, save, and tell you when it is done.
  4. Run npx jev doctor --online. It makes one tiny JEV call and one tiny Claude call to prove both work.

Never ask the user to paste the key into the chat, and never print it. If they paste it into the chat anyway, tell them to delete that key where they made it (console.typesafe.ai/keys or openrouter.ai/keys) and create a new one, because a key in a chat is exposed.

6. Find jobs

Run npx jev discover. It takes about a minute. Show the top 15 as a short table: score, company, role, location, date posted. Say how many jobs are queued and name the main reasons jobs were skipped (npx jev status).

If the user wants other locations or kinds of roles than the ranking prefers, the settings are in src/config.ts (LOCATION_MULTIPLIER, FIT_WEIGHTS, DISCOVER).

7. Rehearse

  1. Run npx jev apply --dry --count 3. A Chrome window opens, three forms are filled, nothing is sent, and the tabs close.
  2. Show the user every answer, grouped by form. Point out anything the tool held and why.
  3. For each answer the user corrects, change the profile: a fact, or a standing answer in answers. Then rehearse again. Do not move on until the user says the answers are right.

Tell the user one thing before the first rehearsal: some job boards save a form as it is typed, so a rehearsal there leaves an unsent draft on that board. It is not an application and the employer is not told.

8. One application, ready for them to read

Setup ends with a real application filled and waiting, not with a list of commands. Run npx jev apply --count 1 on the best job in the queue that is on Greenhouse, Lever or Ashby. Without --submit the form is filled, checked and left open in the tool's Chrome window.

Tell them to look at that window. Read the answers back to them, name anything the tool held, and ask: "Send this one?" Only when they say yes, run npx jev submit <id>. Then show the row in npx jev log.

That is the whole loop. Say so in one sentence, then stop asking questions.

9. What is next, in three lines

Do not list commands. Say only:

  • /apply sends more: they say how many.
  • /report is the dashboard: what was sent, what needs them, and the switch for a daily run.
  • /status says where things stand at any time.

Mention these once, and only when they apply. Do not set any of them up during setup:

  • If many of the jobs they would want are on Workday (npx jev status lists "workday" among the skip reasons), say that Workday employers each want an account and /accounts handles that when they are ready.
  • If they asked for it to run by itself, /daily.

Two things the tool will not do, so they are not surprised: it does not pass a "prove you are human" check, and it does not sign in anywhere unless they set that up. A form that needs them stays open in the tool's window; /resume picks it up.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.