Skip to content

Getting started

This page takes you from an empty folder to your first checked application. You need Claude Code and a folder for your application data. No developer tooling is required for the steps below; the export to PDF is the one optional step that needs extra software, see Export.

1. Install the plugin

Pick one of three ways.

From the marketplace, per project. Add the marketplace and enable the plugin in the .claude/settings.json of the folder you work in:

{
  "extraKnownMarketplaces": {
    "claude-job-search": {
      "source": {
        "source": "github",
        "repo": "nolte/claude-job-search"
      }
    }
  },
  "enabledPlugins": {
    "claude-job-search@claude-job-search": true
  }
}

Interactively. Inside Claude Code run /plugin, add the marketplace nolte/claude-job-search, and install claude-job-search from it.

From a local checkout. Clone the repository and start Claude Code with claude --plugin-dir <path-to-the-checkout>.

Afterwards the skills are available as /claude-job-search:<skill>, for example /claude-job-search:job-application-solution.

2. Create the configuration

The plugin reads exactly one file at a fixed place: job-application/konfiguration.md in the folder where you run Claude Code. Everything else is found through it. The file has a short frontmatter and may hold free notes below it.

Minimal, one file as Datengrundlage. All application data lives in job-application/; the Datengrundlage is one Markdown file:

---
grundlage: job-application/grundlage/lebenslauf-chronologisch.md
---

# Konfiguration

Directory mode with an Obsidian vault outside the data root. The Datengrundlage is a folder of notes, one per station, plus an index note that carries the maintenance history:

---
wurzel: /home/me/bewerbungen
grundlage: /home/me/vault/lebenslauf
grundlage_index: /home/me/vault/metadata/Lebenslauf.md
---

# Konfiguration

Applications in `/home/me/bewerbungen`, the Datengrundlage in my Obsidian vault.

The keys:

Key Needed Meaning
wurzel optional Root of all application data. Default: job-application/ next to the configuration. May point anywhere on your disk, including a folder inside an Obsidian vault.
grundlage optional Your Datengrundlage: one Markdown file, or a directory of notes. Default: <wurzel>/grundlage/lebenslauf-chronologisch.md.
grundlage_index only for a directory The index note of the directory, with the maintenance history.

Relative paths are resolved against the folder that contains job-application/konfiguration.md. If you start a skill without this file, it asks for the paths once and writes the file for you.

Not inside the plugin repository

The data root must not lie inside a checkout of the plugin itself. If you develop the plugin and apply with it, set wurzel to a folder outside the checkout.

3. Capture your data

Put your raw material into <wurzel>/rohdaten/: old CVs, certificates, a LinkedIn export, notes. If you have connected the optional LinkedIn server, LinkedIn access shows how to bring in your own profile. Then run

/claude-job-search:job-application-profile-capture

The skill reads the raw material, asks where it is unsure, and writes the Datengrundlage (your stations, education, skills, languages, certificates, each with its source) and the derived profil.md. Anything it cannot verify is marked [ERGÄNZEN: …] and stays visible until you resolve it. Later changes go through /claude-job-search:job-application-cv-maintain, which asks only about findings and records every change in the maintenance history.

4. Optional: say what you are looking for

/claude-job-search:job-application-traumjob-interview

A seven-phase interview that records your wishes, priorities and vetoes as wunschprofil.md. Nothing is written before you confirm the summary. With a confirmed Wunschprofil, /claude-job-search:job-application-fit-check compares a posting or a company with it before you spend time on documents, and a veto stops the plan in step 5.

5. The first application

/claude-job-search:job-application-solution

Paste the posting text or give its link. A link alone is enough: the skill cleans it of tracking parameters, loads the page through the optional open-webSearch server and shows you the extracted posting for confirmation before anything is saved. If the page cannot be read (removed, log-in wall, bot check, needs JavaScript, no server connected), you are asked to paste the text. LinkedIn links are never loaded. The skill

  1. checks that your Datengrundlage and profile are current,
  2. creates the application folder and analyses the posting (which requirements you meet, with evidence, which you meet partly, which are not covered),
  3. classifies the counterpart (employer directly, recruiter, agency, platform, referral), see Counterparts,
  4. shows you a plan: which documents in which format, which stations and bullets it will use, which gaps remain,
  5. waits for your approval,
  6. creates the documents in order (Lebenslauf, then Anschreiben, then the e-mail if you apply by e-mail), and
  7. runs the quality gate.

The plan is a gate, not a formality: no document is written before you say yes. If the plan is wrong, say what to change, and the skill shows a new plan.

6. What is in the application folder

Each application gets one folder under the data root, named after company and position, for example musterfirma-senior-backend/:

File Content Written by
vorgang.md Company, position, reference number, channel, status, dates, the approved plan, the history the skills for their steps; you for sending and everything after
notizen.md Your free notes: impressions, questions for the interview, names from calls only you; no skill changes it
stellenanzeige.md The posting, verbatim posting analysis
analyse.md Requirement match with evidence, the counterpart, the companies involved posting analysis
passung.md Fit against your Wunschprofil, only when you ran the fit check fit check
lebenslauf.md The Lebenslauf for this application CV skill
anschreiben.md The Anschreiben cover letter skill
motivationsschreiben.md Only when asked for motivation letter skill
email.md Only when you apply by e-mail e-mail skill
pruefbericht.md Which stations were selected and why; the result of the quality gate CV skill, quality gate
export/ PDF, DOCX, HTML, only when you ask for an export export skill

Next to the application folders sit grundlage/ (if your Datengrundlage lives there), profil.md, wunschprofil.md, rohdaten/, firmen/ (one file per company with stable facts and your notes), and the CV variants. A variant is a Lebenslauf with its own focus that you reuse across applications, for example lebenslauf-basis/ or lebenslauf-teamleitung/. Each variant folder holds its lebenslauf.md, its pruefbericht.md, frozen versions in versionen/ and their exports in export/.

7. Status and overview

vorgang.md carries a status field with a fixed set of values. The skills set the first values; from sending onwards you set them yourself:

Set by the skills Set by you
erfasst (posting captured), analysiert, geplant (plan approved), erstellt (documents created), geprüft (quality gate passed) or in Arbeit (blockers open) versendet, rückmeldung, gespräch, angebot, verhandlung, zusage, absage, abgelehnt (you declined), zurückgezogen, ohne Antwort

Every status change gets one line in the Verlauf section of vorgang.md with the date, so the history stays readable. When you send the application, record the date, the channel and the attachments there too; wiedervorlage holds the date of your next follow-up. Free text goes to notizen.md.

Overview across all applications. If your data root is an Obsidian vault (or a folder in one), a Bases file lists every application by status without any extra plugin. Save this as bewerbungen.base in the vault:

filters:
  and:
    - 'file.name == "vorgang"'

properties:
  firma:
    displayName: Firma
  stelle:
    displayName: Stelle
  status:
    displayName: Status
  wiedervorlage:
    displayName: Wiedervorlage

views:
  - type: table
    name: Offen
    filters:
      and:
        - 'status != "zusage"'
        - 'status != "absage"'
        - 'status != "abgelehnt"'
        - 'status != "zurückgezogen"'
        - 'status != "ohne Antwort"'
    groupBy:
      property: status
      direction: ASC
    order:
      - firma
      - stelle
      - status
      - wiedervorlage
      - file.folder

  - type: table
    name: Abgeschlossen
    filters:
      or:
        - 'status == "zusage"'
        - 'status == "absage"'
        - 'status == "abgelehnt"'
        - 'status == "zurückgezogen"'
        - 'status == "ohne Antwort"'
    order:
      - firma
      - stelle
      - status
      - loeschen_ab
      - file.folder

Open the file in Obsidian: the first view groups the open applications by status, the second lists the closed ones with their deletion date. Without Obsidian, grep -r "^status:" */vorgang.md in the data root gives the same list on the command line.

8. Export on request

Documents stay Markdown until you ask for an export. Run /claude-job-search:job-application-export or task export for a PDF, DOCX, HTML or Markdown copy. A text PDF is the default, also for application portals; produce DOCX only when a portal explicitly asks for it. Details, requirements and the container that bundles everything are on the Export page.

9. Retention and deletion

A closed application keeps its folder. For every final status other than zusage, the data is meant to be deleted six months after closure: set loeschen_ab in vorgang.md to that date when you close the application. No skill deletes anything; the deletion is your act, and the Bases view above shows you which dates have passed.

For developers

The rules behind this page: spec/job-application/lokale-struktur/en.md (terms, layout, configuration keys, lifecycle and status values, retention), spec/job-application/konventionen/en.md (fact integrity, data protection, output formats), spec/job-application/lebenslauf-versionen/en.md (variants and frozen versions), spec/job-application/traumjob-interview/en.md (Wunschprofil and fit check), spec/job-application/kontaktklassifikation/en.md (counterparts).