Job Search DE

Personal / Career Intelligence

Find roles in Germany, review the evidence for fit, and track applications in one workbench.

In brief

An agent skill that collects job listings, checks source availability, compares requirements with candidate evidence, and exports a workbench for reviewing roles and tracking applications.

An open-source skill for coding agents, with a standalone browser workbench.

Job discovery, fit assessment, and application tracking often happen in separate tools. This project brings them into one workflow that an existing coding agent can run.

The skill supports job searches across professions in Germany. The public workbench illustrates the workflow with technology roles and sample scores.

A useful shortlist needs more than matching keywords: it needs source links, freshness checks, explicit candidate evidence, and a way to track the next action. Missing information should remain visible rather than becoming an invented match.

What it includes

  1. 01Collects openings from company career feeds, with supplementary job-board sources for broader coverage.
  2. 02Checks URL responses and posting metadata to flag expired, inaccessible, or uncertain listings.
  3. 03Uses criteria-based triage followed by requirement-by-requirement profile evidence, including partial matches and unknowns.
  4. 04Exports a standalone workbench with search, sorting, application tracking, notes, and four visual themes.
  5. 05Stores profile and configuration files in the workspace; workbench status and notes persist in browser storage.

How it works

  1. 01
    Build the profile

    extract candidate evidence and record search preferences and constraints.

  2. 02
    Discover roles

    collect company career feeds and supplementary sources into a common format.

  3. 03
    Check listings

    inspect URL responses and available posting dates; retain uncertainty when evidence is missing.

  4. 04
    Evaluate fit

    triage against criteria, then compare individual requirements with profile evidence.

  5. 05
    Review and act

    export the workbench and report, review the shortlist, and track applications.

Explore the pipeline illustration
SYS // SPEC 014Job Search DE / workflow
Illustrative workflow
From discovery to application

Find roles. Review fit. Track what follows.

Collect openings, check their sources, and compare requirements with profile evidence before adding them to your application shortlist.

01 / Discover

Company career feeds

Normalize company career feeds and supplementary discovery sources.

Greenhouse · Example
Senior AI Systems EngineerBerlin · €85k-€105k (Example)
Ashby · Example
ML Platform & Agent InfraMunich · Hybrid (Example)
SOURCE CHECKAt collection
02 / Check

Check listing freshness

Inspect URL responses and posting dates; flag missing or uncertain evidence.

  • URL respondsReachable page; hiring status still needs review
  • Posting metadataRead JobPosting structured data when available
  • Freshness windowCompare dates with the configured threshold
  • Unavailable listingFlag broken or expired source links
EXAMPLE: SOURCE CHECKED
03 / Evaluate

Review fit evidence

Triage first, then record matches, transferable skills, gaps, and unknowns.

CRITERIA TRIAGERole · Location · Constraints
PROFILE GROUNDING
Example fit score94/100
Rubric scoreReview evidence
EXAMPLE PROFILE EXCERPT

“Built production LLM pipelines using LangGraph and PyTorch.”

04 / Track

Application workbench

Filter the shortlist, inspect details, and record your next application step.

THEMEEditorialDarkIndustrialQuartz
TO APPLY3
Senior AI Systems94/100 · Berlin
APPLIED2
ML Platform Lead89/100 · Munich
INTERVIEW1
Staff SystemsRound 2 Tech
Draft an application opening
2 Evaluation stagesSource-linked listingsBrowser-saved tracking4 Workbench themes

Illustrative roles, scores, and profile evidence. This diagram does not fetch live jobs.

Build notes

Note 01

Keep listing provenance visible

source verification
Context
Duplicate and outdated listings make a shortlist difficult to trust.
Approach
Prefer company career feeds, retain source links, and check available posting metadata. Supplementary sources broaden discovery.
What it enables
Listings can be traced back to their source. A successful URL check does not establish that an employer is actively hiring.
Note 02

Separate triage from evidence review

evidence scoring
Context
One overall match score can hide missing requirements and unsupported assumptions.
Approach
Use an initial criteria-based pass, then record direct matches, partial matches, gaps, and unknowns for the shortlist.
What it enables
The score has an explanation that a candidate can review. It is a rubric score, not a statistically calibrated probability.
Note 03

Export a browser workbench

standalone workbench
Context
A terminal report is awkward for comparing roles and maintaining application status.
Approach
Generate an HTML workbench with embedded data, four themes, and browser storage for status and notes.
What it enables
The exported shortlist can be reviewed without rerunning the agent. New discovery and scoring still require the agent workflow.

System capabilities

2
Evaluation stages

DefinitionCriteria triage followed by detailed evidence scoring; scores are prioritization aids, not hiring probabilities.

4
Workbench themes

DefinitionEditorial Craft, Dark Velocity, Industrial Precision, and Spatial Quartz.

Compatibility

  • Coverage depends on configured sources, public endpoint availability, and changing response formats. Freshness checks cannot identify every ghost listing or repost.
  • No separate LLM API key is needed by the skill, but the host agent may have usage costs and may process profile content through its model provider.
  • Workbench notes and status are stored in the current browser and do not automatically sync across devices. Profile changes require saving or exporting the configuration files.
  • The public demo contains illustrative roles, dates, and scores; its links lead to company career pages and do not verify the displayed openings. Web fonts may require a network connection.
Next project15Material ModelAcademic Research / Scientific Machine Learning

Project image

Interactive demo