Navigation and fit use different thresholds
questions.py separates careers navigation (0.55), job detection (0.5), relevance (0.7) and saving (0.7). A single universal confidence cutoff would change this workflow.
Inspect the supporting source ↗REPOSITORY ANALYSIS
JevScout
A demo job-search Skill for coding Agents that browses company careers pages in Chrome, uses Jev to screen AI and software-engineering roles, and saves the results.
E2 · source reviewedPublic material or source evidence was reviewed. A fixed-version source review requires checking a pinned revision; this is not a reproduction, performance endorsement, or safety guarantee.
- Stars
- 33
- Forks
- 7
- License
- —
- Last commit
- 2026-09-18
What Jev does here
Jev uses Choice to classify pages and Noul to assess careers links, role relevance, filter controls and candidate fit; local thresholds select navigation, job details and saved results.
Connects careers navigation, role screening and detail matching in one CLI workflow, with local JSON and Markdown reports.
Demo MVP reviewed at a fixed commit; no browser or live API run. Scanning has count and text limits. Mock scores and latency are simulated, while mock mode can still operate real Chrome. The workflow is not guaranteed read-only. Site coverage, matching accuracy and performance were not validated; no open-source license was confirmed.
What task does this project fit?
Turn a candidate profile into a reviewed shortlist from company careers sites.
Material analysed：
From input to action
- Classify the current page and score visible links.
- Filter likely software/AI jobs, then read details.
- Save a posting only when its overall-fit result passes the local policy.
Implementation details that matter
Mock inference does not isolate the browser
The README describes a Chrome-based demo; mock scores avoid model calls but the runner can still navigate real pages. Testing must control the browser surface separately.
Inspect the supporting source ↗When to choose it
Useful for inspecting a staged browser-screening pipeline. Treat the saved list as candidates to review, not as verified job availability or an automated application decision.
How to inspect it yourself
Follow policy.py from page_kind to candidate de-duplication and decide_detail. Check whether a failed navigation is distinguishable from a genuinely irrelevant company.
Limits of this review
- No Chrome session or paid inference was executed here.
- Link-count and text limits can omit relevant jobs; matching quality was not measured.