Repository navigation
Releases: inamiqbal2/jobkit
Release list
jobkit v1.0
v1.0: the structure release. One settings schema (userconfig.SCHEMA drives defaults, validation, template comments and docs); nothing reads a file at import (claims_loader; import cycle gone); one eligibility module shared by the corpus arm, screener and critic; daily.py split into stage functions with a tested dedupe/mirror core; one exit-status table and --help on every CLI. Behaviour verified unchanged with a live A/B run. Fixes the ' rejected me' crash from v0.9. Details: system/docs/DEVELOPER.md 'What changed in v1.0'.
jobkit v0.9
v0.9: second code audit. Safer fetches (http/https only), honest run summaries (failures are sentences, errors never counted from job titles), the learning loop fed automatically on 'I applied', faster daily runs (concurrent arms, one TF-IDF fit, SmartRecruiters fan-out), fill-rule WARN for thin profiles, Windows path fixes. Details: system/docs/DEVELOPER.md 'What changed in v0.9'.
jobkit — 2026-10-20
Download jobkit.zip below, unzip it, open the jobkit folder in the Claude desktop app (Code tab), and type set me up. Already using jobkit? Copy your My Job Search folder into the new one. (Or use the one-click page: https://inamiqbal2.github.io/jobkit)
What's new (program v0.8): a code audit, and its fixes
An independent review of the engine found places where the honesty audit could be bypassed. All are closed, each with a test that reproduces the original gap and must now fail.
- The audit reads every part of a document. Text in tables, headers, footers and text boxes is now checked; before, a claim placed there was invisible to it.
- The audit fails closed. A document's kind (resume, CV, letter) comes from a stamp the builder writes, never from its filename; an unknown file is checked as a resume, the strictest path.
- Headings and labels are recognised by structure, not by capital letters, so a claim cannot hide as a 'heading' or next to your email.
- Expired postings are set aside (they were slipping through to the judges).
- Your auto-skip setting now works; it was fixed at 40 before.
- Employer sites that fail to answer are reported instead of looking searched.
- Places you said you would never move to are enforced by the program, not only by Claude.
- Many smaller fixes: consistent posting IDs in the ledger, no duplicate ledger lines on a same-day re-run, one shared definition of 'fresh / nearby / thin posting', instructions that match what the audit allows, dead code removed. 36 tests, 18 new.
The engine status reads "Engine: original (v0.8)".
jobkit — 2026-10-19
Download jobkit.zip below, unzip it, open the jobkit folder in the Claude desktop app (Code tab), and type set me up. Use this zip rather than the green Code button: it includes the program history that modify: and revert to original need. Already using jobkit? Copy your My Job Search folder into the new one.
What's new (program v0.7): you decide, jobkit follows
- Location is your call. Setup asks how you think about it (stay local; local first but would move for a strong fit, and how strong; location doesn't matter; or not sure yet), plus remote, places you'd never move, and other countries you'd work in. It tells you what each answer changes before you pick.
- Position types are your call. Internships, co-ops, new-graduate programs, fellowships, research or unpaid work, postdocs, contract, hourly: only what you leave unticked is filtered out.
- You choose what the judges score. Skills and experience always; location, career direction, pay and remote only if you say so.
- Steadier judges. A fixed scoring scale and a quoted line from each posting, checked against the text, so a judge can't skim. In a test on 40 real postings, the lighter Sonnet judges matched the top model on every build decision.
- Light mode for Pro plans (offered at setup): a measured run took about 10 minutes and about $1.40 of usage.
- README: measured costs per run and an honest Limitations section.
- Fixes: defaults inherited from the original author's own search were removed (a recruiter exclusion, a new-grad block, a weekly build quota); foreign cities such as Budapest are now caught.
The engine status reads "Engine: original (v0.7)".
jobkit — 2026-10-16
Download jobkit.zip below, unzip it, open the jobkit folder in the Claude desktop app (Code tab), and type set me up. Use this zip rather than the green Code button: it includes the program history that modify: and revert to original need. Already using jobkit? Your My Job Search folder carries over: copy it into the new folder.
What's new (program v0.6)
- Reads far more postings. Morgan Stanley and other Eightfold sites, and shared Workday sites (Wells Fargo, Fidelity), now return the real job text. In a test run for a finance professional, unreadable postings went from 452 to 0 and postings reaching the judges' shortlist from 113 to 151.
- Cheaper judging. The second blind judge now reads only the borderline postings (18 of 40 in that test).
- Safer decisions. A strong posting can't be skipped without quoting the requirement it fails; a recent high score is remembered for 14 days.
- Your words, the posting's words. Wordings you confirm mean the same thing for you can be used on the resume and still pass the audit. Never-claims stay forbidden.
- Smaller fixes: company-age blurbs no longer read as experience requirements, US-first multi-site postings are kept, 'High Temp' is not a temp job, 'Sr.' titles match, Nodi job boards, LinkedIn/Adzuna backoff.
The engine status now reads "Engine: original (v0.6)".
jobkit — 2026-10-12
Download jobkit.zip below, unzip it, open the jobkit folder in the Claude desktop app (Code tab), and type set me up. Full instructions are in the README.
Use this zip rather than the green Code button: it includes the program history that modify: and revert to original need.
What's in this release: the jobkit engine as first shipped (status shows "Engine: original (v0.5)"), plus the new README.