Repository navigation
jobdork 0.16.0
Download
| For | Download |
|---|---|
| Mac with Apple Silicon (M1 or later) | jobdork-0.16.0-macos-arm64.dmg |
| Mac with an Intel processor | jobdork-0.16.0-macos-x86_64.dmg |
| Windows 10 or 11, 64-bit | jobdork-0.16.0-windows-x64-setup.exe |
| Docker (a NAS or a server) | docker pull jessengolab/jobdork:0.16.0 |
| Checksums of the installers | SHA256SUMS.txt |
Mac: open the .dmg and drag jobdork into Applications. Windows: run the setup; it installs for you alone, no administrator needed. Not sure which Mac you have? Apple menu → About This Mac: "Chip" is Apple Silicon, "Processor … Intel" is Intel. Linux, or from source: the ways to run it. New to jobdork? The user guide shows every page of the app.
The first time you open it
The installers are not signed yet, so each system warns once:
- macOS says it cannot check jobdork for malicious software. Click Done,
open System Settings → Privacy & Security, scroll down to the line about
jobdork and click Open Anyway, then Open Anyway again. - Windows shows "Windows protected your PC". Click More info, then Run anyway.
Verify Build Package
An installer: compare its SHA-256 with SHA256SUMS.txt, or with the GitHub CLI:
gh attestation verify jobdork-0.16.0-macos-arm64.dmg --repo jessn-dev/jobDorkEach installer's signed attestation is here too, as <installer>.sigstore.json, to check it without asking GitHub:
gh attestation verify jobdork-0.16.0-macos-arm64.dmg --bundle jobdork-0.16.0-macos-arm64.dmg.sigstore.json --repo jessn-dev/jobDorkThe image jessengolab/jobdork:0.16.0 is exactly:
jessengolab/jobdork@sha256:9c22765efbb38b02855923be99fa549d96091fbfb871410f053b320ba58a3679
Built, scanned, checked and signed by this repository's workflow after approval.
Verify it before you run it, then pull it by that digest:
cosign verify jessengolab/jobdork@sha256:9c22765efbb38b02855923be99fa549d96091fbfb871410f053b320ba58a3679 \
--certificate-identity-regexp '^https://github.com/jessn-dev/jobDork/\.github/workflows/docker\.yml@' \
--certificate-oidc-issuer https://token.actions.githubusercontent.comThe installers are unsigned: how to open them. Privacy: PRIVACY.md.
What changed
20 changes in this release, newest first (click to read)
Changed — 2026-10-30 — version 0.16.0
0.16.0 carries the 19 entries below, everything since 0.15.4 (28 October).
For you:
- Your resume, tailored and checked. A tailored resume in Indeed's
layout and length, as PDF and Word; how a tracking system reads your
resume file, scored with what to fix; and how its lines are written
(weak openers, "I", long bullets, no numbers). - A new Resume page: your resume and each tailored one as cards beside
a preview that fills the window, with each job post's other documents as
tabs: cover letter, recruiter message, interview prep, review, ATS match. - A job post's ATS match, by script: keywords, required skills, title,
sections and dates, out of 100, with I have this for a skill you have. - Interview prep and a recruiter message, written by the AI and checked
like a cover letter. jobdork never sends the message; you do. - Job posts with the chosen post beside the list, as job boards show
them, on a wide window. - My applications: every job you applied to, by stage, with each
status dated, the next step and its date (past ones flagged on the
Dashboard), the contact, what you sent, and notes. - Your coding agent as the AI, with no key: Claude Code, Codex, Gemini
CLI, Cursor, GitHub Copilot or Claude Desktop does the writing through
jobdork's tools, and jobdork checks and keeps it. With guardrails: what it
adds to your record waits for your OK, job adverts are data and never
instructions, your contact details stay with you, limits an hour, and
every change it makes on the Dashboard. - Fixed: the Windows app's AI settings (no time zone database); any
reader or checker moves or goes, the first too; Settings no longer makes
the Mac ask for Documents and Downloads; the resume's full path; keys from
your computer's environment named; AI providers by their company names.
Updating from 0.15.4 uses the update window, as before. On first start the
database moves from schema 13 to 16 by itself (confirmed skills, the dated
history of statuses, applications and their notes, and what a coding agent
suggested); an older jobdork will not open it after. The config keeps
working as it is. On Windows, if 0.15.4's Settings, Updates does not
answer (the time zone fix above), install 0.16.0 from the release page over
it: your data stays. Windows also gains jobdork-agent.exe beside the app,
which a coding agent starts.
The documentation is brought up to date for all of it: the user guide (its
home page, overview, getting started and glossary, and a new My
applications page), the README, PRIVACY.md and SECURITY.md (what a coding
agent reads, and what a job post cannot make it do), the terminal commands
(jobdork prep, message, mcp, agent), the config reference
(provider: agent), ARCHITECTURE.md, the desktop page, and a decision in
DECISIONS.md on how a coding agent is used.
Fixed — 2026-10-30 — The Windows app's AI settings, which needed a time zone
On Windows, jobdork 0.15.4 could not load the part that keeps free Gemini
and Groq keys within their limits (ai/limits.py): it reads the Pacific
time zone, where Gemini's day turns over, and Windows has no time zone
database of its own; the app carried none. The dashboard's AI settings,
which the page asks for all the time, failed with it. Found by the new MCP
check in the Windows build (#105).
tzdatais a dependency on Windows (pyproject.toml,
requirements-desktop.lock) and travels with the Windows app
(packaging/jobdork.spec).- Without a database anyway, the module uses Pacific standard time all
year instead of failing (the day turns an hour early in summer). - The app's self-test looks a time zone up, so a build without one stops
before it becomes an installer.
Added — 2026-10-29 — Guardrails for a coding agent
A coding agent (web/agent.py) is held to the rules a model is, and more:
- What it adds to your record waits for your OK. A project, education
entry or skill it suggests is kept apart (suggested_by, schema 16) and
counts for nothing (fit, checks, tailoring, the record notes) until you
Approve it at the top of the Resume page; Turn down removes it.
The menu counts what waits. An agent cannot take back a skill you
confirmed, nor undo one; yours approves its suggestion of the same skill.
At most 25 wait at once. Without this, an agent could add a skill and
then cite it, and every check would pass. - Job board text is data, not instructions. Hidden characters are
removed from every title, company, place and advert the agent reads, and
from its briefs; every answer carrying them says they are from strangers;
the agent's instructions say never to follow them, run a command or open a
link because of them. - Contact details stay with you, as for a hosted model: email, phone
and profile links are removed from the resume, the record and the briefs
the agent reads (llm.redact);get_resumegives the file's name, not
its place on your computer. - Limits: 30 documents and 60 job post scores an hour, counted from what
agents saved, so they hold across sessions and the command line; answers
up to 64 KB. - A record of what it changes. Every change is a run on the Dashboard
from the coding agent, refused ones too, with what it did; its answers
are counted with the AI calls under its name; statuses and notes it set
say by Coding agent · in the Application tab (by, schema 16).
Guide: the AI page's guardrails, the Resume page's suggestions, with a
picture; demo data has one suggested skill. Tests: 7 more in
tests/test_agent.py.
Added — 2026-10-29 — A coding agent as the AI, with no key
A third way to give jobdork its AI, beside Ollama and a keyed model: your
own coding agent (Claude Code, Codex, Gemini CLI, Cursor, GitHub Copilot,
Claude Desktop), after Resume Matcher's agent mode (srbhr/resume-matcher). jobdork
never calls the agent; the agent calls jobdork's tools, and jobdork asks no
model of its own.
- The tools (
web/agent.py): overview, list and read job posts, the
resume and record, a post's ATS match;get_brief, exactly what jobdork
would send a model for a tailored resume, cover letter, interview prep,
recruiter message, resume review or a post's AI score;save_document
andsave_verdict, which put the agent's answer through every check a
model's gets (gates, PDF and Word, the record) and return the checks so it
can fix and save again; applications (status, next step, contact, notes);
and projects, education and confirmed skills. - Two doors to the same tools: an MCP server on stdio (
web/mcp.py,
JSON-RPC 2.0 in the standard library;jobdork mcp, or the installed app
with--mcp), with its workflow sent as the server's instructions; and
jobdork agent TOOL JSON(the app:--agent) for agents that only run
commands. The build checks the built app answers MCP (packaging/build.py).
On Windows the agent startsjobdork-agent.exe, the same program built
with a console besidejobdork.exe(packaging/jobdork.spec): the
window's program has no standard streams for an agent to talk on, which
that check found. - How it fits (
ai/llm.py):agentis a provider whose calls are
answered from a queue (the agent's own answer, never passed on). What
jobdork asks for itself (a button, judging after a scan, a page read)
goes to the next reader; with the agent the only reader, the call keeps
its brief and raisesAgentBrief. One reader line at most may be an
agent, with other readers before or after it; never a checker
(core/config.py). - The hallucination check stays a second model's (
ai/guard.py): the
agent never checks its own writing. A checker, or a reader the config had
before, checks; with none, the check is not run and is a check to read. - The AI page offers Coding agent for Reader 1, with the line for
Claude Code, Codex, Gemini CLI (command and settings.json), Claude
Desktop, Cursor, VS Code and the command line,
filled in for this computer, each with Copy. With the agent as Reader
1, a job post's AI buttons show what to ask the agent, with Copy, and
Use instead when another reader is listed. With the agent
the only reader, the dashboard's model jobs are refused with what to ask,
and scans skip AI judging.
Guide: the AI page's new Coding agent section, with a picture. Tests:
tests/test_agent.py (14).
Added — 2026-10-29 — My applications
A new page, My applications (menu, under Job posts), lists every job
you applied to, newest change first: one marked applied, submitted,
interviewing or offer, or rejected or withdrawn after (and one closed or
skipped after applying). Chips narrow it to a stage (All, Applied,
Interviewing, Offer, Closed) with counts; each row has the job, its status,
the day you applied and its next step with the date, red once past. A row
opens beside the list, as on Job posts (in the window when narrower), on a
new Application tab that any job post you applied to now has:
- Stages: each status with its day. jobdork now dates every status
change (status_history); a post applied to before this version starts
from the status it had, dated when it was set. - Next step and its date, Contact (name, email, link) and
Documents sent (which of the post's tailored resumes and cover
letters), kept by Save application (applications). - Notes: dated, filed under the status at the time, newest first, with
Delete (asks first) (application_notes).
Like a status, all of it belongs to the job, so copies of one job share it,
and it goes with the job post when that is deleted. A follow-up past its
date is counted in red beside My applications in the menu and named on the
Dashboard, with Open My applications. Schema 15. API: /api/applications,
/api/application, /api/application/note; the job post's own
(/api/role) carries its application. Guide: a new My applications page,
and the Dashboard and Statuses pages; demo data with four applications.
Changed — 2026-10-29 — Job posts: the list with the chosen post beside it
On a window 1180 pixels wide or more (the app's own size included), Job
posts shows the list on the left and the post you chose on the right, as
job boards do: click a card's title or any blank part of it and the whole
post (the same tabs as its window) opens beside the list, scrolling on its
own, with the list still there to choose the next. The chosen card is
outlined; its Open button, status list and note step aside, as the post has
them. The post's header gains Apply (opens the posting and marks it
applied, as the card's does) and a Note field under the status buttons.
Close empties the space. A narrower window, a list with nothing to open
(posts dropped by your filters) and other pages (the Resume page's Job
post) still open a post in a window over the page. Guide: Job posts and
the job post window, pictures retaken.
Changed — 2026-10-29 — The Resume page's preview uses the whole right side
On a wide window the Resume page kept to 1060 pixels and its PDF to 68% of
the window's height, leaving a band of blank page. The page now widens to
1600 pixels, the preview takes the larger share of it, and the PDF is as tall
as the window less the preview's tabs. The PDF opens fitted to the
preview's width without the viewer's page list beside it, and a long
Stored at path wraps instead of running under the preview.
Added — 2026-10-29 — A job post's documents beside its tailored resume
On the Resume page, a tailored resume's preview now has a tab for each of
its job post's other documents, as Resume Matcher's builder has: Cover
letter, Recruiter message (with Copy message), Interview prep,
Resume review when there is one, and the post's ATS match, then the
resume's Checks. Everything for one application is in one place. A
document not written yet says so, with Open AI tools (the job post's
window, where it is written). Choosing another tailored card keeps the tab,
to compare two cover letters, say. /api/resumes gives each card the newest
of each document for its job post (docs); the job post's window and the
preview draw a document the same way (docHtml). The resume's Job post,
PDF and Word moved beside its title, so the tabs have the row.
Guide: Resume page, with a new picture of the interview prep tab.
Added — 2026-10-29 — Interview prep and a recruiter message, by the AI
Two more documents from a job post's window, written by the AI-page model
and checked like a cover letter (ai/writer.py; jobdork prep UID and
jobdork message UID; /api/prep, /api/message).
Interview prep (AI): why you fit (three to five points), questions
about your resume (five to eight, each with what it tests and answer points
from your resume), follow-ups on your work (three to six), gaps to prepare
(three to five things the advert wants that your resume does not show, how
to answer honestly) and talking points (five to eight). What it says about
you gets the hallucination and figures checks; a new gaps not claimed
check is READ when a gap's preparation reads as if you have it. It uses the
Resume page's projects and education too. Saved as interview-prep-ai.md.
Recruiter message (AI): 70 to 100 words for LinkedIn or an email, its
first sentence on something from the advert (never "I'm reaching out"),
one on your strongest match, a figure only from your resume, and a small
ask ("Worth a quick chat?"); no "excited" or "passionate", no greeting, no
placeholders. A word count check joins the letter's checks (and its
not-empty check takes 150 characters, not 400). Copy message copies it:
jobdork never sends it. Saved as recruiter-message-ai.md.
Both follow Resume Matcher's interview prep and outreach message
(srbhr/resume-matcher, Apache-2.0), rewritten with jobdork's rules.
Interview prep reads your whole record beside the resume: projects,
education and the skills you confirmed with I have this (store_notes).
The job window's heading is now The main buttons (there are more than four);
the Resume page's link to it follows, and tests/test_guide.py now fails on
any guide link to a part of a page that is not there (the docs build only
warned).
Added — 2026-10-29 — A job post's ATS match, and "I have this"
Each job post's window has a new ATS match tab: your resume file
against its advert the way an applicant tracking system searches it, out of
100, by script (search/ats.py, /api/ats). Keywords 40% (the advert's
skills your resume names), required skills 20% (those under its
Requirements or Qualifications), job title 15% (at the top of your resume,
in its summary, or some of its words), sections 15% (contact details and
the summary, experience, education and skills headings) and dates in one
format 10%; a part that cannot be judged is left out. The parts and
weights follow Resume Matcher's ATS score (srbhr/resume-matcher,
rewritten); where it asks a model for the advert's keywords, jobdork uses
the skills list the Fit score uses, so it stays offline and free.
Below, the advert's skills in three groups: on your resume; safe to
add, in your projects, education or confirmed skills but not the file (a
tailored resume can name them); and missing, each with I have this.
A confirmed skill is kept in a new table (schema 14, confirmed_skills,
/api/skills/confirm) and joins the record the Resume page's projects and
education already make (resume_doc.store_notes): Fit scores count it,
job posts are re-scored, and a tailored resume may name it. take back
removes one. What to do says what would raise the score.
Added — 2026-10-29 — How a resume's lines are written, checked
A new script check (writing/style.py, no AI) reads a resume's bullets as
a recruiter does: a bullet that starts by saying you were near the work
("Responsible for", "Helped with", "Assisted", "Worked on"; hard), first
person ("I", "my"), over 35 words, three or more bullets of one job
starting with the same word, and no figure. Each line scores 100, less 40
for the hard rule, 10 for each soft one and 5 for no figure, and the score
is their mean (Resume Matcher's writing rules and weights,
srbhr/resume-matcher, rewritten; its long-dash and AI-phrase rules are left
to the checks jobdork already has). It shows under your resume's
Checks as How its lines are written, each rule once with its count
and three of its lines; from a PDF, whose bullet marks are usually drawn,
the Experience section's lines are read as bullets. Tailored resumes and
CVs get it as the resume writing check (READ only for the hard rule),
and the tailoring instructions now forbid those openers and the first
person and ask for varied verbs.
Changed — 2026-10-29 — The Resume page: your resumes as cards, beside a preview
The Resume page is laid out after Resume Matcher's dashboard and builder
(srbhr/resume-matcher), in jobdork's own colours, fonts and dark mode.
Across the top, a card for your resume (its readability score as a bar),
one for each tailored resume and CV, newest first (the job's initials,
title and employer, the date, Ready or Read 2 for checks to look at,
and its readability), and Tailor for a job, which opens Job posts.
Below, the details (the upload, Projects, Education, Resume review and
Cover letters) sit beside a preview of the chosen card: Resume shows
it as it will print, Checks its checks, and a tailored one has Job
post, PDF and Word. On a narrow window the preview comes first.
New /api/resumes lists the cards. The guide demo gained two tailored
resumes, and the guide two pictures.
Added — 2026-10-29 — How a tracking system reads your resume
The Resume page now checks the uploaded file the way an applicant tracking
system's text reader does, on this computer and with no AI, and scores it
out of 100 (search/readable.py, /api/resume/check, and in the upload's
answer). It looks for a text layer (not a scan), letters that come out as
letters, pages that are pictures, two columns, Word tables and text boxes,
contact details only in a page header, repeated header or footer lines,
pictures, more than two pages, an email and phone near the top, the
Experience, Education and Skills headings, and dates written one way. Each
problem shown says what to do. Every tailored resume gets the same check on
its own PDF, as the new readable by tracking systems gate.
The rules, thresholds and penalties (30, 12 and 4 points, half for a
warning) follow Resume Matcher's parse check (srbhr/resume-matcher,
Apache-2.0), rewritten here with pypdf rather than pdfminer and no code
copied. Tried on two dozen real resumes, three things were added to it: a
sidebar of skills of even length is still a column (Resume Matcher's rule
took it for dates at the right margin, so dates must now be dates);
letter-spaced text ("J O H N") fails as unreadable; and a heading split by a
large first letter ("E xperience") is named as such, as are plural
headings and ones wrapped onto two lines ("Education and"), which were
missed.
Fixed — 2026-10-28 — Any reader or checker can be moved or removed, the first too
On the AI page and setup's AI step, only reader 2 onward (and checker 2
onward) had Earlier and Remove: the first line could not be taken
out or moved down, so swapping two meant retyping both. Now, once there are
two or more, every line has Remove, every line but the first has
Earlier, and every line but the last has the new Later. Removing the
first makes the next one first. A single line is still turned off with
Off (readers) or None (checkers).
Fixed — 2026-10-28 — Settings no longer makes the Mac ask for Documents and Downloads
Opening Settings in the Mac app made macOS ask to let jobdork into the
Downloads folder and then the Documents folder. Two sections looked there
as the page opened: Remove jobdork from this computer listed the
installers Update had left in Downloads and the letters in
Documents/job-applications, with their sizes, and Folders checked that
the documents folder exists to enable its Open button. Both now wait
for you, on every system (they are your folders, not the app's): the remove section shows those two items as not looked
at yet with an Also look in Documents and Downloads button (they can
be ticked without it), and Open checks the folder when clicked. macOS
still asks the first time jobdork writes a letter into Documents, or saves
an update into Downloads: that is when it needs them.
Fixed — 2026-10-28 — The resume's full path, and keys from the environment named
The Resume page said "Stored at data/resume.pdf" (or "Currently:" on a
later visit): the setting as written, relative to wherever jobdork started.
It now shows the full path of jobdork's copy on disk, and right after an
upload the name of the file you chose with it.
Enter your keys, on a start with keys held in memory only, asked for
some keys and silently skipped others. A key set in the computer's
environment outside jobdork (OPENAI_API_KEY in a shell profile, or
launchctl setenv on a Mac, which every app opened afterwards inherits)
counts as held, so it was never asked for, and nothing said so. The window
now lists each such key with its variable and a box to paste another for
this run; the AI page names the variable too ("using OPENAI_API_KEY from
your computer's environment").
Fixed — 2026-10-28 — AI providers by their company names, and a save message that says which line
The AI page and setup's AI step named two providers by their chat apps: they
are now Anthropic (was Claude) and OpenAI (was ChatGPT), with
Anthropic's own mark in place of Claude's spark; Ollama, Gemini and Groq keep
theirs. The names change wherever a provider is named: the provider buttons,
the key boxes, each model's label ("OpenAI · gpt-5.5"), error messages, the
guide and the docs. Claude Code, which drafts documents, keeps its name.
Config is unchanged (anthropic, openai).
Saving a line that cannot be used yet said only "Saved. no model chosen for
Groq · checked by ChatGPT · gpt-5.5", which read as if the save had chosen
Groq. It now names the line: "Saved, but reader 1 cannot be used yet: no
model chosen for Groq." A later reader or checker that cannot be used is
named the same way after a working first one, and a checker shown as
"checked by" is one that is ready. Save and test used to stop there
without a word; it now adds "Not tested until then."
Added — 2026-10-28 — A tailored resume in Indeed's layout and length, as Word too
A tailored resume is now laid out like Indeed's free resume template and held
to Indeed's advice on length
(https://www.indeed.com/career-advice/resumes-cover-letters/how-long-should-a-resume-be).
The name, then Phone | Email | Location and the website line, centred;
then Professional summary, Experience, Skills, Education, Certifications and
a new Awards & volunteer work section (each award checked against the resume
like an employer); each job's dates at the right margin with Company,
Location under the title; skills separated by |; Arial's equal at 11
points, 0.7 inch margins. Students still lead with Education and projects
still go by career stage.
Length is enforced in code (writing/length.py), not only asked of the
model: one page for up to ten years of experience, two after that, three
past 25 years or for academic and medical jobs; jobs older than 15 years
left out; seven bullets for the newest job down to three for older ones;
and while the PDF runs over, an older job's last bullets go first, then
extra projects, awards and the end of a long skills list. A new page
count check says how many pages it fills and lists each cut.
Besides the PDF, jobdork tailor and Write tailored resume (AI) write
the same resume as a Word file (FirstName_LastName_JobTitle_Resume.docx),
and the document's tab has Download Word. The .docx is written by
output/docx.py with no new dependency, keeps any script (the PDF still
spells non-Western letters with the nearest ones), and has no tables, text
boxes or headers for a tracking system to trip on. The template's own file
is not in the repository: the layout is rebuilt. Claude Code drafts
(CV.md) get the same layout and length rules.
Changed — 2026-10-28 — Workflow names that say what each one does
The Actions page listed "Test, scan and publish", "Desktop app", "guide",
"Scorecard" and job names like build and analysis, which said little
about what a run was for. Every workflow now says what it does and when, and
every job what it checks:
- CI: test, scan and build every change; release on a tag (docker.yml),
with "Installers: needed for this change?", "Release: publish the Docker
image (after your approval)" and "Release: create the GitHub Release page". - Installers: build and self-test the Mac and Windows apps (desktop.yml):
"Build the Mac (Apple Silicon) installer", "... Mac (Intel) ...",
"... Windows ...", "WinGet: install and uninstall the Windows build for
real" and "Windows: photograph the installer and uninstaller". - Guide: build the user guide website; publish it from main.
- Boards: weekly re-check of the built-in employer boards.
- PR check: no vulnerable new dependencies.
- Scorecard: weekly supply-chain security score.
The four checks main requires keep their names (test (3.10),
test (3.14), security, image): main's ruleset names them, and a
renamed one would never report, blocking every pull request.
Changed — 2026-10-28 — The installers built when they need to be
The Mac (Apple Silicon and Intel) and Windows installers were built on every
pull request and every push to main, with the WinGet test and the Windows
pictures: five runners each time, which kept GitHub's queue full and every
other check waiting. Now a small packaging job in docker.yml decides: a
release tag always builds them; a pull request or push builds them only when
it touches what they are built from, packaging/,
requirements-desktop.lock, pyproject.toml (so a release pull request,
which sets the version there, builds them before the tag), the two
workflows, or the code only the installed app runs (the window, the update
check, notifications, the keychain, uninstalling). desktop.yml's Run
workflow button builds any branch by hand.
The Windows pictures (the installer's pages and the uninstaller's window,
added for 0.15.4's high-scaling fix) are taken only when the installer
script itself, packaging/jobdork.iss, changes, or when Run workflow's
"Photograph the Windows installer and uninstaller" box is ticked; never on a
release.