Repository navigation
Finding jobs
Four ways to find jobs worth applying to, and one to tell which of them fit: the assistant searching for you, the Search Builder's queries for the job boards, the watchlist of companies you'd work for, and recruiters' posts.
AI Assistant (/discover) is where your assistant works for you. Its presets ask it to find jobs matching your CV, find jobs for requirements you give (titles, where, skills, seniority, minimum salary, words to leave out...), run your saved searches, check which postings closed, score your jobs, or tidy up applications with no answer. Each preset saves a request (ai_requests) and shows a short prompt to paste into your assistant, or the command that starts it with the prompt (AssistantPrompts in web). The page also gives a terminal loop that picks up every request on its own: it asks JobWarden every five minutes whether one waits (GET /claude/requests?profile=…, a plain number), and only then starts the assistant, afresh each time (claude -p or codex exec), with JobWarden's tools and the web allowed. Waiting costs no tokens, and no request reads through the ones before it, as it would in one long session; Auto-apply, which needs you at the browser, isn't among them. The page follows each request as it goes, lists the jobs found for it with why each fits, and adds them to the cards to save, apply to or discard.
Search Builder (/search) builds Warden Queries, JobWarden's name for what recruiters call Boolean searches, from job titles, locations, skills, work and employment types and terms to exclude: for Google limited to the job boards companies use (Lever, Greenhouse, Ashby...), LinkedIn, Reddit, X and the remote job boards. Google reads only the first 32 words of a query, so the generator leaves out the terms that don't fit, last ones first, and says so. A description of the job you want ("senior backend engineer, remote Europe, contract") fills in the filters. The address keeps the filters, so a search can be bookmarked; with a profile, Save search keeps them under a name, and Saved filters in the header opens them again with every filter filled in, or deletes them (the same saved searches the assistant finds jobs for). The page fills the window: the filters on the left, and on the right every platform's query at once, its operators, fields, phrases and exclusions in colour (QuerySyntax in web), each with Copy and Open search. The rules are in the search module (QueryGenerator, DescriptionParser).
Watchlist (/watchlist, in the sidebar): the companies you'd work for, read with no AI tokens at all. Paste a link to one of a company's jobs, or to its careers page, on Greenhouse (boards.greenhouse.io/…, job-boards.greenhouse.io/…), Lever (jobs.lever.co/…, jobs.eu.lever.co/…) or Ashby (jobs.ashbyhq.com/…), several at once if you like, one per line; or click the eye in the extension's popup on one of its jobs. The page also suggests the companies whose boards your jobs are on. JobWarden reads the boards' public feeds itself, which need no key, every six hours (jobwarden.watchlist.every, each board once for every profile watching it, a few at a time, with a timeout, a size cap and a JobWarden/<version> User-Agent), or now with Check now. The postings that match one of your saved searches go to the cards to review, like the jobs the assistant finds, with the search's name; those whose links you have are left out. The matching needs no AI: every word of a title term in the title, in any order (developer, engineer and programmer count as one); no excluded word in the title or the location; a location term in the location's words, or for a region or a country, a place in one of its countries (Regions, in the search module, knows the regions the Search Builder suggests, the countries by their names in several languages, and the big cities and US states boards name alone); remote, hybrid or on-site and the kind of contract as the feed says them; a skill in the description; and the search's recency. Where the feed leaves something out, the posting isn't held against it: one with no description passes the skills, one with no date passes the recency, one with no work type counts as on-site or hybrid. A reading adds at most 25 new postings per company, the newest; when a board has more matches, the next reading (or Check now) adds the next ones. The assistant helps fill the list: its steps for finding jobs name the companies watched, so it looks elsewhere, and watch_company adds the boards it comes across.
Who's hiring (/hiring): recruiters' and hiring managers' posts saying they're hiring, the way LinkedIn feeds show them. The assistant finds them when the page asks (a FIND_HIRING_POSTS request, for the roles, places and time you give): it searches LinkedIn, X and Reddit, looks at the people you follow first (list_recruiters), and adds the real posts with add_hiring_posts; JobWarden logs in nowhere. The Discover tab shows every post, the latest first, filtered by time, country, hashtag or a search; Following, only those of the people you follow; Recruiters, everyone, with their company, their number of posts and how they came. Recruiters come from four places: the authors of the posts; the assistant, asked on the Recruiters tab (a FIND_RECRUITERS request for the roles and places you give, starting with the companies in your Jobs), which adds them with add_recruiters and a note on why each is worth following; the contacts you save on your jobs, with a name and a profile link, which become recruiters you follow the first time the page sees them, linked to their jobs; and those you add by hand. Follow a recruiter from any of their posts, remove one for good (the assistant won't add them again), hide a post for good, or save it as a job, bookmarked, with the post as its description (the hiring module: recruiters and hiring_posts).
Fit: apply better to fewer. On a job's page (and its side panel), the assistant scores how well you fit the job (a SCORE_JOB request; Score my jobs on the AI Assistant page scores up to 15 saved and applied jobs at once, SCORE_JOBS). It lists 3 to 15 of the posting's requirements, each a must-have or a plus, met, partly met or missing by your CV, and saves them with save_job_fit. JobWarden works out the score from them, the same way for every job, so scores compare: 1 plus 4 times how much of the requirements the CV meets, a must-have weighing three times a plus and a partial match half, and a missing must-have keeping it at 3 at most (JobFit.score). 4 and up is a strong fit, 3 to 3.9 a partial one. The score shows next to each job in the list (which sorts by it) and the pipeline. With no AI, JobWarden also flags a posting that rules out sponsoring a visa when your Auto-apply answers say you need one, a job posted again (the same title at the same company under another link), and a posting up for more than 60 days (its job data's datePosted): maybe a ghost job (job_fits, in jobs).
JobWarden · Releases · Discord · Suggest a feature · Report a bug · MIT License