Job search agents
A daily agent that finds senior roles across about eighty sources, reads each advert properly, scores it against my profile and emails me the best.
The problem
Many of the roles I want never appear on job boards. They sit with executive search firms or on a target company's own careers page. Therefore, a general job search tool misses most of them.
What I built
I started from an open-source job search project, which was written for a non-UK jobseeker and a short list of boards. I extended it for a senior UK search that runs without me.
Every morning a scheduled Claude Code task scrapes, ranks and emails. It never applies for anything or contacts anyone. It finds, scores and tells me why.
The sources are about eighty places: the main UK job boards, a sweep across Adzuna, roughly thirty-five recruitment and executive search firms, and twenty-four target company careers pages.
What happens at 08:15
- ScrapeBoards, search firms, target companies
- MatchSeniority word plus function word
- RankFull advert read, scored 0 to 100
- EmailDigest of the best, with reasons
How AI was used
A rule instead of a phrase list
The original tool searched a hand-written list of job titles. A review of the first two hundred results showed only eight literally said the title I had in mind. I replaced the list with a rule: a title matches if it has a seniority word and a function word. It catches titles nobody thought to type.
Research the market, then target it
I used AI to help decide which companies deserved a place on the target list, based on the parameters of my search, then added their careers pages as sources.
Read every advert
Scoring is based on reading the full advert against my profile, giving a score out of a hundred and a short reason. The reasons let me check the thinking.
Skills that follow me into the application
The same toolkit tailors a CV and cover letter for roles I choose, and helps prepare me for interviews. I will always tailor these before I use them in earnest, but it helps save a lot of time.
A reviewer skill that checks every claim
A drafter and reviewer pair of skills writes and then independently critiques each CV. The reviewer checks every claim against my profile file and flags anything the profile does not support. I confirm or reject each point, and anything I confirm goes into the profile so it stays the single source of truth.
What I think is clever
- Tiered scanning: search firms that have produced jobs are checked daily, the long tail weekly. Firms that only recruit confidentially are expected to show nothing.
- Unglamorous problems, solved. Some sites block automated readers, and one served a hidden decoy link, so the agent reads those in a real browser. An advert it cannot read is marked deferred, not expired, so nothing disappears silently. After a run stopped part-way and sent an email full of raw code, I added a run log and a check that blocks broken emails.
- The search grows from my real network. A recruiter portal I could log in to, a private directory of search firms and a friend's list of recruiters each became new sources.
- A graceful fallback. Where a site blocks automated access, or no browser is available in the unattended run, the agent drops back to web search.
- Nothing leaves the machine unreviewed. The agent reads and reports, and I decide.
What I learned
- Replace brittle lists with rules that apply identically to every source.
- Measure what you actually get. Counting how many results matched the title I expected changed the design.
- Write the explainer as you go. A plain-English description of how it works is also the best thing to show someone else.
The detail behind this project is personal, so this page describes the method only.