Neptune Vibes

Home · Agents

How it works: Job Scout

For a daily search for remote roles: reading only the job sources that allow it, ranking each role by pay and flexibility with a reason, and never applying on its own.

Its devlog

Philosophy

  • Read only what's allowed. Every source's terms were checked before any code was written. Job Scout uses documented feeds and public job-board APIs, reads each once a day, and links back to every original listing. Sites that forbid automated reading aren't touched.
  • Every score comes with a reason. A number alone can't be argued with. Each job shows why it ranked where it did, and red flags such as fixed core hours, contract-to-hire or state-limited remote are labelled, not hidden in the score.
  • Flexibility and pay lead. The ranking weighs how much freedom a role gives first, then what it pays. A perfect title can't rescue a job that demands fixed hours and office days.
  • Finding is not applying. Job Scout never applies, submits or sends anything. It finds and prepares; applying stays a separate, manual step.
  • Check the number against the listing. Pay parsed from free text is compared with the original posting, and every bug fix ships with a test shown to fail on the old code first.

The workflow

Job Scout: the workflow as a flowchart1Start the morning runthe schedule starts one search a day.daily routine2Read the approved sourcesjob feeds and company job boards, once each, keeping every job'slink.↳ A source doesn't answer → noted on the list page, and the runcarries onJob Scout3Filter the resultsdrop duplicates, on-site roles, roles outside the country and paybelow the floor; keep hybrid, marked as a fallback.Job Scout4Score and explain0–100 for each job, with a short reason and labelled red flags.Job Scout5Update the list pagenew jobs first, then everything still waiting for a decision.Job Scout6Send the morning summaryhow many new jobs, and the best one.daily routine7Review the listopen a listing, then mark it "I applied", Save or Skip.↳ A job worth a closer look → Save, and it waits in the Saved tabowner8Draft the applicationa tailored résumé and short cover note, drafts only (being built).Job Scout9Applyread the draft, change what needs changing, and apply on theemployer's own site.ownerThe draft needsworkJob Scout: the workflow as a flowchart1Start the morningrunthe schedule starts one search a day.daily routine2Read the approvedsourcesjob feeds and company job boards, onceeach, keeping every job's link.↳ A source doesn't answer → noted on thelist page, and the run carries onJob Scout3Filter the resultsdrop duplicates, on-site roles, rolesoutside the country and pay below thefloor; keep hybrid, marked as a fallback.Job Scout4Score and explain0–100 for each job, with a short reasonand labelled red flags.Job Scout5Update the list pagenew jobs first, then everything stillwaiting for a decision.Job Scout6Send the morningsummaryhow many new jobs, and the best one.daily routine7Review the listopen a listing, then mark it "I applied",Save or Skip.↳ A job worth a closer look → Save, andit waits in the Saved tabowner8Draft the applicationa tailored résumé and short cover note,drafts only (being built).Job Scout9Applyread the draft, change what needschanging, and apply on the employer's ownsite.↩ The draft needs work → step 8owner

What it owns

  • The daily search: reading the approved sources, removing duplicates, filtering and scoring.
  • The list page: ranked jobs sorted into To review, Saved and Done, each with its pay, reason and flags, and a link to the original listing.
  • The record of which sources allow automated reading, and on what terms.
  • A demo data set of made-up jobs, which every public picture of the tool is made from.

What it hands off

  • Project Manager: runs the daily schedule and turns each run's summary into the morning notification.
  • Art Director: the spyglass mark and the tool's colours.
  • A person: every application, and what the search should favour.

Guardrails

  • Never applies, submits a form or sends a message for anyone.
  • Reads only sources whose terms allow automated reading: no scraping, no logged-in reading, no CAPTCHA workarounds, and never an application endpoint.
  • Never publishes the search itself. Public pictures and posts use made-up demo jobs, and the real results stay local.
  • Never guesses silently: a job with no posted pay is kept and labelled, not dropped or estimated.

At work

A daily run on the made-up demo data: the summary arrives line by line, opening with the counts and the best new match, then the top five with their reasons.

A daily run on made-up demo data