{% extends "settings/_layout.html" %} {% block title %}Sources — roledesk{% endblock %} {# One page, three sections: the `search` plugin's job board queries, the `ats` plugin's company boards, and the runs that read both, with the message log as their last block. One is shown at a time, the one the URL fragment names (`#runs`) or contains (`#messages`), or Job boards without one; sources.js hides the rest and the rail picks one. Hidden rather than separate pages, because each section keeps the element ids it had as a page of its own and its script (sources.js, companies.js, jobs.js, messages.js) is unchanged: it finds its root or does nothing. The ids are `SECTIONS` in app/web/pages.py, which the rail links to. #} {% block content %}
Sources are where roles come from: queries on job boards that span companies, and the boards of the companies you follow. Each section is one plugin; the runs read them all.
{# The Job boards form is a `` because only one edits at a time, so its ids stay unique: sources.js puts it in place of a row on Edit, and on top of the list on New, as agents.js does for the Search agents cards. #}A query on Jobindex, Adzuna, Platsbanken, freehire, Jobnet or Akademikernes Jobbank finds roles at any company. Switching one off closes its roles until you switch it back on.
Every company roledesk has seen, and how many of its roles are open. Switching a board off stops reading it and closes its roles; hiding a company keeps reading it but leaves its roles out of the list and unscored. Like or favourite raises a company's roles a little in the cut and in the feed order, in each search agent's own ranking (a liked company about 0.05, a favourite about 0.1), and passes nothing the fetch gate rejects. All countries and All roles are for every agent: the first skips the place part of the fetch gate, the second the fetch terms and the titles to leave out.
Liking a company needs a search agent. Add one under Search agents.
On the schedule below the pull reads every board and every search, embed and score rank every role and fix each agent's cut, fetch text reads the own page of the cut roles that came without text, extract fills in what the cut postings say, score runs again with what it filled, judge scores the cut against each search agent's CV, and summarise writes the summaries for the top of the feed.
{# The stages with a switch, switched here rather than in config.yaml (`GET /api/stages`); jobs.js draws the rows into #stages-list. #}Extract, summarise and judge ask a model, which costs tokens on the API; fetch text asks none. Turning one on starts nothing: the next pull runs it, and Run the pull now does so at once. How deep each reads is set per search agent, under Search agents: fetch text, extract and judge read the agent's cut, and summarise reads its own size.
What the server said that nobody has cleared, newest first. A message goes when you clear it, or a fortnight after it was last said.
What prep could not place. A place text no city or country resolves leaves its role without that place, and a host word no table maps leaves the level, contract or remote empty. Tell Claude what one is, such as "Überlingen is a town in Germany" or "Home Based - Americas is remote", or paste the copied list. The fix goes into the code, and the next pull applies it to every open role.