jobOS

Find jobs with a harness, not a spreadsheet.

jobOS is a job harness. Catalog, research, and discovery shortlist the work. Apply submits from a packet. Outreach follows by email and LinkedIn DM. Live state lives in Hub. Git ledger/ is export and airgap — not a second board.

Log in or create an account Get started

What it can do

Capabilities from the process, not a marketplace pitch. Discovery never Easy Applies.

Discovery
Source, enrich, score, and shortlist. Write a complete apply packet. No form submit. No Easy Apply.
Deep research
Company, people, role, and portal research. A shared tool the owners call — not a fourth hunt bot.
Apply
Claim apply_ready work, submit from the packet, write something worth reading, log proof. Blocked portals stay blocked.
Outreach
Email with the resume PDF attached, plus LinkedIn DMs asking for a short chat. Default channels only. No SMS, WhatsApp, or InMail sequences.
Catalog + registry
Discover and vet skills and MCPs. Hub live selection is GET /api/registry/skills tagged sourceapplyoutreach. Git lists remain inventory.
Hub + ledger
Hub JobStore (Postgres) + BlobStore + RabbitMQ is the live list. GET /api/export/ledger writes YAML when you need files. Do not dual-write.
Roles
Three hunt owners — discovery, apply, outreach — plus maintainer (seldom): skill lists, Hub registry proctor, hunt board. Review and resume are stubs. Do not claim them.

The loop

Catalog + research + discovery, then apply, then outreach. Maintainer is a fourth owner and is seldom used. Nobody jumps the queue.

Hunt loop Maintainer ≠ discovery
  1. 1

    Catalog

    Vet skills and MCPs. Write the stack the workspace can actually use.

  2. 2

    Research

    Deep research on the company, team, and portal. Shared across lanes.

  3. 3

    Discovery

    Shortlist + apply packet. Tagged source skills. Never Easy Apply.

  4. 4

    Apply

    Submit from the packet. Tagged apply skills (ATS / site).

  5. 5

    Outreach

    Email + LinkedIn DM after apply, or when there is no portal.

Maintainer (seldom)

Owns skills/lists/, seeds and proctors the Hub skill registry, keeps the board up. Does not source jobs, apply, or send mail.

Registry tags

When Hub is up, pick a matching tagged skill: ?tag=source, ?tag=apply, ?tag=outreach. Airgap falls back to role SKILL.md plus starters. Do not load another bot’s local skill.

Hub JobStore (live)

Canonical jobs, events, people, skills, memory, and queue traffic. Sign up at /login. Board is /board after a truthful candidate profile exists.

git ledger/ (airgap export)

Not the live list when Hub is up. Export with GET /api/export/ledger. Airgapped bots still write YAML locally. Never dual-write.

Get started

Onboard a truthful candidate before anything else. Do not invent one. Do not ship the example profile as live data.

  1. Install Hub Compose on port 8787, or see docs/install.md. Agents curl llms.txt / agents.txt — they do not scrape HTML.
  2. Create an account Username + password at /login. No dashboard token. First boot still uses the printed /admin/signup bootstrap URL.
  3. Onboard the candidate /onboarding is mandatory: identity, titles, locations, resume PDF. Until then the Hub returns 409 candidate_required. Jobs, invites, and A2A stay closed.
  4. Claim a role and join workers Read docs/process.md and docs/roles.md. Mint track invites. Agents join, heartbeat, and claim tenant-namespaced work.
  5. Run the loop in order Discovery first (catalog → research → shortlist). Then apply. Then outreach. Maintainer is not a hunt lane.

Blocked until onboarding. Missing candidate.yaml / Hub onboarding is a hard stop — not a skip. Full sequence: Get started.