Career-Ops Agentic OS

An autonomous job-search loop that scans, evaluates, tailors and applies while I sleep.

The problem

A serious job search is a part-time job. Scanning boards, judging fit, tailoring a CV and filling the same application forms eats hours every day, while companies use AI to filter candidates on their side.

The open source Career-Ops project gave me a strong multi-agent pipeline, but it still needed a human at the keyboard for every run. I wanted the whole loop to run unattended on my Mac.

How it works

1. Cloud half, daily at 09:001

   Scan

   API and Playwright browser scans across LinkedIn, Indeed, CV-Library, Reed and WTTJ.

2. 2

   Evaluate

   Each role gets an A to G fit score against my profile.

3. 3

   Tailor

   High-fit roles get a tailored CV generated for them.

4. 4

   Queue

   Approved roles land in a queue file as the handoff point.

5. Local half, every 30 minutes5

   Auto-fill and submit

   A persistent Playwright profile drains the queue and fills forms from my data pack.

6. 6

   Daily digest

   A zero-LLM Gmail summary reports what was scanned, scored and submitted.

What I built

• A two-half autonomous loop on macOS launchd: a cloud half that runs daily and a local half every 30 minutes.
• A Playwright scanner over five job boards using a persisted logged-in browser profile.
• A to G fit scoring and tailored CV generation through Claude Code multi-agent modes.
• A queue handoff between the halves with a full audit log of every submission.
• Safety rails: a kill switch, dry-run window, score thresholds, dedup against past applications, daily caps and anti-bot pacing.
• A zero-LLM daily email summary over Gmail SMTP with nodemailer.

Outcome

The search runs itself. Applications go out daily with scores and an audit trail, and my involvement drops to reading the morning digest and taking interviews.