Argopia

Job-search automation for engineers, inside Claude Code.

Hand it your CV. It scans the boards, scores every opening against you, and shows its reasoning. Free, on your laptop, no API keys.

Argos Panoptes — the hundred-eyed watcher of Greek myth — guarded what mattered. Never sleeping, never blinking, every eye on a different horizon. Argopia is the same idea, narrower: many eyes on many job boards, watching for the role you actually want, while you do something else.

---

What it does

Scanning the same handful of job boards week after week — sifting through repetitive listings for the handful that actually fit — is the kind of low-signal work that bleeds time and quietly stops happening when life gets busy.

Argopia runs that loop for you. It derives filter rules from your CV, surveys public boards, scores every survivor against your profile with a transparent rubric, surfaces matches in a local dashboard, and reflects back on what the market wants vs. what your CV shows as more reviews land. Your hour goes into the application, not the search.

$ /argopia-onboard ./my-cv.pdf
✓ Onboarded with the default template
Profile (extracted from your CV): who you are, what you've built, what you know
Criteria (derived from your profile): what you want, what you won't accept

$ /argopia-survey
✓ 8 sources, 47 listings → 12 queued after filter

$ /argopia-review
✓ 12 scored — 2 apply (≥70), 8 research, 2 skip

$ npm run dashboard
→ http://localhost:4242

dashboard.mp4

How it works

Two stages, two cost profiles.

1. Survey runs in deterministic Node — fetches listings from public boards, drops 70-90% via keyword + region + staleness filters, and caches the survivors as content-addressed posting files. Cheap, predictable, no tokens spent.
2. Review then runs inside your existing Claude Code session — reads each cached posting, scores it against your CV with a two-stage rubric (binary gates → 3-criterion fit score), and appends one JSON line per opening to data/reviews.jsonl. This is where Claude spends — only on what the filter let through.

The dashboard renders that ledger; /argopia-advise reflects on it to suggest CV, criteria, or source edits.

Re-runs cost only what's new since last time.

Principles

• CV is the spine. Every config — keywords, deal-breakers, target levels, excluded companies — is derived from your CV, not the other way around.
• Free everything. Public boards + your existing Claude Code session. No paid APIs, no Anthropic billing, no cloud storage.
• Human ranks, you apply. Argopia surfaces, scores, and explains; you decide whether to apply. No spray-and-pray.
• Yours, not the bot's. working/ is your editable source of truth. Templates are starting material; you make them yours.

Quick start

Prerequisites: Node 20+ and Claude Code.

git clone https://github.com/samson-ailabs/argopia.git
cd argopia
npm install   # auto-runs scripts/install.mjs (env check + dir setup)
claude .      # open in Claude Code

Then inside Claude Code:

/argopia-onboard ./your-cv.pdf

From there: /argopia-survey → /argopia-review → npm run dashboard. See The pipeline below for what each command does.

The pipeline

Command	What it does
/argopia-onboard <cv>	Parse CV → populate working/profile.yaml and criteria.yaml
/argopia-survey	Discover URLs from enabled sources, fetch + cache JD postings, filter, queue openings
/argopia-review [--limit N]	Score each queued opening against your CV; append one JSON line per opening to data/reviews.jsonl
/argopia-advise	Reflect on reviews.jsonl → recommended edits, longer-horizon patterns, and evidence backing
npm run dashboard	Build reports/dashboard.html — sortable, filterable triage view of every review

Configuration

Three files in working/, three matching files in templates/. Each owns one concern:

File	Role	Filled by
profile.yaml	Identity — who you are, what you've built	profile-extractor, then you
criteria.yaml	Preferences — what you want / won't accept	criteria-deriver, then you
sources.yaml	Where to look — one entry per board	Argopia defaults, then you

The scoring rubric lives inside /argopia-review — not as a separate file the user maintains.

Customizing. templates/ ships starter scaffolds for your tech / role / region. Edit them before running /argopia-onboard, or edit the copies in working/ after onboarding. No code changes required — if you find yourself reaching into scripts/, that's a bug; open an issue.

Updates are safe. working/, data/, reports/ are gitignored — your personal state stays put. git pull only updates Argopia's code (templates, schemas, scripts, slash commands, agents).

Roadmap

v0.1 (current — public alpha). Full pipeline works end-to-end: onboarding, the curated api + html boards in templates/sources.yaml, the scoring rubric, the dashboard, and on-demand advise.

Next up, roughly in order:

1. SPA boards via browser MCP (Indeed, Wellfound, etc.) — JS-rendered listings, no auth needed but bot-protected.
2. Auth-walled boards (LinkedIn, Glassdoor, etc.) — login + session management on top of browser MCP.
3. Domain templates beyond speech-AI — frontend, backend, ML research, data engineering

Speculative / depends on demand:

• Multi-CLI support (Gemini CLI, OpenCode, Qwen)

Contributing

Code contributions welcome. The issue tracker holds bugs and feature requests; starter-friendly ones are tagged good first issue. Two especially-welcome paths:

• Add a board to templates/sources.yaml (api or html; SPA / auth boards await browser-MCP support).
• Add a domain template if your tech stack isn't speech-AI — the three working/ files plus a starter rubric is all that's needed.

For larger work, open an issue first so we can sanity-check the scope.

Sponsor

I built Argopia for my own job search. If it saves you hours too, sponsoring on GitHub returns some of that value and accelerates the roadmap above — more boards, more templates, faster issue triage.

License

MIT — use, fork, modify freely. Attribution appreciated.