jobforge

README

JobForge

A Windows-first Go CLI for running a real job search end to end - without surrendering control of any account or letting a tool send an application on your behalf.

JobForge pulls listings from public ATS APIs, ranks them against your profile with an LLM, drafts outreach through a multi-agent pipeline, schedules interviews on your Google Calendar, and watches your inbox for replies. Every state change passes a human approval gate.

What it does

• Aggregates real jobs. Greenhouse, Lever, Ashby, SmartRecruiters, Workable, Recruitee, and RSS from public endpoints, plus opt-in scraper connectors for LinkedIn, Indeed, Glassdoor, and Handshake (optional CloakBrowser for fewer blocks). Scrapers never log in, bypass CAPTCHA, or submit applications.
• Ranks them. jobforge feed scores each posting 0-100 against your profile + active resume via your configured LLM, with a one-sentence rationale per job.
• Drafts outreach. A researcher -> writer -> reviewer agent pipeline produces a cover-letter-style draft. You see it, edit it, and gate on y/n/e before anything saves.
• Tracks the lifecycle. Add interviews and application deadlines to Google Calendar (OAuth Desktop client). Poll an IMAP inbox for confirmations, interview invites, rejections, and offers - JobForge classifies each via LLM and proposes status updates one by one.
• Stays out of the way otherwise. Bare jobforge drops into an interactive REPL with a live / palette (like Claude Code's). Plain English works too: type what you want and the LLM resolves it to a command, which you confirm before it runs.

Install

Pre-built binaries for Windows / Linux / macOS on amd64 / arm64.

Windows:

iwr https://raw.githubusercontent.com/0xcryptj/jobforge/master/install.ps1 -useb | iex

Linux + macOS:

curl -fsSL https://raw.githubusercontent.com/0xcryptj/jobforge/master/install.sh | bash

Both installers download the matching binary from the latest GitHub release, put it on your PATH (%LOCALAPPDATA%\Programs\jobforge on Windows; /usr/local/bin on Unix, override with PREFIX=...), and install scraper dependencies by default: Python 3.10+ (Windows: via winget if needed) and CloakBrowser (pip install + ~200MB Chromium download). Skip that step with $env:JOBFORGE_SKIP_SCRAPER_DEPS = '1' (Windows) or JOBFORGE_SKIP_SCRAPER_DEPS=1 (Unix).

Other paths (go install, winget, scoop, direct download, build from source): see USAGE.md -> Install.

Quickstart

# one-time setup: profile, CloakBrowser scrapers, LLM, default boards, optional import
jobforge onboard

# or step by step:
jobforge init
jobforge profile setup
jobforge sources seed
jobforge jobs import --all

# discover jobs (scrapers fill URLs — no manual entry)
jobforge sources add-scrapers
jobforge jobs sync

# rank them against your profile + resume
jobforge feed

# draft outreach for the top match (id from `jobs list`)
jobforge apply --llm <id>

# (after `calendar auth`) schedule an interview
jobforge calendar add-interview <id> --when "2025-06-10 14:00" --duration 45m

# poll the inbox for replies and propose status updates
jobforge inbox poll

# or just type `jobforge` for the interactive REPL
jobforge

See USAGE.md for the full command reference, OAuth and IMAP setup, source configuration, scraper caveats, troubleshooting, and the data-on-disk layout.

Scrapers and CloakBrowser

LinkedIn, Indeed, Glassdoor, and Handshake imports use a headless browser (chromedp). By default JobForge picks CloakBrowser when its Chromium binary is installed (stealth fingerprints, fewer blocks); otherwise it falls back to system Chrome.

jobforge onboard              # installs CloakBrowser when you opt in
jobforge scraper setup        # scraper-only wizard
jobforge doctor               # verify profile, LLM, and browser readiness

Optional config.json (scraper.engine: auto, cloak, chrome, or cdp). Details: USAGE.md -> Scraper sources.

Scrapers remain read-only, opt-in, and fragile (HTML changes, rate limits, host ToS). CloakBrowser improves fingerprinting; it does not solve CAPTCHAs or replace proxies for aggressive anti-bot sites.

Privacy posture

JobForge is built around an explicit promise:

It never submits an application on your behalf, never sends a message from your account, and never marks remote state as read or modified behind your back.

Concretely: the apply flow saves a .txt you copy and send yourself. IMAP reads use BodyPeek so the server never sees Seen flags. Google Calendar gets only the calendar.events scope. Every state change - locally on disk, on your calendar, or in your job tracker - runs through a default-deny [Y/n] gate.

Secrets stay local: the Google refresh token and the IMAP password each live in their own file under %APPDATA%\jobforge with restrictive permissions, never logged or transmitted anywhere except the corresponding service.

Job importing supports public source connectors plus explicit opt-in scrapers for LinkedIn, Indeed, Glassdoor, and Handshake. Scrapers are read-only search result importers: no credentials, no CAPTCHA bypass, no form filling, and no application submission.

Status

v0.6. Builds and tests green on Go 1.26+ across Windows, Linux, and macOS. Public source paths (Greenhouse / Lever / Ashby / SmartRecruiters / Workable / Recruitee / RSS) are the stable import surface; LinkedIn / Indeed / Glassdoor / Handshake scrapers are opt-in and fragile. Use jobforge onboard or jobforge scraper setup for guided CloakBrowser installation.

License

MIT — see LICENSE.