Network Jobs

Local-first agent skill suite: import your resume and preferences, import LinkedIn connections, discover company career pages, build a personal job corpus, search openings, and draft warm intros — on your machine.

Works with any agent that loads Agent Skills (Claude Code, Cursor, Codex, OpenCode, Muse, Pi, Gemini CLI, Grok Bot, and others).

For agents

If you're an agent working from this repo rather than a human reading it:

1. Install the skills into yourself — npx --yes @hirefrank/network-jobs@latest setup --agent auto (or npx skills add hirefrank/network-jobs -g -a …, then network-jobs setup for data + launcher).
2. Start with network-jobs-setup — it turns the user's resume into profile.json + preferences.json via an interview. Everything downstream reads those two files.
3. Follow the skills table below — each SKILL.md is self-contained and states its inputs/outputs. The day-to-day pair is careers-discover + jobs-ingest (build the corpus) and network-jobs (search it).
4. For routing, run network-jobs routing — it prints a snippet to drop into a project instruction file so future sessions pick the right skill.

Always read SCHEMA.md for data paths and JSON shapes. Daily use lives in the agent; the CLI covers setup, import, and quick search / companies / refresh. Release notes are in CHANGELOG.md.

Install

npm i -g @hirefrank/network-jobs@latest
network-jobs setup --agent auto

This creates ~/.network-jobs/, installs agent-browser when missing (and upgrades it when it is below the floor declared in package.json), puts network-jobs on PATH (~/.local/bin), and symlinks all six skills into each detected agent’s skills directory.

Search works out of the box with keyword scoring. Semantic matching is optional: network-jobs embed-setup adds resume↔job similarity on top of keyword scoring for better fit ranking. It pip-installs fastembed (and dependencies such as onnxruntime) into your Python, downloads a one-time ~60MB model (needs network), and uses a local Ollama instead when one is already running. Setup does not install it; skip it if you're offline or want a lean install.

Zero-install try (no global install):

npx --yes @hirefrank/network-jobs@latest setup --agent auto

Bleeding edge from git main:

npx --yes 'github:hirefrank/network-jobs#main' setup --agent auto

network-jobs setup --agent claude-code,cursor,codex
network-jobs setup --agent all
network-jobs setup --force          # replace skill dirs from an older install
network-jobs setup --no-browser     # skip agent-browser install
network-jobs setup -v               # print every path
network-jobs update                 # fetch latest + relink
network-jobs doctor

Requires: curl, jq, unzip, python3, and a working npm (for agent-browser). Optional: pdftotext for PDF resume extraction.

agent-browser is only needed for JavaScript-rendered career pages — skip it with setup --no-browser and static boards still work. network-jobs doctor prints the installed version and warns when it is below agentBrowser.minVersion in package.json; network-jobs update upgrades only below that floor. See skills/careers-discover/references/agent-browser.md for the command reference (subcommands are grouped: network har, not har).

Optional alternate skill placement: npx skills add hirefrank/network-jobs -g -a … then still run network-jobs setup for data + launcher.

Then use your agent

• “Set up network jobs from my resume” → profile + dynamic preferences interview
• “Import my LinkedIn zip at ~/Downloads/….zip”
• “Find open roles at my top companies” → review triage → “ingest that batch”
• “Senior PM jobs in NYC” → “Draft an intro to Jane”

Daily use is in the agent. The CLI also covers install, update, doctor, resume/LinkedIn import, corpus clear, and quick search / companies / refresh.

Tips

• Pipeline stop: careers-discover only stages under triage/. Review, then explicitly ask to ingest.
• Resume + prefs: network-jobs profile import ~/resume.pdf, then in your agent run setup from the resume. Interview covers remote/hybrid/onsite (onsite scoped to specific cities, not every office), location, IC vs manager, former employers include/exclude, stage/domain, and other gaps the resume implies.
• Model choice (agnostic): prefer a stronger model for career-page discovery; mid-tier is usually enough for local search and intro drafts.
• Refresh: network-jobs update
• Health: network-jobs doctor
• Semantic matching (optional): network-jobs embed-setup pip-installs fastembed (one-time ~60MB model, uses a local Ollama if already running) so search/rank also score resume↔job similarity. Keyword scoring is unchanged when it's not installed.
• Data: everything under ~/.network-jobs/ — no hosted API.

Pipeline

Resume
    → profile import + setup skill   (profile.json + preferences.json)
LinkedIn ZIP
    → network-jobs-import            (connections + companies graph)
    → careers-discover               (stage openings under triage/)
    → you confirm
    → jobs-ingest                    (promote to corpus/)
    → network-jobs                   (search local corpus using prefs defaults)
    → intro-email-generator          (forwardable warm intro; uses local resume)

Always read SCHEMA.md for paths and JSON shapes.

Quick start

1. Install — npm i -g @hirefrank/network-jobs@latest && network-jobs setup --agent auto

2. Resume

network-jobs profile import ~/Downloads/Resume.pdf

Then in your agent: “set up network jobs from my resume” (profile + prefs, including whether to consider roles at former employers).

3. LinkedIn connections
   Settings → Data privacy → Get a copy of your data → Connections → download ZIP.

network-jobs import ~/Downloads/Connections.zip

4. Discover + ingest in the agent
   “Find open roles at my top 5 companies” → review triage → “ingest the Stripe batch”

5. Search + intro in the agent
   “Find senior PM roles in NYC” → “Draft an intro to …”

Skills

Skill	Role
network-jobs-setup	Resume → profile, preferences interview, LinkedIn export help
network-jobs-import	ZIP → local company graph
careers-discover	Model-driven career page discovery + staging
jobs-ingest	Triage → normalized corpus shards (after you confirm)
network-jobs	Search local corpus (prefs as defaults)
intro-email-generator	Forwardable warm intro emails

CLI

network-jobs setup | update | doctor | which | agents | routing
network-jobs report [--title T] [--dry-run] [--yes]
network-jobs import <zip-or-csv>
network-jobs profile import <resume-file>
network-jobs profile show
network-jobs profile clear [--resume] [--yes]
network-jobs corpus clear [--triage] [--yes]
network-jobs reset [--data] [--purge-cache] [--yes]
network-jobs uninstall [--yes]    # alias for reset --data

# Day-to-day (also available in your agent):
network-jobs search [--query "..."] [-k N] [--company-cap N] [--include-stale] [-v]
network-jobs intros [--k-roles N] [--k-forwarders N] [--title T] [--url U] [--company C]
network-jobs companies [--sort connections|name|crawl]
network-jobs refresh [--company NAME] [--limit N]
network-jobs review-matches --triage-dir DIR [--veto PHRASE] [--json]
network-jobs build-pack <zip-or-csv> [--label NAME]
network-jobs fetch-pack <url> [--as NAME]
network-jobs demo [-k N] [--company-cap N]   # synthetic jobs, no network/disk
network-jobs embed-setup                     # one-time fastembed for semantic search

Skills self-describe for routing. network-jobs routing prints an optional snippet for a project instruction file.

Wipe	Keeps
corpus clear	profile, prefs, resume, LinkedIn graph
profile clear	corpus, LinkedIn graph (--resume also drops resume files)
uninstall / reset --data	nothing under ~/.network-jobs/ — the npm package and the launcher stay installed; both commands print how to remove them

Design notes

• Provenance-style scraping: orchestration + optional recipes — no ATS adapter matrix.
• Shared data: one ~/.network-jobs/ for every agent on the machine.
• Symlink installs: one suite copy; update re-points agent skill links.
• Prefs-aware discovery/search: after extract, match-prefs.py writes index/matches.json (confirm matches by default; ingest all / department slice still available). Search runs rank-jobs.py and reports Showing K of N instead of dumping shards.
• Crawl budget: listing-set hash + lastCrawl on the company; skip unchanged boards; paginate with page/browser caps; ingest is a fingerprint delta. Expiry requires pagination.complete and a status that is not truncated/blocked/captured (truncated can be forced).

Fixtures

export NETWORK_JOBS_HOME=/tmp/nj-test
npx --yes @hirefrank/network-jobs@latest setup --agent auto --no-browser
export PATH="$HOME/.local/bin:$PATH"   # if setup just installed the launcher there
SUITE="$(network-jobs which | sed -n 's/^suite=//p')"   # fixtures ship in the suite
network-jobs import "$SUITE/fixtures/Connections.csv"
python3 "$SUITE/tests/test_pipeline.py"   # prefs, ranker, pagination, fingerprint, classifier, crawl, intros

License

MIT