hunter-scout

hunter-scout

Instead of sending thousands of resumes, send a few to the roles that fit. The scout reads the job boards companies publish for exactly this purpose, keeps the listings that match the titles and places you want, and ranks them with a score you can check line by line.

What the user typed after /hunter-scout: $ARGUMENTS

Tool in this folder (standard library Python, no keys, no login):

python3 "${CLAUDE_SKILL_DIR}/jobs.py" --help

What it reads and writes

• Reads hunter/profile.json: target_roles, skills, locations, remote, relocate, exclude_titles, years_experience, salary_floor, dream_companies, work_authorization. If it is missing, copy skills/hunter-apply/profile.example.json (the reel's fictional Jordan Reyes) to hunter/profile.json and fill it in with the user, one true fact at a time.
• Reads boards.json in this folder: 44 companies and the board each one really uses, every token checked live on 2026-10-15. Add any other company with --board "Name=greenhouse:token" (or lever: / ashby:). The token is the last part of the company's job board address, for example job-boards.greenhouse.io/discord is discord.
• Writes hunter/jobs.json: { "as_of", "scanned", "matched", "jobs": [ {"id", "company", "title", "location", "url", "source", "posted", "score", "why"} ] }.
• pick adds chosen listings to hunter/pipeline.json as roles at stage saved, which /hunter-tracker and the other skills read.

Which careers sites it can and cannot read

Company	How	Why
Greenhouse, Lever, Ashby boards (Stripe, Discord, OpenAI, Ramp, Palantir and 34 more)	automatic	public job board APIs published for job seekers and aggregators
NVIDIA	automatic, Workday board	robots.txt allows the career site; the tool never touches nvidia.com, whose terms forbid scrapers
Netflix	automatic, careers search	robots.txt on explore.jobs.netflix.net allows /api/apply
Google	manual	no public feed, and robots.txt disallows crawling paged job results
Meta	manual	Meta's automated data collection terms require written permission
Amazon	manual	amazon.jobs links Amazon's Conditions of Use, which exclude robots and data mining

For a manual company, open its careers search in the user's own browser (jobs.py boards prints the link). Claude in Chrome can read the results page with the user present, one page at a time, and the user can save the search or turn on alerts in their own account. Never loop through pages, never script it.

The fit score

score = title 35 + skills 30 + location 15 + seniority 10 + pay 5 + freshness 5

• Title: 35 when a target role (or a known alias, like Software Development Engineer for Software Engineer) is in the title, 30 when every word is, less for a partial match. Titles under 15 are dropped unless --any-title.
• Skills: the share of the profile's skills found in the posting, full marks at six. Feeds without descriptions (Workday, Netflix) get a neutral 10 and a note to open the posting.
• Location: 15 for a profile location or remote in the user's country, 8 when a move is needed and the user is open to relocating, 0 otherwise.
• Seniority: 10, or 3 when the posting asks for more years than the user has (plus one), or 4 for a Senior title under five years.
• Pay: 5 when the posted range clears salary_floor, 0 when it does not, 2 when no range is posted. Freshness: 5 within a week, 3 within 30 days.

Every point comes with a reason in why. A score ranks listings against the profile; it never predicts an offer. Say so when you show it.

Steps

1. Check hunter/profile.json exists and is true. Ask for anything missing (target titles, places, salary floor) rather than guessing.

2. Search. Default boards are the profile's dream_companies; --company picks others, --all scans every automatic board (about 40 polite calls):

   python3 "${CLAUDE_SKILL_DIR}/jobs.py" search
   python3 "${CLAUDE_SKILL_DIR}/jobs.py" search --company Stripe --company Discord --days 30
   python3 "${CLAUDE_SKILL_DIR}/jobs.py" search --all --remote --min-score 70
   

   Other filters: --title "Backend Engineer", --location "Seattle, WA", --limit 25, --json. Exit 1 means nothing passed the filters.

3. Show the top listings with score, why, the posted date and the link, and the manual route for any dream company without a feed. Open the top few postings and read them before calling one a fit.

4. Ask which ones to pursue, then add exactly those:

   python3 "${CLAUDE_SKILL_DIR}/jobs.py" pick <id> <id>
   

5. Suggest the next skills for each pick: /hunter-research, /hunter-resume, /hunter-linkedin, /hunter-connector.

If a board fails (HTTP 404), the company moved boards: jobs.py check --company Name re-verifies, and the user can pass the new token with --board.

Example: the reel

Jordan Reyes (fictional) is a Software Engineer at Brightline Labs in Chicago with three years of Go, Python and distributed systems, open to relocating. His dream companies are Google, Meta, NVIDIA, Netflix and Amazon. jobs.py search reads NVIDIA's Workday board and Netflix's careers search, then prints the manual route for Google, Meta and Amazon. Jordan searches Google's careers site with Claude in Chrome and finds the Software Engineer role on Google Cloud, Infrastructure in Mountain View. Netflix's "Software Engineer (L6) - Compute Runtime" (remote, United States) scores 75: title match, remote, posted 20 days ago, skills to check on the posting. He picks it and the Google role, and both land in hunter/pipeline.json as saved.

What it never does

• Never logs in anywhere, never reads a site whose robots.txt or terms say no, and never pages through results a site asks crawlers not to page through.
• Never hammers a server: one request per board, at most a few detail calls, a pause between calls, and a User-Agent that says what it is.
• Never invents a listing, a salary or a posting date. Unknown means unknown.
• Never applies. Applying is /hunter-apply, and the user submits.