hunter-tracker

hunter-tracker

The Pipeline Tracker is the memory of the whole search. Every other hunter skill reads hunter/pipeline.json and writes its part back; this skill owns the file's shape and checks it.

${CLAUDE_SKILL_DIR} below is this skill's folder (where this SKILL.md lives).

python3 "${CLAUDE_SKILL_DIR}/pipeline.py" init
python3 "${CLAUDE_SKILL_DIR}/pipeline.py" add-role --company Google --title "Software Engineer" --url <posting URL>
python3 "${CLAUDE_SKILL_DIR}/pipeline.py" move google-software-engineer applied --today 2026-10-23
python3 "${CLAUDE_SKILL_DIR}/pipeline.py" update-role google-software-engineer --next-step "Recruiter screen" --next-date 2026-10-30
python3 "${CLAUDE_SKILL_DIR}/pipeline.py" add-contact --name "Maya C." --company Google --role "Engineering Manager"
python3 "${CLAUDE_SKILL_DIR}/pipeline.py" summary --today 2026-10-16
python3 "${CLAUDE_SKILL_DIR}/pipeline.py" next --today 2026-10-16 --days 7
python3 "${CLAUDE_SKILL_DIR}/pipeline.py" stale --today 2026-10-16 --days 14
python3 "${CLAUDE_SKILL_DIR}/pipeline.py" show google-software-engineer
python3 "${CLAUDE_SKILL_DIR}/pipeline.py" export --what roles --out hunter/pipeline-roles.csv
python3 "${CLAUDE_SKILL_DIR}/pipeline.py" validate

Every command takes --file (default hunter/pipeline.json) and --today YYYY-MM-DD where dates matter, so results are reproducible.

The schema (this skill owns it)

{ "roles":    [ {"id", "company", "title", "url", "stage", "applied", "next_step", "next_date", "notes"} ],
  "contacts": [ {"id", "name", "role", "company", "linkedin", "email", "email_confidence", "source", "notes"} ],
  "threads":  [ {"contact_id", "role_id", "channel",
                 "messages": [ {"date", "direction": "out|in", "text"} ],
                 "status", "next_followup"} ] }

stages: saved, applied, screen, interview, onsite, offer, rejected, withdrawn.
thread status: drafted, sent, replied, booked, declined, closed.
dates: YYYY-MM-DD. A message date may carry a time (2026-10-12T09:30).
ids: roles company-title (google-software-engineer), contacts name-company (maya-c-google); a clash gets -2. Ids from /hunter-scout are kept as they are.
email_confidence: a number from 0 to 1 (a pattern guess from /hunter-connector), or high, medium, low, unknown, or published. published (or verified) is only valid when source is the URL of the public page where the address appears; everything else is a guess.
channel: email, linkedin, phone, in-person or other.
optional fields kept by /hunter-connector (part of the schema): thread invite (true while a LinkedIn invitation is pending; it gets no follow-up), thread draft ({"kind", "subject", "text", "created", "hash"}, a message waiting for the user's yes), thread approval ({"reply", "date", "hash"}, that yes word for word), thread booked ({"when", "minutes", "where"}), and contact do_not_contact (true after a stop request: never contacted again).
a pending draft is never in messages: messages holds only what really went out or came in. (A hand-made file may instead flag a message "draft": true or use status drafted with the draft as its last outgoing message; both are read the same way.)
notes: free text; the tool appends dated lines (2026-10-16: applied -> screen) so the history stays readable. "DO NOT CONTACT" in a contact's notes closes every thread with them for good.
Missing lists are read as empty. Top-level keys starting with _ are comments. Any other extra field is kept and reported as a warning.

validate exits 1 on an error (unknown stage or status, bad date, a thread pointing at a contact or role that does not exist, duplicate ids, an invalid email, a "published" email with no source URL) and lists warnings (no applied date on an applied role, a follow-up date on a closed thread, messages out of order).

Who writes what

skill	writes
`/hunter-scout`	new roles at stage saved, from `hunter/jobs.json`
`/hunter-resume`, `/hunter-cover`	a dated note on the role
`/hunter-apply`	a note that the application is filled and waiting for the user; `move ... applied` once the user submits
`/hunter-connector`	contacts and threads: drafted, sent, replied, booked
`/hunter-followup`	follow-ups it recorded after the user's yes, replies, stop requests
`/hunter-interview`, `/hunter-salary`	next steps and dates

After any skill writes, run validate.

Steps

First time: init, then add what the user already has in flight (roles they applied to, people they talk to). Ask for the applied dates; do not guess them.
Each time something happens, update it right away: move for a stage change, update-role --next-step ... --next-date ... for the next action.
When the user asks where things stand: summary. It shows counts by stage, the next 7 days (overdue first) and stale roles: active, nothing scheduled, no activity in 14 days.
For a spreadsheet: export --what roles (or contacts, threads) and the user opens the CSV in Google Sheets or Excel.

Example: Jordan's pipeline on 2026-10-16

pipeline.example.json is the reel's story: five roles at the hardest companies, Maya C. at Google, the email thread booked for coffee and the LinkedIn request.

Pipeline on 2026-10-16: 5 roles (5 active, 2 applied), 1 contact

By stage:
  saved      3
  applied    1
  screen     1

Conversations: 1 sent, 1 booked

Next 7 days:
  2026-10-19  Amazon, Software Development Engineer II: Online assessment
  2026-10-19  Meta, Software Engineer, Product: Tailor the resume (hunter-resume)
  2026-10-22  Google, Software Engineer: Coffee chat with Maya C. (Google Meet, 10:00)

Stale (no activity and nothing scheduled):
  netflix-senior-software-engineer (Netflix, saved): 22 days idle
  nvidia-deep-learning-software-engineer (NVIDIA, applied): 16 days idle

The Google role is still "saved": the application is filled but not sent, and the coffee chat comes first. That is the point of the reel: an in at the company before the application goes out.

What it never does

Never sends anything or contacts anyone: it is a record.
Never guesses a date, a stage or a reply the user did not report.
Never stores private information about a contact: public professional facts only.

Consent and truth

Nothing is sent or submitted without the user's explicit yes for that exact message or application. Claude drafts, the user approves, then it goes out from the user's own accounts: email as a Gmail draft first; LinkedIn messages through Claude in Chrome with the user present, one at a time, at human pace, the user clicking Send; application forms filled and left for the user to submit.
No bulk or automated messaging, no scraping behind logins, no tricks around a site's limits. LinkedIn's terms forbid bots and automated messaging.
Contact research uses public professional information only (the person's public profile, talks, posts, papers, the company site), never personal life, family, home address or private accounts. Email addresses are guesses from public patterns, marked with a confidence, never called verified unless found published.
Stop rules: stop at the first "no" or any sign of disinterest, max 2 follow-ups, no message after 21 days of silence, and a request to stop is honoured forever.
Never invent experience, titles, degrees, dates or numbers: rewrite and reorder what is true, and ask when something is missing.
Salary benchmarks and interview material cite their sources. Nothing here guarantees an offer or an outcome.