hunter-apply

Most of an application is the same twenty questions. This skill answers those from the user's own saved answers, asks about the rest, fills the form in their browser, and stops with the Submit button untouched. The user reads it, fixes anything, and submits it themselves.

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

Tool in this folder (standard library Python; it never submits anything):

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

What it reads and writes

• hunter/profile.json: the user's facts and reusable answers. The shape every hunter skill reads is profile.example.json in this folder (the reel's fictional Jordan Reyes). Anything left null is asked, never guessed.
• hunter/applications/<role-id>.json: the answer plan for one form.
• hunter/pipeline.json: the role gets "application": {"status": "draft", ...}, next step "Review the application draft and submit it yourself"; after the user submits, stage applied and the date.
• Reads hunter/resumes/<role-id>.md, hunter/covers/<role-id>.md and hunter/companies/<company>.md for the checklist.

The rules

• Only the user's truth. Every filled answer comes from the profile or a document the user approved. Nothing is invented: no experience, title, degree, date, number, referral or reason.
• Some answers are the user's alone. Voluntary self-identification (gender, race or ethnicity, veteran, disability) is answered only with the user's own saved choice, or left for them. Salary history is the user's call (several US states and cities bar employers from asking): never estimate it. A salary expectation comes from the user, informed by /hunter-salary.
• No accounts, no passwords, no CAPTCHAs. If a site needs a sign-in (Workday often does), the user signs in. Claude never creates an account, types a password or solves a CAPTCHA.
• Stop before Submit. Never click Submit, Apply or Send on a form.

Steps

1. Checklist first:

   python3 "${CLAUDE_SKILL_DIR}/answers.py" checklist --role <role-id>

   Tailored resume (/hunter-resume), cover letter if the form takes one (/hunter-cover), research brief (/hunter-research), work authorization in the profile, and ideally a warm intro (/hunter-connector).

2. With the user present, open the application in their browser with Claude in Chrome and read the form. Write its questions, as labelled, to a JSON list (with options for selects and required where marked).

3. Map them:

   python3 "${CLAUDE_SKILL_DIR}/answers.py" fill --questions form.json --role <role-id> --out hunter/applications/<role-id>.json
   python3 "${CLAUDE_SKILL_DIR}/answers.py" answer --question "Will you now or in the future require sponsorship?"

   fill rows go in as they are. ASK rows: ask the user, and offer to save the answer to hunter/profile.json for next time. USER rows: the user answers them. ROLE rows: the approved cover letter or a short answer written for this role from the research brief.

4. Fill the form field by field at a human pace. Attach the tailored resume the user picks. Review every page with the user.

5. Stop before Submit. If the site offers "save for later" and the user is signed in, use it only with their yes. Record the draft:

   python3 "${CLAUDE_SKILL_DIR}/answers.py" draft --role <role-id> --form-url <url> --answers hunter/applications/<role-id>.json

6. The user reviews and clicks Submit. When they say they did:

   python3 "${CLAUDE_SKILL_DIR}/answers.py" submitted --role <role-id> --confirm "I submitted it"

   The role moves to applied, with a follow-up date a week out for /hunter-followup.

Example: the reel

Jordan Reyes (fictional) applies for the Software Engineer role at Google after his coffee chat with Maya C. is booked. The checklist shows the tailored resume, the cover letter and the Google brief are done. On the form, fill answers 9 questions from his profile (name, email, LinkedIn, "authorized to work in the United States": Yes, "require sponsorship": No, relocation: Yes, never worked at Google: No), asks him for his phone number, his salary expectation and a non-compete question it cannot know, leaves gender, veteran status and current salary to him, and marks "Why Google?" for the cover letter. Claude fills the form in his browser and stops. The pipeline shows "Draft · not sent" until Jordan reads it and clicks Submit himself.

What it never does

• Never submits an application, clicks Submit or Apply, or sends a form.
• Never invents an answer, a referral, a salary history or a demographic answer.
• Never creates an account, types a password or solves a CAPTCHA.
• Never applies in bulk or on a schedule: one application at a time, with the user there.

Consent

• Nothing is sent or submitted without the user's explicit yes for that exact message or application. Claude fills the form and leaves it for the user to submit from their own browser and their own account.
• No bulk or automated applying, no scraping behind logins, no tricks around a site's limits or CAPTCHAs.
• The user's personal data goes only into the form they chose to fill, and voluntary questions stay theirs to answer.
• Outreach that comes with an application follows the connector's stop rules: stop at the first no, at most 2 follow-ups, nothing after 21 days of silence, and a stop request is honoured forever.