---

name: "job-campaign-machine-pack" version: 0.1.0 description: "Playbook for running a high-volume autonomous job application campaign."

Job Campaign Machine Pack (opinionated, free)

What

The operating playbook for running an aggressive, high-volume job application campaign end to end: discovery, vetting, tailoring, submission, and counting — with the agent doing everything except the steps that genuinely need the human. Built to convert volume into interviews without burning the candidate's reputation or the agent's credibility. An agent reading this can run the campaign from tonight onward.

When to use

• Setting up or operating a job-search campaign for a professional candidate (medical, technical, executive).
• Auditing an existing campaign: run the gates to find where sends are leaking, duplicating, or going to dead listings.

How

1. Standing authorization (agree it once, in writing)

• Get blanket authorization in the candidate's own words: what to apply to, whether tailoring is allowed, whether review is needed before submit. Without it, every application is a new permission question and the campaign never reaches volume.
• The human touches only the manual gates: CAPTCHAs (hand over the page with full briefing), account passwords, attestations, and decisions. Everything else is the agent's job.

2. Discovery and vetting

• Official keyless JSON endpoints first. ATS vendors (Greenhouse, Lever, Ashby and more) expose structured job feeds — scrape those before touching JS-heavy career pages. Structured beats scraped every time.
• Vet before you write. Disqualify listings that require an active license the candidate does not hold, on-site presence in a non-commutable state, or anything outside the agreed target list. A disqualification needs one written reason; a qualified lead gets the full pipeline.
• Dead listings die fast. If the employer-side portal says "no longer available," the aggregator listing is stale — close it, do not count it, do not retry it.

3. Application craft

• Tailor the cover letter to the role. One template with the job's real nouns swapped in — never a generic blast. Mirror the posting's language for the title, the top three requirements, and the outcome the team is hiring for.
• Write in the candidate's first-person voice. No assistant self-identification in any field, email, or form. The candidate is the applicant; the agent is invisible.
• Opt yes on AI screening questions. Answer demographic/EEO questions once and consistently (the candidate's answers, not the agent's).
• Never claim credentials the candidate does not hold. No active license claims, no invented experience. One false credential torpedoes the whole campaign.

4. Submission gates

• Every application ends in a verified send: a confirmation page, a confirmation email, or a reference number — witnessed, not assumed.
• A failed submit is retried only after proving the original did not go through. A duplicate application to the same employer is worse than a missed one.
• If a submission needs something only the human has (a code, a document, a decision), hand it over with the full briefing — role, employer, what is already filled, exactly what the click does — and stop. Never leave the handoff half-briefed.

5. Counting (the discipline that makes volume trustworthy)

• The verified-send tally is mechanical. Re-run the count script every time you quote it. Inbound emails (rejections, thank-yous) are never new sends. Yesterday's number is never today's number.
• Keep a per-role ledger: employer, title, listing source, date, outcome witness. The ledger is the campaign's memory — the alternative is the agent's memory, which drifts.

6. The daily rhythm

• Discovery and vetting run continuously; submissions batch through the day; the count is quoted fresh in every status update with real numbers, never reassurance.
• Every rejection is data: which roles reject, which titles convert, which sources are stale. Feed it back into vetting weekly.

References

• ATS JSON endpoint pattern: vendor career-page APIs return structured postings without authentication — query them directly before any browser scraping.
• Campaign ledger: one file, one row per application, witnessed outcomes only.