It runs the search.
It never presses send.

One install gives your agent eight modes for the part of a job search around the resume: a shortlist deduped against your brief, fit scores with the JD line behind every number, one plain tracker file you can open and audit, drafts for the roles that went quiet, briefings before an interview, and a weekly read of the funnel. It stops at the send button, and a row only becomes submitted when you say it did.

8 modes 9 ground rules · 0 submits ~385 lines of instruction 0 API keys · 0 telemetry

for coding agents · free, open, no auth

install

Copy one line. Get a search you can actually review.

Detects Claude Code, Cursor, Codex, Qoder and more, then writes the skill into the right folder. Add --dry-run to see the plan first.

npx skills add arm-pz/application-pilot --skill application-pilot

the vocabulary

Eight modes. One hard stop.

Or skip the modes. Hand it a resume and say "run my job search", and it confirms the brief first, then works the loop and stops before any material writing.

The desk 8

Markdown and CSV only, sized for a search one person can review: roughly 40 tracked rows and a 25-row shortlist.

  • 1 search brief
  • 2 discover
  • 3 score fit
  • 4 packet
  • 5 track
  • 6 follow-up
  • 7 briefing
  • 8 funnel

The hand-offs 2

Skills sharing one subject must not share triggers. Writing belongs to applicant-kit, judging to recruiter-lens, and this skill turns their verdicts into next actions without re-judging them.

  • applicant-kit produces
  • recruiter-lens judges

What it will not do by design

No CLI, no cloud state, no telemetry, no browser automation, no submit path, and it does not acquire those to be more convenient. Ask for "fill and submit these 40" and it hands off rather than half-implementing automation. The workflow shape comes from a public prompt thread by @HeyAirahAi on X (October 2026); its central move, sending hundreds of applications overnight, is deliberately not implemented here.

  • never submits
  • never invents
  • no fake recency
  • no status without evidence
  • rubric, not an ATS
  • volume is not the lever
  • respects site access rules
  • your data stays local
  • reviewable files only

house rules

Because this runs under your name.

Nine ground rules ship with the skill. Three of them are the reason it stops where it does.

It prepares, you dispatch

The final materials and the exact target are shown, then it waits for your approval of that specific action. No batch is ever opened, filled or sent unattended, and the tracker records a submission only after you say it happened.

Nothing is filled in

No invented experience, dates, metrics, recruiter names, application statuses or salary figures. A follow-up that depends on a fact not in evidence keeps the slot bracketed, [recruiter name?], and says the draft is not ready to send.

The spray is not the plan

Automated bulk applying breaks LinkedIn's and Indeed's terms, both of which run detection that suspends accounts, and employers now filter generated applications. A reviewed shortlist with specific material outperforms a spray at the same hour of effort.

Run the search like someone will read it.

Free, open, and it runs entirely inside the agent you already use.

Install application-pilot Read the source