Greenhouse

Promptfree, for everyone, and the only version there is

Copy the prompt for Claude CodeCopy the prompt for CodexCopy the prompt for Cursor

Copy

Build the applicant tracking I actually need instead of Greenhouse.

This is a database of people moving through stages, and the reason to build it carefully is legal as much as practical: **hiring records are evidence.** Design it as an archive first.

STACK
- Node 20+ with Fastify, server-rendered HTML with a little vanilla JS
- SQLite through better-sqlite3, WAL mode, FTS5 over CVs
- Caddy in front

THE DATA MODEL
- jobs: id, title, department, location, description_md, status, opened_at, closed_at
- candidates: id, name, email, phone, source, created_at
- applications: id, job_id, candidate_id, stage, status, applied_at, rejected_reason
- stage_changes: id, application_id, from_stage, to_stage, actor, at — **append-only, and this is the table that answers 'why was this person rejected'**
- interviews: id, application_id, kit_id, interviewer_id, scheduled_for_utc, held_at
- scorecards: id, interview_id, attribute, rating, note, submitted_at — submitted once and never edited
- attachments: id, application_id, kind, path, sha256

STRUCTURED INTERVIEWING, WHICH IS THE WHOLE VALUE
- A kit per stage: the attributes being assessed, the questions to ask, and the rating scale. Written before the first interview, not after
- **An interviewer cannot see other scorecards until they submit their own.** Enforce it on the server. Without this, the second interviewer agrees with the first and you have one opinion recorded four times
- Ratings on a fixed scale with a written definition of each point, and a required note. A rating without a note is a feeling
- A hiring decision recorded with the reason, referencing the scorecards

APPLYING, WHICH MUST NOT BE HOSTILE
- One page, a CV upload, and the fields you genuinely need. Every extra field costs applicants, and the ones it costs most are the ones with the least free time
- Acknowledge immediately, and **reply to everybody, including rejections.** An unanswered application is the most common complaint about every hiring process and it is free to fix
- No account to apply. A token link to check status

THE LEGAL SIDE, WHICH YOU MUST NOT SKIP
- Store what you need and no more. Do not collect date of birth, marital status, or a photograph
- Demographic data, if collected at all, is separate, optional, and **never visible to anybody making a decision** — a different table with different access, and it is only ever read in aggregate
- Retention with an actual sweeper: candidate data deleted on a stated schedule after the process ends, unless they consented to be kept, and consent recorded with a date
- A subject access export per candidate, because they can ask, and a deletion that reaches the attachments

COORDINATION
- Free-slot lookup across interviewers, every instant in UTC and working hours as local times with a named zone
- Reminders to interviewers with the kit attached and the scorecard link
- A list of scorecards not yet submitted, which is the only nagging this system needs to do

WHAT MATTERS MOST
Blind scorecards and a real retention sweeper. The first makes the hiring better; the second keeps you on the right side of the law.