Playbooks · Everyday tasks · last verified 2026-10-25

Prep for interviews and track outcomes with ai-job-search

For a job seeker × preparing for interviews and tracking what happened × with ai-job-search in Claude Code, after applications are already submitted.

Before you start

You need ai-job-search set up and at least one application already tracked through it — the interview prep builds from the application's archive, so an interview for an application the framework never tracked has nothing to build from. The setup is on our ai-job-search install page. You also need your own experience to draw on: the prep maps questions to your real examples, and it cannot invent any for you. Time and cost: NOT_MEASURED. No outcome is promised: good prep improves an interview; it does not guarantee an offer.

What "done" looks like

At the end of one sitting you have a prep pack for one real scheduled interview — built from the exact posting, CV, and cover letter the interviewer read — a mock interview done, and every application that has resolved recorded in the tracker with its materials archived, so the next round of applications learns from this one.

Steps

When an interview is scheduled, run the prep command. Open Claude Code in the project directory and run /interview for the tracked application. It builds a stage-specific prep pack from the application's archive: the exact posting, the CV and cover letter the interviewer actually read, and feedback recorded from earlier rounds. It then researches the company and the interviewers under a verify-before-use rule, maps likely questions to your STAR examples, and offers a mock interview following the roleplay protocol in the project's interview-prep guide.
Check the prep pack against the real application before you study it. Confirm the posting quoted is the one you applied to, the CV quoted is the tailored version you sent, and the company research is verified rather than plausible. If the pack quotes a generic CV or a different posting, stop — you are preparing for the wrong interview, and the rest of the pack is built on the wrong foundation.
Do the mock interview and fix weak examples with real experience only. Take the offered mock interview. Where an answer is thin, fill it from something you actually did. Gaps get honest bridge answers, never invented experience — the project's own rule, and the only one that survives a follow-up question in a real interview. A polished story about work you never did is worse than an honest bridge, because the interviewer will ask one question further.
After each stage, record what happened. Run /outcome and record the stage: interview stages, offers, rejections, silence. It archives the submitted CV, cover letter, and posting text into the application's folder and keeps the outcome record in the format the setup step can parse later. Recording promptly matters — a stage remembered a month later is a stage misremembered.
Chase quiet applications with drafts, not sends. Run /outcome followup to surface open applications that have gone quiet — ten days by default. It drafts a short follow-up in your writing style, using only claims from the materials you already submitted, and offers a thank-you note in the same turn an interview stage is recorded. Drafts only, never sends, and at most twice per application: after two follow-ups with no reply, the application is done, and further chasing is pestering, not persistence.
Feed the results back into the next round. Once a few applications resolve, return to /setup and calibrate the fit framework from what actually got interviews. The tracker now holds evidence — which postings turned into interviews, which did not — and that evidence is worth more than the scoring guesses you started with.

Check the agent's work

The prep pack quotes the posting and CV you actually submitted, not a generic version of either. A pack built on the wrong documents prepares you for the wrong interview.
Company and interviewer research is verified before use. An unverified fact about the company, repeated confidently in the interview, is a credibility loss you cannot take back.
Every STAR example in the pack is something you actually did. If you cannot tell the story behind an example without notes, it is not your example.
Follow-up drafts contain only claims from materials you already submitted. A new claim in a follow-up is a new fact the employer never saw, and it will be checked.
Tracker updates match what actually happened. A wrong outcome record corrupts the calibration in step 6, which is the whole point of tracking.

What can go wrong

The interview has no tracked application behind it. The prep command builds from the archive. An untracked application means no archive, no tailored pack — track the application first, even by hand, before running the prep.
You already sent two follow-ups. The cap is two per application. A third follow-up does not show persistence; it shows you are not reading the silence. Move on.
An outcome is recorded wrong. A rejection logged as silence, or a stage logged against the wrong company, misleads the next round's calibration. Fix the outcome record by hand as soon as you notice — the tracker is only as honest as its last entry.
The mock interview flatters you. A mock that never pushes on a weak answer is rehearsal, not preparation. If every answer lands cleanly, ask for harder follow-ups on the thinnest example before you call it done.

Limits

Prep is only as good as the archive. A thin application record produces a thin prep pack, and no amount of company research fixes that.
A mock interview is practice, not prediction. The real interviewer will ask something the pack did not map, and the bridge-answer habit from step 3 is what carries that moment.
Follow-ups are drafts. Nothing here sends email on your behalf, and nothing here applies to jobs for you — the framework never submits an application without you, and this playbook does not change that.

Verification

Checked 2026-10-25. This page was not run end to end; its claims rest on the author's own documentation fetched this run. Sources: the ai-job-search README cloned this run (the /interview prep pack: stage-specific, archive-based, verify-before-use research, STAR mapping, mock interview; the /outcome command: stage recording, archive layout, outcome record format, /outcome followup quiet-application default of ten days, drafts-only rule, twice-per-application cap, thank-you note, and /setup recalibration). Omitted: the framework's Gmail sync and HTML report commands exist but are not covered here — they are separate workflows, not steps in this one. No claim is made that any interview outcome follows from using these commands.

Related skills: ai-job-search-dev/ai-job-search

Back to: Everyday playbooks hub · Skills index