/interview-prep

Usage: /interview-prep <job folder or company + role> [interview date/type] (or just ask in plain English).

Job: $ARGUMENTS.

1. Read the job folder (posting, fit, resume, application answers), profile/story-bank.md, knowledge/interview-questions.md (question bank by competency), knowledge/field/role-questions.md (role-specific technical questions; if it still says STATUS: not generated, look up real interview reports for the role instead), knowledge/field/field-primer.md, knowledge/field/background.md, and references/formats-and-hard-questions.md (what interviewers score, each format, hard questions, salary wording, sources); for a take-home, also references/take-home-assignments.md, and for an accommodation, references/accommodations-and-disclosure.md. If the user only has a take-home or an accommodation question, handle just that part (the matching bullet in step 4 and step 6), not a whole brief. If a knowledge file still says STATUS: not generated, do a quick web lookup for what you need instead and offer to build it.
2. Ask the user one question at a time, skipping anything they already said: (a) when it is (date, time, time zone); (b) what kind of interview it is, in plain words (for example "a phone call with a recruiter, a video call, a recorded video, an onsite day, or something else?"; you'll sort it into recruiter phone screen, video, one-way recorded video, AI interviewer, onsite, panel, site or office tour, technical presentation, case or skills exercise, or take-home assignment); (c) who they'll meet, names and titles if known; then, on its own, (d) the accommodation offer, once, neutrally, as something anyone can ask for: "Anything that would make the format work better for you, like questions in writing or extra time? No reason needed, and it's totally optional." Never ask why.
3. Run the company-researcher agent: what they make or do at this site or team, recent news (12 months), anything public a candidate in the user's field should know (e.g., a regulated industry may have public enforcement or recall records), competitors, culture signals, the team's likely work. Look up interviewers only through public, professional info (company pages, publications, talks). Don't browse LinkedIn for the user. If a company-data app is connected, look up recent deals, investors and competitors yourself and pass them to the agent with source and date (knowledge/data-connectors.md, which also covers a draft deck for a presentation round).
4. Write jobs/<folder>/interview-prep.md:
   • Company in 60 seconds: 5 bullets the user can say out loud.
   • Why this company / why this role: their real reasons, drafted from their preferences and fit; ask them to confirm.
   • Walk me through your resume: a 90-second version tuned to this role.
   • Likely questions: 8 behavioral (each mapped to a story-bank story; flag any with no good story) + 6–10 technical or role-specific ones from the field primer and the posting's stated requirements, picked by role type. Interviewers usually score the posting's skills on a rubric and then probe, so under each story note answers to: "What was your role?", "What was the result?", "What would you do differently?"
   • Their likely weak spots for this job (from fit.md gaps) and an honest way to answer each. If the resume shows a gap, layoff, firing or career change, draft a short answer from the user's own facts (wording in the reference, section 3) and have them say it out loud until it's easy.
   • Questions to ask them: 6 specific ones, matched to who they're meeting (recruiter: process and range; hiring manager: "How will you measure success in this role?", "What are the biggest challenges someone here will face?"; peers: a typical week). Always end with "What are the next steps and the timeline?" (the follow-up date depends on it).
   • Recruiter phone screen: a filter (must-haves, work authorization, location, start date, pay). Prepare the 90-second pitch, availability, and the salary answer.
   • Salary question: look for the pay range in the posting, and check whether the job's location requires employers to post ranges or bans asking about salary history (look up the current rules; they change). Where salary-history questions are banned, the user never has to share past pay. If a range is posted, don't ask for it; first compare it with the user's salary floor in preferences (same currency and period; any conversion is an estimate). Whole range at or above the floor: "The posted range of $X–$Y works for me; given <one real strength from the master resume> I'd hope to be in the upper half, but I'm looking at the whole package." Floor inside the range: "Given <strength>, I'd be looking at the upper part of that range, around <a number at or above the floor>." Whole range below the floor: never say it works; tell the user plainly before the call, and script a question instead: "The posted range is a bit below what I'm looking for. Is there flexibility for someone with <strength>?" (or help them decide whether to go ahead at all). Floor not set: use the market range below as the check. If no range is posted: "I'd love to hear the range you have budgeted for this role first," and if pressed, give a researched range: the "Market pay" block in fit.md if it's recent, otherwise build one per .claude/skills/offer/references/pay-data.md and save it there. Base the user's number on that range and their floor in preferences, never a guess. Put it in the brief, with the sources so they can say where it comes from. If pressed, they give a range (bottom at or above their floor), never one number.
   • One-way video (HireVue-style): usually 3–6 questions, about 30 seconds to think and 1–3 minutes each, often one retake or none. HireVue says its scoring reads only a transcript of what's said, so a complete story with a clear result matters more than polish. Prepare "why this company / why this role" (plus "which rotations" for rotational programs) and 3 STAR stories under 2 minutes, check camera and lighting, and offer /mock-interview <job> one-way.
   • AI interviewer (a bot by phone, video or chat): treat it as a structured interview: answer each question fully in one turn, with an example and a number; speak clearly; never read AI-written answers. If the user lives in Illinois or New York City, or has a disability, tell them their notice and accommodation rights (reference, section 2). Offer /mock-interview <job> ai-screen.
   • If the format includes a technical presentation: pick one project, write a 7-slide outline timed to finish about a minute under the slot they gave (about 9 minutes for a 10-minute slot) (problem, context, approach, data, result, what I'd change, Q&A), scrub confidential material from past employers (no customer, client or product names, normalized charts; check with the former manager if unsure), list 6 likely probes with answers, and bring backups (USB, emailed copy, printed handouts). Offer /mock-interview <job> presentation.
   • Site or office tour: follow any safety or dress rules exactly (e.g., PPE on a factory floor or lab), don't touch equipment, stay with the host, ask one or two questions about how the work actually runs, and note one real observation to mention in the thank-you.
   • References: before a final-round or onsite interview, make sure 2–3 references are lined up in profile/references.md (former managers, or for students, co-op/internship supervisors or professors), with their permission, and offer to draft a heads-up note with the job description into outbox/.
   • Take-home assignment (full guide: references/take-home-assignments.md): first pin down scope, time limit, AI-use rules, due date, and who owns the work; ask the recruiter for whatever's missing with a short scope question drafted into outbox/ (draft only). Suggest a time box (commonly 2–4 hours; the employer's stated time if given) and a short "what I'd do with more time" note. Flag red flags plainly: it looks like real backlog or customer work, uses real data, is unpaid multi-day work, has a vague spec, or has no interview planned around it. The user keeps their own copy in jobs/<folder>/take-home/ and, where the work could be reused, adds a "sample for evaluation only" line. The deliverable must be the user's own work under the employer's AI rules: if AI isn't explicitly allowed, you coach only (clarify, plan, explain, rehearse the walkthrough), and never write or solve the deliverable. Prep them to present and defend it.
   • Accommodations (full guide: references/accommodations-and-disclosure.md): if the user wants one, disclosure is their choice and timing, and no diagnosis is needed. Draft a short, plain request into outbox/ (draft only) naming the specific adjustment (questions in advance, extra time, written format, breaks, captioning, a quiet room, a human instead of an AI interviewer), sent early. Mention their local rights only from the reference's country notes (US: ADA, askjan.org; UK: Equality Act, Access to Work; elsewhere: check local law). Store only what they choose; never store a diagnosis unless they ask. Offer a /mock-interview set up the same way.
   • Integrity and ID checks: many employers now verify identity on camera, and some forbid AI help during interviews and assessments. Prepare the user to do the live interview on their own, without AI, and to have photo ID ready if a verified employer asks. Never send ID numbers by email before an offer.
   • If the user is employed: suggest early, late or lunchtime slots and batching rounds, a personal device and a private place for calls, a short forward-looking answer to "why are you leaving?", and references who aren't the current boss (knowledge/searching-while-employed.md §2-3). Hourly, academic or contract roles: the matching file in the playbook's "Special situations" covers their interview formats (open-hiring days, academic job talks and campus visits, agency then client calls).
   • Logistics: time zone, link/address, what to wear (match the company and any tour rules), what to bring (printed resume, notebook, questions), 15-minute-early plan.
5. Run truth-checker, then voice-editor, on the first-person parts of interview-prep.md (the pitch, "why this company / role", weak-spot answers, presentation outline); re-check anything changed. Details still marked [ASK] stay out of answers until the user confirms them. Then offer a printable one-page brief (/pages brief <folder>) and flashcards for this job (/pages flashcards).
6. Update the tracker only if an employer has actually scheduled an interview (the user said so, or an email shows it): for a recruiter/HR phone or video screen or a one-way video screen, status=screening; for a hiring-manager, technical, panel or onsite interview, status=interviewing. Either way next_action=Interview <date> (<type>) and next_action_date=<date>, and log it in history/activity.md. If the user is just preparing ahead, leave the status alone. Take-home received: next_action=Take-home due <YYYY-MM-DD> (the real due date) and next_action_date = the day to start (two days before the due date by default, or today if sooner); keep the status (if it's still applied, set screening); log it. Two dated commitments on one job (an interview booked and a take-home due, or two interviews): follow the playbook's "Two commitments at once" rule (Also: token at the start of notes); never overwrite one with the other. Offer /mock-interview with this job.