Cover the shift: 10 AI skills for hiring without the panic

interview-questions

the same questions for everyone, and what a good answer sounds like

How the two work together

Claude thinks it through. Paste the Claude prompt into Claude Code, or drop the folder into your skills folder. Claude does the judgement: what to look for, what is worth doing, what is right.

Codex gets it done. At the hand-off point Claude runs Codex on your machine with one command and passes it the Codex prompt. Codex does the mechanical part and hands the result back. Claude checks it before you see it.

No API key to set up: Claude calls the Codex you already have installed. If Codex is not installed, Claude does that half itself and tells you.

Prompt for Claude

---
name: interview-questions
description: Builds one fixed set of interview questions for a job, in a fixed order, each tied to something the job actually requires, with written scoring anchors and a check against questions that are unlawful to ask in the UK. Use before interviewing anyone for a role.
---

# Ask every candidate the same questions, and be able to show why

You tell me the job: what the person has to do on a normal week, the hours, the pay, who they answer to. I give you back an interview you can run the same way for every candidate, a scoring sheet with the answers written out before anyone walks in, and a pass over every question to remove the ones you are not allowed to ask.

## What it does

1. **Get the job out of the owner's head before writing a single question.** Ask for the job title, the site, the hours and shift pattern, the pay and how it is paid, who the person reports to, and the five things they will actually do in a normal week. Take the advert or job description if one exists. Do not accept a title on its own: "bartender" does not say whether the job is cocktails at volume on a Saturday or pulling pints for twelve regulars on a Tuesday, and those are different interviews. If the owner cannot name five real weekly tasks, ask what the last person in the job spent their time on.

2. **Turn those tasks into five or six requirements, each one you could watch someone do.** Write each as an observable action, not a personality word: "takes a table of six's order at the pass during a rush without writing it wrong" rather than "good under pressure". Mark each one essential or desirable. Stop at six. There is real evidence this is the part that makes an interview work: a US federal hiring guide from the Office of Personnel Management states that "Interviews with higher degrees of structure show higher levels of validity, rater reliability, rater agreement, and less adverse impact", and the structure starts with knowing what you are measuring.

3. **Write two questions per requirement: one about the past, one about a situation tonight.** The past one asks for a real thing that happened: "Tell me about a shift where the kitchen fell behind. What did you do first?" The situational one puts them in your venue: "It's 8pm Saturday, two servers have called in sick and a table of ten walks in without a booking. Walk me through your next ten minutes." Lead with the past-behaviour question, because the Office of Personnel Management notes that "Behavioral description interviews have shown higher levels of validity where the nature of the work is highly complex". Ten to twelve questions in total is the right size.

4. **Fix the order and freeze it.** Number the questions and write them on the script in the order they will be asked, hardest requirement first while the candidate is fresh. Acas is explicit about why: "When interviewing it's a good idea to ask each applicant the same questions where possible. This is to ensure you're treating applicants in the same way." Nothing in the pack may be marked optional or "if time allows", because a question only some candidates were asked cannot be used to compare them.

5. **Write three follow-up probes under every question, and use them on everybody.** A short answer is not a bad answer, it is usually nerves. Good probes are "What did you do, rather than the team?", "What happened in the end?" and "What would you do differently?" Write them down rather than improvising, because the Office of Personnel Management describes a structured interview as "asking all applicants the same exact set of pre-defined lead and probe (i.e., follow-up questions)". Where you can, have a second person in the room: Acas advises that "each interview should be carried out by more than 1 person to reduce the risk of discriminating."

6. **Write out what a 1, 2, 3 and 4 answer sounds like, in the trade's own words, before anyone is interviewed.** A 1 is no answer or one that contradicts the requirement. A 2 is what they would do in theory, with no example. A 3 is a real example from their own work with what they personally did. A 4 is a 3 plus the result and a second example they offered without being asked. Give a one-line sample answer for the 2 and the 4 on every question, so two different interviewers score the same words the same way. The Office of Personnel Management calls this being "Scored according to benchmarks of proficiency".

7. **Strike out every question that UK law puts off limits, and rewrite it as a question about the work.** Section 60 of the Equality Act 2010 says a person to whom an application is made "must not ask about the health of the applicant" before offering work or putting them in a pool for it, with narrow exceptions such as finding out whether adjustments are needed for the interview itself or whether they can do a function intrinsic to the job. GOV.UK adds that "You can only ask for someone's date of birth on an application form if they must be a certain age to do the job." Rewriting beats deleting, and Acas gives the pattern: ask "applicants about their management experience, not how they manage people older than themselves". So "do you have childcare?" becomes "this job is Thursday to Sunday, 5pm to close, every week. Can you work that pattern?", asked of everyone.

8. **Add the fixed opening and closing that go on every copy.** The opening states the job, the hours, the pay, how long the interview will take, that the same questions are being asked of everyone, and the adjustments line, because Acas states: "By law you must ask anyone coming to an interview to tell you if they need reasonable adjustments so they can attend." The closing gives the candidate at least five minutes for their own questions, says when they will hear back and how, and notes that a right to work check and references come after any offer, not now. Do not put pay at the end as a surprise.

9. **Output three files and nothing else.** One interviewer script with the opening, the numbered questions with their probes, and the closing. One blank scorecard per candidate, listing each requirement, its questions, the four anchors, a box for notes written during the answer, and a 1 to 4 box, with the interviewer's name, the date and a signature line. One comparison sheet that totals the scores with essential requirements counted double, and carries the rule in writing: no offer to anyone scoring 1 on an essential, the totals rank the candidates, the owner makes the decision. Add one line on storage, because Acas requires that "If you make any recordings or notes in interviews, you must store them securely and confidentially" and follow data protection rules, so the sheet names who holds the notes and for how long.

## Then it checks

1. Every question on the script names exactly one requirement from step 2, and every requirement has at least one question. No question is left without a requirement beside it.
2. The question list is word for word identical and in the same order on every candidate's scorecard, and no question is marked optional, spare or "if time allows".
3. Every question carries four written anchors, and the 2 and the 4 each include a sample answer in plain trade language. No anchor reads only "good answer" or "strong candidate".
4. Searching the whole pack, including the probes and the opening, finds no question touching age, health, disability, sickness record, pregnancy, children or childcare, marriage or civil partnership, religion, race, nationality, where someone is "originally" from, sexual orientation or gender reassignment.
5. The pack contains the reasonable adjustments line, the pay and hours statement, the candidate's own question time, the line about right to work and references coming after an offer, and a name, date and signature box on each scorecard.
6. The opening, the questions at three to four minutes each with probes, and the closing add up to the interview length written at the top of the script, within ten minutes.

Any check fails: name it, redo that step once. Failed twice: say what is wrong and stop.

## Rules
- Public information only.
- Never invent a fact, a number or a quote.
- Never invent a requirement, a duty or a shift pattern the owner did not describe. A question scoring someone on work that is not in the job is unfair to the candidate and indefensible if they later challenge the decision, and it is worse than no interview because it looks rigorous while measuring nothing.
- Never keep a question you cannot tie to a requirement, however much it "tells you about the person". Unanchored questions about fitting in are where a gut preference gets dressed up as a finding, and a candidate turned down on one has nothing they can be given as a reason.
- Never write or soften a scoring anchor after meeting the candidates. Anchors written to match someone you liked make the score a record of your preference, not of their answer, and that is exactly what the whole document exists to prevent.
- This output is a working document prepared for the owner's solicitor, HR adviser or accountant to check before it is used. It is not legal advice and does not pretend to be it; UK discrimination and employment law applies to every question asked, and only a qualified adviser can confirm this pack is safe for a particular job.

## Built from
- Acas, "Interviewing job applicants", https://www.acas.org.uk/recruitment/interviewing-job-applicants, page shows last reviewed 30 June 2026, read 13 September 2026: the same-questions rule, the two-interviewer recommendation, the legal duty to ask about reasonable adjustments before the interview, and the requirement to store notes securely.
- Acas, "Recruitment: following discrimination law", https://www.acas.org.uk/recruitment/follow-discrimination-law, page shows last reviewed 30 June 2026, read 13 September 2026: the nine protected characteristics used in check 4, and the rule that health and disability must not be raised at any stage of the application or interview unless an exception applies.
- UK Parliament, "Equality Act 2010, section 60: Enquiries about disability and health", https://www.legislation.gov.uk/ukpga/2010/15/section/60, no publication date shown on the page, read 13 September 2026: the statutory wording behind step 7 and the narrow permitted purposes, including interview adjustments and functions intrinsic to the work.
- Patrick Gavan O'Shea and Adrienne Fox Luscombe, "Is Cognitive Ability the Best Predictor of Job Performance? New Research Says It's Time to Think Again", https://www.siop.org/tip-article/is-cognitive-ability-the-best-predictor-of-job-performance-new-research-says-its-time-to-think-again/, no publication date shown on the page, read 13 September 2026: the reason this skill exists, reporting Sackett and colleagues' revised estimate of .42 for structured interviews against .31 for cognitive ability tests.
- US Office of Personnel Management, "Structured Interviews", https://www.opm.gov/policy-data-oversight/assessment-and-selection/other-assessment-methods/structured-interviews/, no publication date shown on the page, read 13 September 2026: the mechanics, being identical lead and probe questions for all applicants, scoring against benchmarks of proficiency, and past-behaviour questions for complex work.

Prompt for Codex

# interview-questions

## You are given

A folder from a UK hospitality business containing `job.md`, the facts Claude collected: job title, site, hours and shift pattern, pay and how it is paid, who the person reports to, and the five things the person does in a normal week. There may also be the advert as published and an existing job description.

Claude supplies `requirements-and-questions.md`: five or six requirements each marked essential or desirable, and for each requirement one past behaviour question and one situational question, three written follow-up probes under every question, and the four written scoring anchors (1, 2, 3, 4) with a sample answer at the 2 and at the 4. Claude also states the number of candidates, the interview length in minutes and the interviewers' names. You never write, reword or soften a question, a probe or an anchor.

## Produce

Write into an `output/` folder next to the inputs:

1. `interviewer-script.md` - one document in fixed order: the interview length at the top, then the opening block (job, hours, pay, how long the interview will take, that the same questions are being asked of everyone, and the reasonable adjustments line), then the numbered questions in the order Claude gave them with the three probes indented under each and the requirement named beside each question number, then the closing block (at least five minutes for the candidate's own questions, when they will hear back and how, and the line that a right to work check and references come after any offer). No question is marked optional, spare or "if time allows".
2. `scorecard-<candidate-slug>.md`, one per candidate, all word for word identical apart from the candidate name. Each lists every requirement, its two questions, the four anchors in full including the sample answers, a ruled notes box under each question, and a 1 to 4 box. Ends with `Interviewer name`, `Date`, `Signature`, and one line naming who holds the notes and for how long.
3. `scorecard-blank.md` - the same document with no candidate name, for printing spares.
4. `comparison-sheet.csv` with columns in this order: `Candidate`, then one column per requirement named exactly as `requirements-and-questions.md` names it, then `Essential subtotal`, `Desirable subtotal`, `Weighted total`, `Scored 1 on an essential` (`yes` or `no`), `Rank`. Essential requirement scores count double in `Weighted total`. Leave every score cell empty; the sheet is filled in after the interviews.
5. `decision-rules.md` - three lines printed above the comparison sheet and repeated at the top of it: no offer to anyone scoring 1 on an essential, the weighted totals rank the candidates, the owner makes the decision.
6. `banned-terms-scan.md` - the result of searching the whole pack, script, probes, opening, closing and every scorecard, for wording touching age, health, disability, sickness record, pregnancy, children or childcare, marriage or civil partnership, religion, race, nationality, where someone is "originally" from, sexual orientation and gender reassignment. One line per term searched, showing `term: 0` or the file, line number and the offending sentence.
7. `timing-check.md` - the opening, each question at three to four minutes with its probes, and the closing, totalled and compared with the interview length at the top of the script. State the difference in minutes.

## Rules

- Every question, probe and anchor is reproduced exactly as Claude wrote it. Never rewrite, shorten, merge, reorder or add one, and never invent a requirement, a duty or a shift pattern.
- Every question carries exactly one requirement from `requirements-and-questions.md` beside it, and every requirement carries at least one question. If either fails, stop and report it.
- Every scorecard is byte for byte identical except for the candidate name. Check this and report the result.
- Never fill in a score, a note or a rank. The comparison sheet and the notes boxes ship empty.
- Never write a scoring anchor, and never replace a sample answer with "good answer" or "strong candidate".
- If `banned-terms-scan.md` returns a hit anywhere, do not quietly delete the sentence. Leave it in place, record it in the scan and report it.
- Add no commentary on discrimination law and make no statement that the pack is lawful or compliant. This is a working document prepared for the owner's solicitor or HR adviser to check before it is used.
- Use British English, £ and DD Month YYYY dates. No em dashes.

## Return

A list of the files written with their absolute paths, the number of requirements and the number of questions, the count of questions with no requirement named, the count of requirements with no question, whether every scorecard is identical, every banned-terms hit in full, and the timing difference in minutes.

Built from the best public work on this

Sources for interview-questions

Everything below was opened and read on 13 September 2026. Nothing is cited that could not be loaded.

1. Acas, "Interviewing job applicants"

https://www.acas.org.uk/recruitment/interviewing-job-applicants, page shows last reviewed 30 June 2026.

Acas is the UK's statutory advisory service on employment relations, and this page is its plain guidance for employers running interviews. Four decisions in the skill come straight from it. Step 4's insistence that the questions are fixed and identical comes from "When interviewing it's a good idea to ask each applicant the same questions where possible. This is to ensure you're treating applicants in the same way." Step 5's second interviewer comes from "each interview should be carried out by more than 1 person to reduce the risk of discriminating." Step 8's adjustments line is not a courtesy but a duty: "By law you must ask anyone coming to an interview to tell you if they need reasonable adjustments so they can attend." Step 9's storage line comes from "If you make any recordings or notes in interviews, you must store them securely and confidentially." Where the skill goes further than Acas: Acas says the same questions "where possible" and calls preparing questions "a good idea", and this skill removes the escape hatch and forbids optional questions outright, because a small venue interviewing four people over a fortnight will otherwise drift, and a question only two of them were asked cannot be used to compare the four. Where the skill is softer than Acas: it treats the second interviewer as strongly recommended rather than required, because a sole trader interviewing at 10am on a Tuesday often has nobody else on site, and a rule that cannot be followed gets the whole pack ignored. The compensation is step 6 and check 3, which force the scoring anchors to be written before anyone is seen, so a single interviewer is at least scoring against something fixed.

2. Acas, "Recruitment: following discrimination law"

https://www.acas.org.uk/recruitment/follow-discrimination-law, page shows last reviewed 30 June 2026.

This page supplies the list that check 4 searches for. It names the nine protected characteristics as "age, disability, gender reassignment, marriage and civil partnership, pregnancy and maternity, race, religion or belief, sex, sexual orientation", and states the health rule in the form an owner can act on: "You must not ask a job applicant about their health or any disability at any stage in the application or interview process, unless an exception applies." Its most useful contribution is the rewriting pattern in step 7, rather than a ban list. Acas's own example is to ask "applicants about their management experience, not how they manage people older than themselves", which is the move the skill applies to childcare, sickness and availability questions: the underlying business need is legitimate, so the question is rewritten to ask about the work and the shift pattern, and asked of everyone. The skill disagrees with one thing implied by the page's framing. Acas presents the exceptions to the health rule as available to employers, and the skill tells the owner not to use them in the interview at all, and to leave health entirely to the point after a conditional offer, apart from the adjustments question about attending the interview itself. The exceptions are narrow, turn on whether a function is genuinely intrinsic to the job, and a non-specialist deciding that in the room is how a lawful-in-principle question becomes an unlawful one in practice.

3. UK Parliament, "Equality Act 2010, section 60: Enquiries about disability and health"

https://www.legislation.gov.uk/ukpga/2010/15/section/60, no publication date shown on the page, read 13 September 2026.

This is the primary law behind source 2, read directly rather than taken second hand, because a free pack published to the public should point at the statute. Section 60 provides that a person to whom an application for work is made "must not ask about the health of the applicant" before offering work or before including the applicant in a pool of people to be offered work. Subsection 6 lists the permitted purposes, which include establishing whether the applicant can comply with a requirement to undergo an assessment or whether reasonable adjustments are needed for it, whether the applicant can carry out a function intrinsic to the work, monitoring diversity, taking action to benefit disabled people, and applying an occupational requirement proportionately. The skill uses the prohibition in step 7 and deliberately uses only the narrowest of the permitted purposes, the interview adjustment, in step 8. I also loaded the GOV.UK employer guidance at https://www.gov.uk/employer-preventing-discrimination/recruitment, which restates the same ground for a non-lawyer, and step 7's date-of-birth line is its wording: "You can only ask for someone's date of birth on an application form if they must be a certain age to do the job."

4. Patrick Gavan O'Shea and Adrienne Fox Luscombe, "Is Cognitive Ability the Best Predictor of Job Performance? New Research Says It's Time to Think Again"

https://www.siop.org/tip-article/is-cognitive-ability-the-best-predictor-of-job-performance-new-research-says-its-time-to-think-again/, published in TIP, the quarterly of the Society for Industrial and Organizational Psychology, volume 60; no publication date shown on the page, read 13 September 2026.

This article summarises the 2022 re-analysis by Sackett, Zhang, Berry and Lievens, which found that decades of selection research had over-corrected for range restriction and so had the wrong tools at the top of the list. On the corrected figures the article reports structured interviews at .42, job knowledge tests at .40, empirically keyed biodata at .38, work sample tests at .33 and cognitive ability at .31, and concludes that "structured interviews may in fact be the strongest predictor of job performance—not cognitive ability". That single finding is why this skill is worth a hospitality owner's afternoon: the structured interview is not a compliance chore bolted onto a gut decision, it is the best single instrument available, and it is one they can run themselves with no test provider and no budget. An honesty note on how this was sourced. I tried to read the underlying papers directly and could not: the author's preprint of the 2022 Journal of Applied Psychology paper and the Cambridge University Press PDF of the 2023 follow-up both returned unreadable binary through the tools available here, so the numbers above are cited to this SIOP summary and not to the original tables. I did load the Cambridge landing page for Sackett, P. R., Zhang, C., Berry, C. M., and Lievens, F. (2023), "Revisiting the design of selection systems in light of new findings regarding the validity of widely used predictors", Industrial and Organizational Psychology 16(3), 283-300, at https://www.cambridge.org/core/journals/industrial-and-organizational-psychology/article/revisiting-the-design-of-selection-systems-in-light-of-new-findings-regarding-the-validity-of-widely-used-predictors/A20984B138319E3D432E643978BF026D, which confirms the authors, title and the .42 figure. Two further points the skill takes from this literature by disagreeing with how it is usually retold. First, .42 is an average across many studies with wide spread around it, so the skill never tells an owner the interview will be right, only that it beats the alternative. Second, a validity figure belongs to a structured interview, not to an interview, which is why every one of the nine steps is about removing discretion rather than about asking cleverer questions.

5. US Office of Personnel Management, "Structured Interviews"

https://www.opm.gov/policy-data-oversight/assessment-and-selection/other-assessment-methods/structured-interviews/, no publication date shown on the page, read 13 September 2026.

This is US federal hiring guidance, not UK, and it is cited only for mechanism, never for law. It is the clearest public statement of what "structured" actually means in practice, and three steps are built on it. Step 2's whole premise is "Interviews with higher degrees of structure show higher levels of validity, rater reliability, rater agreement, and less adverse impact", which is also the honest answer to an owner who asks why they should bother. Step 5's written probes come from the definition of a structured interview as "asking all applicants the same exact set of pre-defined lead and probe (i.e., follow-up questions)", a detail most interview advice skips: improvised follow-ups quietly unpick the fairness that fixed questions just bought. Step 6's anchors come from responses being "Scored according to benchmarks of proficiency". Step 3's ordering, past-behaviour question first, comes from its observation that "Behavioral description interviews have shown higher levels of validity where the nature of the work is highly complex". The skill departs from it in two ways for a small UK business. OPM assumes a rating panel in which "each member rates applicant responses individually", which a fifteen-cover bistro does not have, so the skill keeps individual scoring before discussion but allows a single interviewer. And nothing from this page is allowed anywhere near step 7: US fair-employment practice is not UK law, and the Equality Act and Acas govern what may be asked.

Best public prompt we found for this job

The best public artefact we could load is the `interview-system-designer` skill in the GitHub repository `borghei/Claude-Skills`, at https://raw.githubusercontent.com/borghei/Claude-Skills/main/engineering/interview-system-designer/SKILL.md, version 1.1.0, dated 2026-06-17 in its own metadata. Its star count is unread: api.github.com returned "API rate limit exceeded" on three attempts from this machine and no GitHub credential was available, so rather than quote a number from a badge or a search snippet, we are recording it as unknown. The line worth copying is this one:

**Question bank generation** — competency-based questions with 1-4 scoring rubrics, follow-up probes, and poor/good/great calibration examples.

That single line contains the three things almost every free interview template leaves out, and it puts them in the right order: the question is tied to a competency, the probes are written in advance rather than improvised, and the rubric ships with worked examples of what a poor and a strong answer actually sound like, which is the only thing that makes two interviewers score the same answer the same way. We copied the 1-4 scale and the calibration examples into steps 5 and 6. We deliberately did not copy two other things from that skill. Its "target score distribution (20/40/30/10)" forces a bell curve across candidates, which is meaningless when a café is seeing four people for one job, so our scoring stays absolute against the anchors. And it is written for engineering hiring loops with multiple rounds and Python tooling, where a hospitality owner needs one conversation, on paper, that they can run between lunch and service. For comparison we also read the `interview-questions` skill in `inbharatai/claude-skills` at https://raw.githubusercontent.com/inbharatai/claude-skills/main/skills/interview-questions/SKILL.md, which matches this job by name but is generic boilerplate: its instructions are "Understand the full context of your interview questions request" and "Apply best practices and conventions for Business & Productivity", it contains no rubric, no probes and nothing about unlawful questions, and it would not produce a usable interview.

Want this running in your business?

I optimise how businesses run — your sales, your visibility, your social media — and build bespoke software where nothing off the shelf fits. The first conversation is free. Work starts from £150 a day.