--- name: interview-question-compliance source: https://app.decimal.ai/s/interview-question-compliance@1/SKILL.md source_sha256: 3d7f9d85a0a9 --- # Interview question legal-risk review Reviews interview questions before an interviewer asks them and catches the ones that create discrimination risk. A capable base model reliably flags the *blatant* ones ("How old are you?", "What's your religion?", "Are you married?"), but it under-performs on the two things that actually matter here: 1. **Subtle proxies.** A question never has to name a protected class to be risky. "What year did you graduate?" is an age question; "Do you have reliable childcare?" is a caregiver-status question; "Where's that accent from?" is a national-origin question. The base often reads these as friendly small talk and passes them through. 2. **The fix, not just the flag.** The base tends to say "maybe don't ask that." A hiring manager needs the *compliant reframe* — a question that gets the same job-relevant information legally. This skill always supplies the reframe. It also knows that **salary-history questions are now banned in many US states and cities** — a rule the base applies inconsistently. ## When to activate - Review / sanity-check / vet a list of planned interview questions before an interview. - "Is there anything I shouldn't ask?" about a hiring script or interviewer notes. - Clean up questions an interviewer drafted so they are safe to use. Not for: scoring a candidate against a rubric (candidate-screening), or writing a job description. ## The one principle **Ask about the ability to do the job, never about the personal circumstance behind it.** Every compliant reframe replaces a question about who someone *is* (or their family, health, origin, beliefs, age) with a question about whether they can perform the role's essential functions. ## Protected categories and the proxy questions that hide in them Flag a question if it touches any of these — including indirectly. The proxy phrasings are the ones the base misses. - **Age** (ADEA, 40+). Proxies: graduation or birth year, "how long until you retire?", "you're overqualified / this is a young team", "digital native", references to being close to retirement or "energy level". - **Family / marital / caregiver / pregnancy** (Title VII, PDA). Proxies: do you have kids, are you married, is your spouse local, are you planning a family, are you pregnant, who watches your children, childcare arrangements, "how does your family feel about the travel?". - **Disability / health** (ADA). Proxies: any medical condition, prescriptions, past workers'-comp or sick-leave use, mental-health history, how many sick days you took, "do you have anything that would stop you from…". A physical requirement ("lift 50 lbs") is askable ONLY as a stated essential function ("this role requires lifting up to 50 lbs — can you do that with or without accommodation?"), never as a fishing question about health. - **National origin / citizenship / ancestry** (Title VII, IRCA). Proxies: where are you *from originally*, birthplace, first language / mother tongue, "what kind of name is that?", accent, "are you a US citizen?" (ask work authorization instead). - **Religion** (Title VII). Proxies: what holidays do you observe, are you free to work Saturdays/ Sundays framed around belief, what church do you attend, "any beliefs we should know about?". - **Salary history — where banned.** Current or previous pay, "what do you make now?", W-2 or pay stubs. Banned for employers in many US jurisdictions (California, New York, Massachusetts, Washington, Colorado, Connecticut, Illinois, New Jersey, and more — the list grows and varies). Flag it and ask about **expectations for this role** instead. - Also protected, same treatment: **race / color, sex / gender, sexual orientation & gender identity** (Bostock; also many state laws), **genetic information** (GINA — no family medical history), and **arrest record** (ask about relevant *convictions* only, subject to ban-the-box and EEOC guidance). The full statute-by-statute list and the jurisdiction table for salary-history and ban-the-box rules live in **references/jurisdiction-and-topic-catalog.md** — consult it for a jurisdiction or edge case not shown here. ## What to return For each risky question, give three things: 1. **Flag + category** — which protected class it risks. 2. **Why** — one line on why it's a proxy (the interviewer usually didn't intend harm). 3. **Compliant reframe** — a job-related question that gets the legitimate information legally, or "drop — not job-related" when there is no legitimate version. Leave clearly job-related questions ("Walk me through a project you led", "Are you available for the on-call rotation?") unflagged. Do not manufacture risk where there is none. ## Reframe patterns (representative) | Risky question | Compliant reframe | |---|---| | "Are you a US citizen?" | "Are you legally authorized to work in the US?" | | "What year did you graduate?" | Drop — ask about relevant experience, not dates. | | "Do you have kids / childcare sorted?" | "This role has occasional evening shifts and travel — can you meet that schedule?" | | "Where are you from originally?" | Drop — not job-related. | | "Is English your first language?" | "This role requires writing client reports in English — are you comfortable with that?" | | "What's your current salary?" | "What are your salary expectations for this role?" (and it may be illegal to ask) | | "Any health issues that would get in the way?" | "Can you perform the essential functions of this role, with or without reasonable accommodation?" | | "What holidays do you observe?" | "This role occasionally requires weekend coverage — are you able to work weekends?" | ## Notes - This is general guidance for US federal + common state rules and is **not legal advice**; laws change and vary by jurisdiction — confirm current local law before finalizing a hiring script. - Volunteered information is different from a *question*. The skill reviews what the interviewer plans to *ask*.