--- name: interview-loop-designer description: Designs a structured, multi-stage interview loop for a specific open role, assigning exactly one evaluation purpose per interview, generating a scorecard for each stage, and producing a champion-or-veto decision protocol. Use whenever the user is opening a role, rewriting a hiring process, diagnosing bad hires, deciding who should interview whom, or asking how to reduce founder bias in hiring. Trigger even if the user only says "help me hire a [role]." --- # Interview loop designer ## Purpose Most hiring processes fail in one specific way: several people ask the same candidate overlapping questions, form one blended impression, and then vote. Strong personality contaminates the competence read. Strong credentials contaminate the culture read. This skill separates the evaluations so each signal stays clean. ## Step 1: Gather the role inputs Ask the user for these before designing anything. Ask in one message, not one at a time. 1. Role title and the single outcome this person owns in 12 months. 2. The cash component of the budgeted compensation. 3. Who is available to interview, with a one-line note on each person's strongest evaluation instinct (for example: "Ryan reads people and values, weak on technical screening"). 4. The two or three ways a hire in this role has failed before, at this company or elsewhere. 5. Whether the role is located in or hiring into NYC, Illinois, California, Colorado, or Quebec. These have AI-in-hiring rules that change the process. If the user cannot answer #1 in one sentence, stop and help them write it. A loop built on a vague outcome will produce a vague hire. ## Step 2: Assign one purpose per interview Build a loop of four to six stages. Each stage gets exactly one purpose drawn from this set. Never combine two purposes in one interview. - **Competence.** Can they do the specific work? Run by the closest functional manager. - **Problem-solving.** How do they think when they do not know the answer? Run as a live case or a scoped take-home, never as trivia. - **Resume verification.** Go line by line into their claimed history. What exactly did they own, what did the team do, what would a reference say. This stage exists because resume inflation is now trivially easy to generate. - **Culture and values.** What do they care about, what do they refuse to do, what would make them leave. Run by whoever on the team reads people best, even if that person is a poor judge of technical skill. Especially then. - **Mission alignment.** Why this problem, at this company, now. Run late, after the candidate has enough context to answer honestly. Assign each stage to a named interviewer based on the instincts the user described in Step 1. Say explicitly which interviewer should be kept OUT of which stage and why. ## Step 3: Write the scorecard For each stage, output: - Three to five questions specific to this role, not generic. - What a strong answer contains. - What a disqualifying answer contains. - A 1 to 5 rating anchor where 3 is described concretely, not as "meets expectations." Interviewers score only their own stage. They do not see other scores until the round table. ## Step 4: The decision protocol After all stages, run a single round table. Rules: - Each interviewer states their score and the evidence behind it, in the order the stages ran. - Any interviewer may **champion** the candidate, meaning they personally stake their judgment on the hire working out. - Any interviewer may issue a **veto**, meaning they believe this hire is a mistake and will say so on the record. - A candidate with no champion is not hired, regardless of average score. - A candidate with both a champion and a veto goes to one additional targeted interview aimed at the specific concern, then back to the table once. Output this protocol as a short document the hiring manager can paste into the calendar invite. ## Step 5: Compliance flags If the user named a regulated jurisdiction in Step 1, append a plain-language note covering what applies. Do not give legal advice. Flag the obligation and tell them to confirm with counsel. Known triggers as of this writing: - NYC Local Law 144: annual independent bias audit, published summary, and 10 business days' notice to candidates before any automated screening or ranking tool is used. - Colorado: SB 24-205 was repealed and reenacted as SB 26-189, the Colorado Automated Decision-Making Technology Act, effective January 1, 2027. Verify current status before relying on this. - Illinois and California each regulate automated employment decision tools. Check current requirements. ## Rules - AI, including you, may score and rank. AI never nominates, eliminates, or decides. A human owns every gate. - Never design a loop where the person who most wants the hire is also the only person evaluating competence. - If the user asks you to compress the loop to one or two interviews, comply, but state in one sentence which signal they are choosing to go blind on. - Do not pad the loop to look rigorous. Four well-separated stages beat eight overlapping ones.