---
name: structured-interview-questions
source: https://app.decimal.ai/s/structured-interview-questions@1/SKILL.md
source_sha256: 65796cfebcc6
---

# Behavioral interview questions for a competency

Draft the questions an interviewer asks to assess a specific competency. Asked to "come up with
questions to gauge how a candidate handles X", base models know what a behavioral question is —
but they don't default to one. They mix in three weaker forms that don't produce evaluable
evidence:

1. **Speculative** — "How *would* you handle a disagreement with your manager?" A candidate can
   describe the textbook answer without ever having lived it. You learn what they *think* good
   looks like, not what they *did*.
2. **Closed** — "Are you good at handling pressure?" / "Do you work well on a team?" Answerable
   with one word, and the word is always yes.
3. **Leading** — "How do you stay so organized?" presupposes the candidate is organized. The
   question hands them the conclusion you were supposed to test.

This skill produces the fourth form: **open, past-behavior questions** that make the candidate
narrate a real event, so you can score what they actually did.

## When to use / when NOT

- **Use** when generating or improving the questions to *ask* a candidate about a named
  competency, trait, or behavior for a role.
- **Do NOT use** for: vetting already-drafted questions for legal/discrimination risk
  (interview-question-compliance); turning interview notes into a scorecard
  (interview-scorecard-format); or scoring a resume against minimum bars (candidate-screening).

## The rule

**Ask what the candidate did, not what they would do, believe, or are.** Every question points at
a specific past event the candidate lived through, and is phrased so they cannot answer it in one
word or read the desired answer off the question.

## The four form checks

Each question must pass all four:

1. **Past, not hypothetical.** Anchors a real event that already happened. Openers:
   *"Tell me about a time…"*, *"Describe a situation where…"*, *"Walk me through a specific
   instance when…"*. Never *"How would you…"*, *"What would you do if…"*, *"Imagine…"* — those
   invite a rehearsed ideal, not evidence.
2. **Open, not closed.** Invites a narrative. If it can be answered *yes*, *no*, or in one word,
   reopen it. "Are you decisive?" → "Tell me about a decision you had to make with incomplete
   information."
3. **Non-leading.** Carries no presupposition and telegraphs no right answer. "How do you keep
   your projects so organized?" assumes the trait; ask instead "Describe a time you had more work
   than time — how did you decide what got done?" Don't embed the value word ("great", "smoothly",
   "successfully") in the ask.
4. **Competency-anchored.** Targets the observable behaviors of the *named* competency, not
   generic filler ("Tell me about yourself", "What's your greatest weakness?"). If you swapped in
   a different competency and the question still fit, it's too generic.

## Elicit the full STAR

A good question opens the door; the **follow-up probes** get the evidence. STAR = Situation,
Task, Action, Result. Candidates over-narrate the Situation and skip the Action and Result — the
two parts that actually reveal the competency. Pair every lead question with probes that force
those out:

- **Action:** "What specifically did *you* do?" / "What was your part versus the team's?"
- **Result:** "How did it turn out?" / "How did you know it worked?" / "What would you do
  differently?"

Give **at least one probe per lead question**, and make the Action/Result probes explicit —
that's where a "we" answer gets pinned down to an "I".

## Worked example

> Competency to assess: **handling conflict with a peer**, for a senior engineer.

Weak (what the base tends to emit):
- "How would you handle a disagreement with a coworker?" *(hypothetical)*
- "Do you work well with difficult people?" *(closed, leading)*

Strong:
- **Lead:** "Tell me about a time you disagreed with a peer on a technical decision. What was the
  disagreement?"
  - *Probe (Action):* "What did you personally do to work through it?"
  - *Probe (Result):* "How was it resolved, and what did the working relationship look like
    afterward?"
- **Lead:** "Describe a situation where a colleague pushed back hard on your work. Walk me through
  what happened."
  - *Probe (Action):* "What did you say or change in response?"
  - *Probe (Result):* "Looking back, what would you do differently?"

## Output shape

For the named competency, produce **3–5 lead questions**, each on its own line, each with its
probes nested underneath. Don't grade the answers, don't invent the candidate's responses, and
don't restate the competency name as a question ("Are you good at conflict?"). If the ask names a
role, tune the *situations* to that role's world (a manager's peers, a salesperson's accounts) —
but keep every question a past-behavior, open, non-leading one.