Agent Skills: User Interview Skill

User Interview Skill

Transform vague requirements into actionable specifications through structured questioning.

Core Principles

Past behavior over future intentions. Never ask "Would you use X?" — ask "Tell me about the last time you tried to solve this." Ground every discussion in concrete past experiences.

Problems over solutions. When users describe solutions ("add a dropdown"), redirect to problems: "What problem does that solve?" and "What would that enable you to do?"

Specifics over generalities. When users speak abstractly ("it's frustrating"), immediately anchor: "Can you give me a specific example?" or "When did that last happen?"

One question at a time. Never overwhelm with multiple questions. Ask the most important question, wait for the answer, then follow up.

Interview Flow

Phase 1: Context Discovery

Start broad, understand the landscape:

- "What are you trying to accomplish?"
- "Walk me through what happens today when you [do this task]."
- "What triggered this request? What happened recently?"

Phase 2: Problem Validation

Confirm the problem exists and matters:

- "How are you solving this today?"
- "What's the hardest part about that?"
- "What have you already tried?"
- "How much time/money does this cost you?"

Phase 3: Constraint Mapping

Identify boundaries and requirements:

- "Who else uses this? Who else is affected?"
- "What would a successful outcome look like?"
- "What would make this a failure?"
- "Are there any technical/business/time constraints I should know about?"
- "What's definitely out of scope?"

Phase 4: Edge Case Exploration

Surface hidden complexity:

- "What happens when [X goes wrong]?"
- "What's the most unusual case you've encountered?"
- "Are there exceptions to how this normally works?"
- "What happens if [user does unexpected thing]?"

Phase 5: Priority Clarification

Distinguish must-haves from nice-to-haves:

- "If you could only have one thing, what would it be?"
- "What would you cut if we had half the time?"
- "Is this blocking something else?"

Probing Techniques

Use these when responses are vague or incomplete:

| Technique | When to Use | Example | |-----------|-------------|---------| | Echo probe | Encourage elaboration | "It felt 'clunky'?" | | Specificity probe | Vague statement | "Can you give me a specific example?" | | Contrast probe | Unclear criteria | "What would good vs bad look like?" | | Why probe | Surface motivation | "Why is that important?" | | Consequence probe | Understand impact | "What happens if we don't do this?" |

Five Whys: For root cause analysis, ask "Why?" iteratively (usually 3-5 times) until you reach the underlying need.

Validation Signals

Real interest manifests as:

• Time commitment (willing to discuss further, bring others)
• Existing workarounds (already trying to solve it)
• Quantifiable pain (can estimate time/money lost)
• Urgency indicators (deadlines, blocking issues)

Warning signs of unclear requirements:

• "Make it better/faster/easier" without measurable criteria
• "Like [competitor] but different" without specifying how
• Different stakeholders interpret the requirement differently
• Cannot write acceptance criteria
• Requirements describe HOW instead of WHAT

Output: Translating to Specs

When you have sufficient clarity, produce the appropriate artifact or hand off to a downstream workflow:

User Stories - For discrete implementable units:

As a [user type], I want to [action] so that [benefit].

Acceptance Criteria:
- Given [context], when [action], then [result]
- Given [context], when [action], then [result]

Technical Spec - For implementation-focused requirements. See references/tech-spec-template.md

PRD Handoff - For formal PRDs, use the workflow-create-prd skill using the insights gathered here.

Completeness Checklist

Before generating output, verify:

• [ ] Core problem is validated (not just assumed)
• [ ] Success criteria are measurable
• [ ] Edge cases identified
• [ ] Scope boundaries explicit (in AND out)
• [ ] User/actor types identified
• [ ] Constraints documented (technical, business, time)
• [ ] Priorities clear (must-have vs nice-to-have)
• [ ] No conflicting requirements remain unresolved

What NOT to Do

• Don't ask leading questions: "Wouldn't it be great if...?" biases responses
• Don't suggest solutions: Let users describe problems first
• Don't accept first answer: Probe deeper, especially on "why"
• Don't ask hypotheticals: "Would you use X?" yields unreliable data
• Don't ask multiple questions at once: One question, one answer
• Don't assume shared understanding: Clarify ambiguous terms

Quick Reference by Situation

| Situation | Start With | |-----------|------------| | "Build me X" | "What problem does X solve for you?" | | "Make it better" | "Better in what way? What's not working now?" | | "Like [competitor]" | "What specifically about [competitor] do you want?" | | "It's broken" | "Walk me through what happened." | | "Add a feature" | "What would that feature let you do that you can't do now?" | | User describes solution | "Interesting—what problem are you trying to solve?" | | User can't articulate | "Can you show me how you do this today?" |

References

• references/tech-spec-template.md - Technical specification template
• references/question-bank.md - Extended question library by category
• references/failure-modes.md - Common requirements gathering failures and how to avoid them

Integration with Workflows

This skill is upstream of the workflow-create-prd skill. Once the interview is complete and requirements are clarified:

• Suggest using workflow-create-prd to formalize the requirements into a structured PRD.
• Use the Technical Spec template for implementation-focused requirements when a full PRD is not necessary.