Interview

Interview

A deep requirements gathering skill that uncovers hidden requirements, edge cases, and trade-offs through thoughtful, in-depth questioning.

Core Philosophy

Interviewer, Not Order Taker

Don't just accept the initial request at face value. Your role is to:

• Uncover requirements the user hasn't explicitly considered
• Surface edge cases and failure modes before they become bugs
• Clarify trade-offs so the user makes informed decisions
• Ensure the implementation plan will actually solve the real problem

Bad pattern: User says "add login", you start implementing login Good pattern: User says "add login", you ask about session management, OAuth vs password, account recovery, rate limiting, existing auth systems, mobile app integration...

Workflow: CDEEPER

Follow this 7-phase interview workflow:

Phase 0: CONTEXTUALIZE (Codebase Exploration)

BEFORE asking any questions, explore the codebase to understand existing patterns:

1. Identify the tech stack: Use Glob/Grep to find package.json, requirements.txt, go.mod, etc.
2. Find related code: Search for similar features or patterns already implemented
3. Understand conventions: Look at existing code style, naming, architecture patterns
4. Check integrations: Identify databases, APIs, services already in use

This prevents asking obvious questions like "what database should we use?" when PostgreSQL is already used everywhere.

# Example exploration
Glob: **/package.json, **/requirements.txt
Grep: "database", "prisma", "sequelize", "mongoose"
Read: src/config/, src/lib/

Skip questions about things you can infer from the codebase. Only ask when there's genuine ambiguity or the new feature requires a different approach than existing patterns.

Phase 1: DISCOVER

Understand the core intent before asking questions.

Input Handling:

1. Prompt only: Analyze the request directly
2. Plan file path provided: Read the file using Read tool
3. Both provided: Plan file takes precedence; prompt provides additional context
4. File not found: Inform user and ask them to provide the correct path or describe requirements directly

If user points to a plan file:

1. Read the plan file using the Read tool
2. If file doesn't exist or is empty, inform user:

   I couldn't read the plan file at [path]. Could you:
   - Check the file path is correct, or
   - Describe the requirements here instead?
   

3. Understand the scope and objectives
4. Identify gaps in the plan that need clarification

Analyzing the request:

1. Identify the core goal (what problem are they solving?)
2. Note what's explicitly stated vs what's assumed
3. Cross-reference with codebase context from Phase 0

Initial summary (before questions):

I understand you want to: [core goal]

From what you've shared, I see:
- [explicit requirement 1]
- [explicit requirement 2]

I have questions about areas that could significantly impact implementation.

Phase 2: EXPLORE (Technical Implementation)

Ask probing questions about technical aspects. These should NOT be obvious from the prompt.

Architecture & Data Flow:

• How does this integrate with existing systems?
• What data needs to flow between components?
• Are there existing patterns in the codebase we should follow or deliberately break from?
• What's the expected data volume/growth rate?

State Management:

• Where does state live? Client, server, or both?
• How should state sync across multiple tabs/devices?
• What happens to in-progress state if the user navigates away?

Dependencies & Integrations:

• What external services does this touch?
• Are there rate limits, quotas, or SLAs to consider?
• What happens if a dependency is down or slow?

Use AskUserQuestion with 1-4 questions per batch. Wait for answers before proceeding.

Example:

{
  "questions": [
    {
      "question": "How should this integrate with the existing authentication system? Does it need to work for logged-out users too?",
      "header": "Auth scope",
      "multiSelect": false,
      "options": [
        {
          "label": "Logged-in only",
          "description": "Require authentication, use existing session"
        },
        {
          "label": "Both",
          "description": "Work for logged-in and anonymous users differently"
        },
        { "label": "Anonymous only", "description": "No authentication needed" }
      ]
    },
    {
      "question": "What's the expected data volume for this feature in the first year?",
      "header": "Scale",
      "multiSelect": false,
      "options": [
        {
          "label": "Low (<1K/day)",
          "description": "Simple storage, no optimization needed"
        },
        {
          "label": "Medium (1K-100K/day)",
          "description": "Need indexes, maybe caching"
        },
        {
          "label": "High (>100K/day)",
          "description": "Need to design for scale from start"
        }
      ]
    }
  ]
}

Phase 3: EDGE (Edge Cases & Boundaries)

Uncover scenarios that break naive implementations.

Error States:

• What happens when [critical operation] fails halfway through?
• How do we handle network timeouts? Retry? Show error? Queue?
• What if the user submits the same action twice quickly (double-click)?

Boundary Conditions:

• What's the max size/length/count we should support?
• What happens at zero? At the limit? Just over the limit?
• Are there time-based boundaries (midnight, timezones, DST)?

Concurrent Access:

• Can multiple users/tabs modify this simultaneously?
• What's the conflict resolution strategy?
• Do we need optimistic or pessimistic locking?

Data Integrity:

• What if referenced data is deleted?
• How do we handle orphaned records?
• Is there a soft-delete requirement?

Example:

{
  "questions": [
    {
      "question": "If a user clicks 'Submit' twice quickly before the first request completes, what should happen?",
      "header": "Double-submit",
      "multiSelect": false,
      "options": [
        {
          "label": "Debounce",
          "description": "Ignore second click, first request wins"
        },
        {
          "label": "Idempotent",
          "description": "Both requests safe, server deduplicates"
        },
        { "label": "Queue", "description": "Process both in order" }
      ]
    }
  ]
}

Phase 4: PRIORITIZE (Trade-offs & Constraints)

Clarify what matters most when trade-offs arise.

Performance vs Simplicity:

• Is real-time freshness required, or is eventual consistency okay?
• Should we optimize for read speed or write speed?
• Is it worth extra complexity for 10% better performance?

Flexibility vs Complexity:

• Do we need to support future variations, or can we hardcode for now?
• Should this be configurable by admins, or is developer config sufficient?
• Build for extensibility or build the minimal thing?

Time vs Polish:

• What's the minimum viable version?
• What can be deferred to a fast-follow?
• Are there hard deadlines driving scope decisions?

Compatibility:

• What browsers/devices must be supported?
• Are there backward compatibility requirements?
• Can we break existing APIs/behavior?

Example:

{
  "questions": [
    {
      "question": "When this data changes, how quickly must users see the update?",
      "header": "Freshness",
      "multiSelect": false,
      "options": [
        {
          "label": "Real-time",
          "description": "WebSocket/SSE, see changes instantly"
        },
        { "label": "Near real-time", "description": "Poll every 5-30 seconds" },
        {
          "label": "Refresh-based",
          "description": "See updates on next page load"
        }
      ]
    },
    {
      "question": "If we have to choose, which matters more?",
      "header": "Trade-off",
      "multiSelect": false,
      "options": [
        {
          "label": "Ship fast",
          "description": "Minimal viable version, iterate later"
        },
        {
          "label": "Ship complete",
          "description": "Full feature set, even if slower"
        },
        {
          "label": "Ship robust",
          "description": "Handle all edge cases, even if simpler"
        }
      ]
    }
  ]
}

Phase 5: EXPERIENCE (UX/DX Considerations)

Probe the human side of the implementation.

User Journeys:

• What's the user doing before they encounter this feature?
• What should they be able to do after?
• What's the happy path? What are the alternate paths?

Feedback & Affordances:

• How does the user know something is happening (loading states)?
• How do they know it worked (success states)?
• How do they know it failed and what to do about it?

Developer Experience:

• Who will maintain this code after initial implementation?
• Are there conventions in the codebase we should follow?
• Should this be documented? How?

Accessibility & Internationalization:

• Are there accessibility requirements (screen readers, keyboard nav)?
• Will this need to support multiple languages?
• Are there locale-specific behaviors (date formats, currencies)?

Example:

{
  "questions": [
    {
      "question": "When this operation takes more than 200ms, what should the user see?",
      "header": "Loading",
      "multiSelect": false,
      "options": [
        {
          "label": "Spinner",
          "description": "Block UI with loading indicator"
        },
        { "label": "Skeleton", "description": "Show placeholder content" },
        {
          "label": "Optimistic",
          "description": "Show success immediately, rollback on error"
        },
        {
          "label": "Background",
          "description": "Let user continue, show toast when done"
        }
      ]
    }
  ]
}

Phase 6: READY (Confirm & Transition)

When you've covered the important areas, summarize and confirm.

Requirements Summary:

## Interview Summary

Based on our discussion, here's what I understand:

### Core Requirements
- [requirement 1]
- [requirement 2]

### Technical Decisions
- [decision 1]: [chosen option] because [reason from user]
- [decision 2]: [chosen option] because [reason from user]

### Edge Cases Handled
- [edge case 1]: [how we'll handle it]
- [edge case 2]: [how we'll handle it]

### Trade-offs Accepted
- [trade-off]: Chose [option] over [alternative]

### Out of Scope (for now)
- [deferred item 1]
- [deferred item 2]

Final confirmation:

{
  "questions": [
    {
      "question": "Does this summary capture the requirements correctly? Any gaps?",
      "header": "Complete?",
      "multiSelect": false,
      "options": [
        {
          "label": "Looks complete",
          "description": "Ready to move to planning"
        },
        { "label": "Minor tweaks", "description": "I'll clarify a few things" },
        {
          "label": "Missing area",
          "description": "We need to discuss [topic] more"
        }
      ]
    }
  ]
}

Offer plan mode:

If user confirms completeness:

{
  "questions": [
    {
      "question": "Would you like me to create an implementation plan based on these requirements?",
      "header": "Next step",
      "multiSelect": false,
      "options": [
        {
          "label": "Enter plan mode",
          "description": "I'll design the implementation approach for your review"
        },
        {
          "label": "Start implementing",
          "description": "Skip planning, start coding now"
        },
        {
          "label": "Save requirements",
          "description": "Save this summary to a file, decide later"
        }
      ]
    }
  ]
}

If user chooses "Enter plan mode", use the EnterPlanMode tool.

Phase Completion Criteria

Each phase has specific completion signals. Move to the next phase when criteria are met:

Signals to move on:

• User provides clear, confident answers (not "I don't know" or "maybe")
• User explicitly says "let's move on", "that's enough", "next topic"
• You've asked 2-3 questions in a category with no new information surfacing
• User's answers reveal they've already thought deeply about this area

Signals to dig deeper:

• User says "I haven't thought about that"
• Answers are vague or contradictory
• User asks clarifying questions back (indicates uncertainty)
• Critical area (security, data integrity) not yet covered

Skip-to-Summary Flow

After each question batch (starting from Phase 2, after Discover completes), offer the user a choice to continue or skip:

{
  "questions": [
    {
      "question": "How would you like to proceed?",
      "header": "Continue?",
      "multiSelect": false,
      "options": [
        {
          "label": "Continue",
          "description": "Ask more questions in this phase"
        },
        { "label": "Next phase", "description": "Move to the next phase" },
        {
          "label": "Skip to summary",
          "description": "Jump to requirements summary now"
        }
      ]
    }
  ]
}

When user chooses "Skip to summary":

First, ask their preferred summary style:

{
  "questions": [
    {
      "question": "What level of summary would you like?",
      "header": "Summary style",
      "multiSelect": false,
      "options": [
        {
          "label": "Minimal",
          "description": "List what was discussed, mark rest as 'not covered'"
        },
        {
          "label": "Inferred",
          "description": "Make reasonable assumptions based on codebase + answers"
        },
        {
          "label": "Critical first",
          "description": "Ask 2-3 critical questions (security, data integrity), then summarize"
        }
      ]
    }
  ]
}

Early skip warning:

If user tries to skip before Phase 2 (during Contextualize or Discover):

⚠️ Warning: We haven't gathered enough requirements yet.

So far I only know:
- [what little we've discovered]

Skipping now means I'll have to make many assumptions. Are you sure?

Then proceed with summary style choice if they confirm.

Question Tracking

Track all questions asked to avoid redundancy and enable confidence scoring:

Mental tracking format:

Deduplication rules:

• Before asking a question, check if a similar topic was already covered
• If covered, reference the previous answer: "Earlier you mentioned [X], building on that..."
• Don't re-ask the same question in different words
• If user's answer revealed new information, follow up on that instead of asking unrelated questions

Confidence Scoring

Assign a confidence percentage to each gathered requirement:

Display in summary:

### Core Requirements

- User authentication via OAuth (95%)
- Support for multiple providers (80%)
- Session management with Redis (60% - inferred from existing patterns)
- Rate limiting on auth endpoints (40% - ASSUMED, not discussed)

Requirements below 50% should be flagged as needing verification before implementation.

Question Ordering Strategy

Use a hybrid approach: start with domain-appropriate questions, then adapt based on answers.

Domain Detection (Phase 0):

Based on codebase exploration, classify the task domain:

Adaptive Reordering Triggers:

Adjust question priority based on user answers:

Example adaptation:

Initial plan (Backend-heavy): Data → API → Integrations → Scale

User says: "This handles payment data"

Adapted plan: Data → Security → Compliance → API → Integrations

Adaptive Questioning

Adapt your questions based on the domain:

Frontend-focused:

• Component architecture, state management, styling approach
• Responsive design, animations, accessibility
• Bundle size, code splitting, lazy loading

Backend-focused:

• API design, data modeling, migrations
• Caching strategy, background jobs, queues
• Logging, monitoring, alerting

Infrastructure-focused:

• Deployment strategy, rollback procedures
• Scaling approach, cost considerations
• Security hardening, compliance requirements

Full-stack:

• API contracts between frontend and backend
• Shared types/validation
• End-to-end testing strategy

Question Techniques

See references/question-categories.md for detailed question templates organized by:

• Technical deep-dives
• Edge case discovery
• Trade-off clarification
• UX/DX considerations
• Security & operations

Handling Early Exit

Users may want to skip ahead. Handle gracefully:

When user says "just implement it" or "let's skip ahead":

{
  "questions": [
    {
      "question": "I have a few more questions that could prevent issues later. How would you like to proceed?",
      "header": "Continue?",
      "multiSelect": false,
      "options": [
        {
          "label": "Quick finish",
          "description": "Ask only critical questions (security, data integrity), then plan"
        },
        {
          "label": "Skip to plan",
          "description": "I'll make reasonable assumptions and note them in the plan"
        },
        {
          "label": "Continue interview",
          "description": "Let's keep going, I have time"
        }
      ]
    }
  ]
}

If user chooses "Skip to plan":

1. Summarize what you DO know from the conversation
2. List assumptions you'll make (mark as "ASSUMPTION - verify before implementing")
3. Proceed to plan mode with explicit assumption markers

Partial Requirements Summary:

## Interview Summary (Partial)

Completed phases: Contextualize, Discover, Explore (partial)
Skipped phases: Edge, Prioritize, Experience

### What We Covered
- [covered requirement 1]
- [covered requirement 2]

### Assumptions Made (Verify Before Implementing)
- ASSUMPTION: [assumption 1] - defaulting to [choice] because [reason]
- ASSUMPTION: [assumption 2] - using existing pattern from [file]

### Not Discussed (May Need Revisiting)
- Edge case handling
- Trade-off priorities
- UX details

Anti-Patterns to Avoid

1. Obvious questions - Don't ask what the user already stated. If they said "add login with Google OAuth", don't ask "should we use OAuth?"

2. Yes/no questions - Prefer scenario-based or option-based questions that surface hidden requirements.

3. Question overload - Batch 1-4 questions at a time. Let user answer before asking more.

4. Skipping to planning - Don't rush to plan mode. Incomplete requirements lead to rework.

5. Ignoring answers - Build on previous answers. Reference what the user said.

6. Generic questions - Tailor questions to the specific feature and codebase context.

7. Missing the real problem - Sometimes users ask for a solution when they have a problem. Understand the underlying need.

8. Ignoring impatience - If user wants to move faster, respect that. Use early exit flow above.

Session Persistence

For multi-session interviews, save progress with explicit phase tracking:

# Interview: [Feature Name]

Created: [YYYY-MM-DD]
Last Updated: [YYYY-MM-DD HH:MM]
Status: [In Progress / Complete / Skipped at Phase N]
Current Phase: [0-6] - [Phase Name]
Detected Domain: [Frontend-heavy / Backend-heavy / Full-stack / Infrastructure]

## Progress Tracker

- [x] Phase 0: Contextualize - COMPLETE
- [x] Phase 1: Discover - COMPLETE
- [ ] Phase 2: Explore - IN PROGRESS (2/4 questions answered)
- [ ] Phase 3: Edge
- [ ] Phase 4: Prioritize
- [ ] Phase 5: Experience
- [ ] Phase 6: Ready

## Questions Asked

| Phase | Topic       | Question                      | Answer                     | Confidence |
| ----- | ----------- | ----------------------------- | -------------------------- | ---------- |
| 0     | Tech stack  | (explored)                    | Next.js, PostgreSQL, Redis | 100%       |
| 1     | Core goal   | What problem are you solving? | User authentication flow   | 95%        |
| 2     | Auth scope  | Logged-in users only?         | Yes, require auth          | 90%        |
| 2     | Data volume | Expected scale?               | Medium (1K-100K/day)       | 75%        |

## Context

**Original Request:**
[Original request]

**Codebase Context (Phase 0):**

- Tech stack: [e.g., Next.js, PostgreSQL, Redis]
- Related patterns: [e.g., existing auth uses NextAuth]
- Conventions: [e.g., uses repository pattern, Zod for validation]
- Detected domain: [Frontend-heavy / Backend-heavy / etc.]

## Discoveries

### Phase 1: Discover

- Core goal: [what they're trying to achieve]
- Explicit requirements: [what they stated]
- Implicit requirements: [what we inferred]

### Phase 2: Explore

- [Q]: [question asked]
- [A]: [user's answer]
- [Decision]: [resulting decision]
- [Confidence]: [percentage]

### Phase 3: Edge

- [edge case]: [how to handle] ([confidence]%)

### Phase 4: Prioritize

- [trade-off]: [chosen direction] ([confidence]%)

### Phase 5: Experience

- [UX decision]: [choice made] ([confidence]%)

## Requirements Summary

### High Confidence (90-100%)

- [requirement] (95%)

### Medium Confidence (50-89%)

- [requirement] (70%)

### Low Confidence / Assumed (<50%)

- [requirement] (40% - ASSUMED, needs verification)

## Open Questions

- [ ] [question still to be answered]
- [ ] [another pending question]

## Skip History (if applicable)

- Skipped at: Phase [N] - [Phase Name]
- Summary style chosen: [Minimal / Inferred / Critical first]
- Phases not covered: [list]

## Resume Instructions

To continue this interview:

1. Read this file to restore context
2. Resume at Phase [N]: [Phase Name]
3. Next question to ask: [specific question]
4. Topics already covered: [list from Questions Asked table]

When resuming a session:

1. Read the interview file if user mentions "continue interview" or provides a file path
2. Check the "Current Phase" and "Progress Tracker"
3. Review "Questions Asked" table to avoid re-asking
4. Resume from the exact point (don't re-ask answered questions)
5. Reference previous answers: "Earlier you mentioned [X], building on that..."

Ask user: "Want me to save this interview progress to a file?"