Interview Framework

Interview Framework

Adaptive brainstorming dialogue algorithm for all spec phases. Each phase command references this skill and provides its own exploration territory (phase-specific areas to probe).

Option Limit Rule

Each question MUST have 2-4 options (max 4). Keep the most relevant options, combine similar ones.

Intent-Based Depth Scaling

Read .progress.md for intent classification. Scale dialogue depth accordingly:

Completion Signal Detection

After each response, check if user wants to end early:

completionSignals = ["done", "proceed", "skip", "enough", "that's all", "continue", "next"]

if askedCount >= minRequired:
  for signal in completionSignals:
    if signal in userResponse.lower():
      -> SKIP remaining questions, move to PROPOSE APPROACHES

3-Phase Algorithm

Phase 1: UNDERSTAND (Adaptive Dialogue)

UNDERSTAND:
  1. Read all available context:
     - .progress.md (prior phase answers, intent, goal)
     - Prior artifacts (research.md, requirements.md, etc.)
     - Original goal text
  2. Read the exploration territory provided by the calling command
  3. Identify what is UNKNOWN vs what is already decided
     - If prior phases already covered a topic, SKIP it
     - Only ask about what still needs clarification
  4. Set depth from intent:
     - minRequired = intent.minQuestions
     - maxAllowed = intent.maxQuestions
  5. askedCount = 0

  WHILE askedCount < maxAllowed:
    |
    +-- Generate next question from context + exploration territory
    |   (Questions emerge from what you've learned so far, NOT from a fixed pool)
    |
    +-- Context-based skip check:
    |   Read .progress.md holistically. If this topic was already
    |   answered in a prior phase, SKIP it. Log: "Already covered: [topic]"
    |
    +-- Ask single question:
    |   AskUserQuestion:
    |     question: "[Context-aware question referencing prior answers]"
    |     options:
    |       - "[Option 1]"
    |       - "[Option 2]"
    |       - "[Option 3 if needed]"
    |       - "Other"
    |
    +-- askedCount++
    |
    +-- If user selected "Other":
    |   -> Ask context-specific follow-up (see Adaptive Depth below)
    |   -> DO NOT increment askedCount for follow-ups
    |
    +-- Check completion signals (see above)
    |
    +-- Decide: ask another question or move to PROPOSE APPROACHES
    |   (If you have enough context to propose meaningful approaches, move on)

Key rules for question generation:

• Each question builds on prior answers in THIS dialogue AND prior phases
• Reference specific things the user said ("You mentioned X — does that mean...")
• Never ask something .progress.md already answers
• Never ask generic questions — every question must be grounded in the user's context

Phase 2: PROPOSE APPROACHES

PROPOSE APPROACHES:
  1. Synthesize the dialogue into 2-3 distinct approaches
  2. Each approach MUST include:
     - Name (short label)
     - Description (1-2 sentences)
     - Trade-offs (pros and cons)
  3. Lead with your recommendation
  4. Present via AskUserQuestion:

  AskUserQuestion:
    question: "Based on our discussion, here are the approaches I see:

      **A) [Recommended] [Name]**
      [Description]. Trade-off: [pro] vs [con].

      **B) [Name]**
      [Description]. Trade-off: [pro] vs [con].

      **C) [Name]** (if applicable)
      [Description]. Trade-off: [pro] vs [con].

      Which approach fits best?"
    options:
      - "A) [Name]"
      - "B) [Name]"
      - "C) [Name]" (if applicable)
      - "Other"

  5. If user picks "Other":
     -> Ask what they'd change or combine
     -> Iterate until approach is confirmed (max 3 rounds)
  6. Store chosen approach as primary input for the subagent

Approach rules:

• Always present at least 2 approaches (never just 1)
• Maximum 3 approaches (more causes decision fatigue)
• The recommended approach goes first
• Trade-offs must be honest — no straw-man alternatives
• Apply YAGNI: strip unnecessary complexity from all approaches

Phase 3: CONFIRM & STORE

CONFIRM & STORE:
  1. Brief recap to the user:
     "Here's what I'll pass to the [agent name]:
      - [Key decision 1]
      - [Key decision 2]
      - [Chosen approach summary]
      Does this look right?"
  2. If user corrects something, update before storing
  3. Store in .progress.md (see Context Accumulator below)
  4. Proceed to subagent delegation

Adaptive Depth (Other Responses)

If user selects "Other" for any question:

1. Ask a context-specific follow-up (NEVER generic "elaborate")
2. Continue until clarity is reached or 5 rounds complete
3. Each follow-up uses a single question focused on their response

Follow-up questions MUST be context-specific, not generic. When user provides an "Other" response:

1. Acknowledge the specific response: Reference what the user actually typed
2. Ask a probing question based on response content: Analyze keywords in their response
3. Include context from prior answers: Reference earlier responses to create continuity

Do NOT use generic follow-ups like "Can you elaborate?" — always tailor to their specific response.

Example — if user types "We need GraphQL support" for a technical approach question:

AskUserQuestion:
  question: "You mentioned needing GraphQL support. Is this for the entire API layer, or specific endpoints only?"
  options:
    - "Full API layer - replace REST"
    - "Hybrid - GraphQL for new endpoints only"
    - "Specific queries for mobile clients"
    - "Other"

Example — if user types "Security is critical" for success criteria:

AskUserQuestion:
  question: "You emphasized security is critical. Given your earlier constraints, which security aspects matter most?"
  options:
    - "Authentication and authorization"
    - "Data encryption at rest and in transit"
    - "Audit logging and compliance"
    - "Other"

Context Accumulator Pattern

After each interview, update .progress.md:

1. Read existing .progress.md content
2. Append new section under "## Interview Responses"
3. Use descriptive keys that reflect what was actually discussed
4. Include the chosen approach

Storage Format

### [Phase] Interview (from [phase].md)
- [Topic 1]: [response]
- [Topic 2]: [response]
- Chosen approach: [name] — [brief description]
[Any follow-up responses from "Other" selections]