deep-interview

Deep Interview

Ouroboros-inspired Socratic questioning with mathematical ambiguity scoring. Replaces vague ideas with crystal-clear specifications by asking targeted questions that expose hidden assumptions.

Pipeline

deep-interview → ralplan (consensus refinement) → omg-autopilot (execution)

When to Use

• User has a vague idea and wants thorough requirements gathering
• Task is complex enough that jumping to code would waste cycles
• User wants mathematically-validated clarity before execution

When NOT to Use

• Detailed specific request with file paths → execute directly
• Quick fix → delegate to @executor or /ralph
• User says "just do it" → respect their intent

Phases

Phase 1: Initialize

1. Parse the user's idea
2. Detect brownfield vs greenfield (use @explore to check codebase)
3. For brownfield: map relevant codebase areas
4. Initialize ambiguity score at 100%

Phase 2: Interview Loop

Repeat until ambiguity <= 20% or user exits early:

1. Generate question targeting the WEAKEST clarity dimension
2. Ask ONE question at a time with current ambiguity context
3. Score ambiguity across dimensions:

• Goal Clarity (40% weight for greenfield, 35% brownfield)
• Constraint Clarity (30% / 25%)
• Success Criteria (30% / 25%)
• Context Clarity (N/A / 15% for brownfield)

1. Report progress with dimension scores and gaps
2. Track ontology (key entities, stability ratio)

Phase 3: Challenge Agents

• Round 4+: Contrarian - challenge core assumptions
• Round 6+: Simplifier - probe for complexity removal
• Round 8+: Ontologist - find the essence (if ambiguity > 30%)

Phase 4: Crystallize Spec

When ambiguity <= threshold, generate spec to .omc/specs/deep-interview-{slug}.md:

• Goal, Constraints, Non-Goals, Acceptance Criteria
• Assumptions Exposed & Resolved
• Ontology (Key Entities) with convergence tracking
• Interview Transcript

Phase 5: Execution Bridge

Present options:

1. Ralplan → OMG Autopilot (Recommended): consensus-refine then execute
2. Execute with omg-autopilot (skip ralplan)
3. Execute with ralph: persistence loop
4. Execute with team: parallel agents
5. Refine further: continue interviewing

Rules

• Ask ONE question at a time
• Target the WEAKEST clarity dimension each round
• Gather codebase facts via @explore BEFORE asking user
• Score ambiguity after every answer
• Do not proceed until ambiguity <= threshold (default 20%)
• Hard cap at 20 rounds, soft warning at 10