Deep Interview

Overview

Deep Interview combines Ouroboros-inspired Socratic questioning with mathematical ambiguity scoring. It asks targeted questions to surface hidden assumptions, measures clarity across weighted dimensions, and refuses to proceed until ambiguity falls below a threshold (default 20%).

Designed as the first step of a 3-stage pipeline: deep-interview → ralplan → autopilot

Usage

/oh-my-claudecode:deep-interview "vague idea or description"

Also activated via magic keywords.

deep interview me about this project idea
ouroboros help me clarify what I want
I have a vague idea about a task management tool

Ambiguity Score

Clarity is measured across 4 dimensions during the interview.

Dimension	Greenfield Weight	Brownfield Weight
Goal clarity	40%	35%
Constraint clarity	30%	25%
Success criteria clarity	30%	25%
Context clarity	-	15%

Brownfield (modifying existing code) additionally measures understanding of the existing system.

Score Interpretation

Range	Meaning
0-10%	Fully clear, proceed immediately
10-20%	Sufficiently clear (default threshold)
20-40%	Some gaps, continue interview
40-60%	Significant gaps, focus on weak dimensions
60-80%	Very unclear, reframing needed
80-100%	Almost nothing is known

Challenge Agents

The questioning perspective shifts at specific rounds.

Mode	Activated	Purpose
Contrarian	Round 4+	"What if the opposite is true?" — challenge core assumptions
Simplifier	Round 6+	"What is the simplest version?" — strip complexity
Ontologist	Round 8+ (ambiguity > 30%)	"What is this really?" — find the essence

Each mode is used only once.

Execution Phases

Phase 1: Initialization

Parse the user's idea
Detect Brownfield vs. Greenfield (explore agent)
Initialize state

Phase 2: Interview Loop

Generate questions targeting the weakest dimension
Only one question at a time
Calculate and display ambiguity score after each answer
Codebase facts gathered first via the explore agent

Phase 3: Challenge Agents

Perspective shifts at rounds 4, 6, and 8.

Phase 4: Spec Crystallization

When the ambiguity threshold is reached, generate a spec file: .omc/specs/deep-interview-{slug}.md

Phase 5: Execution Bridge

Present execution options:

Ralplan → Autopilot (recommended): consensus refinement then autonomous execution
Autopilot (fast): direct autonomous execution without consensus refinement
Ralph: continuous loop execution
Team: N parallel agents
Continue interview: keep interviewing for greater clarity

3-Stage Pipeline

deep-interview → ralplan → autopilot
  ambiguity ≤ 20%   consensus plan   execution

Each stage provides a different quality gate.

Deep Interview: clarity gate (does the user know what they want?)
Ralplan: feasibility gate (is the approach sound?)
Autopilot: correctness gate (does the code work?)

ralplan - consensus-based planning (stage 2)
autopilot - autonomous execution (stage 3)
plan - strategic planning

Reference

Item	Value
Invoke	`/oh-my-claudecode:deep-interview`
Magic Keywords	`deep interview`, `ouroboros`, `socratic`
Category	Workflow
Spec Path	`.omc/specs/deep-interview-{slug}.md`

On this page