Agent Skills
› zereight/gitlab-mcp
› deep-interview
deep-interview
GitHub通过苏格拉底式提问和模糊度评分，将用户模糊需求转化为清晰规范，辅助复杂任务的需求澄清与规格定义。
Trigger Scenarios
用户有模糊想法需要深入挖掘
直接编码可能浪费精力的复杂任务
执行前需验证清晰度
Install
npx skills add zereight/gitlab-mcp --skill deep-interview -g -y
SKILL.md
Frontmatter
{
"name": "deep-interview",
"description": "Socratic deep interview with mathematical ambiguity gating. Activate when user says: deep interview, interview me, ask me everything, don't assume, make sure you understand, ouroboros, socratic, I have a vague idea, not sure exactly what I want.\n",
"argument-hint": "[--quick|--standard|--deep] <idea or vague description>"
}
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
- Parse the user's idea
- Detect brownfield vs greenfield (use @explore to check codebase)
- For brownfield: map relevant codebase areas
- Initialize ambiguity score at 100%
Phase 2: Interview Loop
Repeat until ambiguity <= 20% or user exits early:
- Generate question targeting the WEAKEST clarity dimension
- Ask ONE question at a time with current ambiguity context
- 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)
- Report progress with dimension scores and gaps
- 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:
- Ralplan → OMG Autopilot (Recommended): consensus-refine then execute
- Execute with omg-autopilot (skip ralplan)
- Execute with ralph: persistence loop
- Execute with team: parallel agents
- 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
Version History
- a8f5d4f Current 2026-07-25 10:03
