Council ThinkDeep Skill

Multi-step investigation and deep reasoning with both Codex (GPT-5.2) and Gemini using extra-high reasoning.

When to Use

• Complex problems requiring step-by-step analysis
• Architecture decisions
• Performance deep-dives
• Security analysis
• When user asks for "deep", "thorough", or "comprehensive" council analysis

Reasoning Level

xhigh (always - this skill is for deep analysis)

Execution

1. Understand the problem fully

2. Gather all relevant files and context

3. Formulate a structured prompt:

   Investigate this step by step. Form hypotheses, test them, and build confidence.
   
   Question: <user's question>
   
   Context:
   <relevant file contents>
   
   Please:
   1. Break down the problem
   2. Form initial hypotheses
   3. Analyze each hypothesis
   4. Reach a well-reasoned conclusion
   

4. Run BOTH commands in parallel:

   Codex:

   codex exec --sandbox read-only -c model_reasoning_effort="xhigh" "<prompt>"
   

   Gemini:

   gemini -s -y -o json "<prompt>"
   

5. Synthesize the comprehensive analyses

Response Format

## AI Council Deep Analysis

### Codex (GPT-5.2) Analysis:
**Problem Breakdown:**
[Step-by-step analysis]

**Hypotheses Explored:**
[Each hypothesis and findings]

**Conclusion:**
[Final recommendation with confidence level]

---

### Gemini Analysis:
**Problem Breakdown:**
[Step-by-step analysis]

**Hypotheses Explored:**
[Each hypothesis and findings]

**Conclusion:**
[Final recommendation with confidence level]

---

### Council Synthesis:
**Shared Insights:** [What both models discovered]
**Unique Perspectives:** [Insights unique to each model]
**Combined Recommendation:** [Synthesized recommendation]
**Confidence:** [High/Medium/Low based on agreement]

---
*Session IDs: Codex=[id], Gemini=[id]*