Compare Contrast

Overview

Run a multi-agent debate using CLI subagents. Each agent argues one perspective with evidence. Synthesize and recommend a path.

Inputs

• Topic or decision
• Perspective A
• Perspective B

Workflow

1. Clarify the decision, constraints, and success criteria.
2. Run six subagents (Codex, Gemini, Claude) using the commands below.
3. Group findings by perspective and compare evidence strength.
4. Recommend a winner and note when the other option is preferable.

Subagent Commands

Codex A

codex exec -m gpt-5.2 -s read-only -c model_reasoning_effort="low" --skip-git-repo-check \
  "Argue for {PERSPECTIVE_A} on {TOPIC}. Find codebase evidence with file paths and line numbers."

Codex B

codex exec -m gpt-5.2 -s read-only -c model_reasoning_effort="low" --skip-git-repo-check \
  "Argue for {PERSPECTIVE_B} on {TOPIC}. Find codebase evidence with file paths and line numbers."

Gemini A

CLOUDSDK_CORE_PROJECT="" GOOGLE_CLOUD_PROJECT="" GCLOUD_PROJECT="" GEMINI_API_KEY=${GEMINI_API_KEY} \
  gemini -m gemini-3-pro-preview -o text "Argue for {PERSPECTIVE_A} on {TOPIC}. Provide evidence with file paths and line numbers."

Gemini B

CLOUDSDK_CORE_PROJECT="" GOOGLE_CLOUD_PROJECT="" GCLOUD_PROJECT="" GEMINI_API_KEY=${GEMINI_API_KEY} \
  gemini -m gemini-3-pro-preview -o text "Argue for {PERSPECTIVE_B} on {TOPIC}. Provide evidence with file paths and line numbers."

Claude A

claude --model haiku -p "Argue for {PERSPECTIVE_A} on {TOPIC}. Provide evidence with file paths and line numbers."

Claude B

claude --model haiku -p "Argue for {PERSPECTIVE_B} on {TOPIC}. Provide evidence with file paths and line numbers."

Output

• Side-by-side comparison
• Evidence-backed recommendation