Socrates, Feynman, and Torvalds enter your terminal

What it does council-of-high-intelligence is a shell-based deliberation framework that plugs into Claude Code or Codex CLI. Typing /council dispatches your question to 18 persona-prompted agents — from Aristotle to Torvalds — and runs a structured debate across multiple LLM providers. The output is a verdict that surfaces disagreement instead of smoothing it over.

The interesting bit The project treats unanimous AI agreement as a failure mode. It enforces genuine disagreement through “polarity pairs” (Socrates destroys assumptions while Feynman rebuilds from first principles), hard-separates those voices across different providers, and triggers dissent quotas if too many agents agree too early. It is essentially very thoughtful glue code around existing LLM CLIs.

Key highlights

• 18 personas with deliberate counterweights (e.g., Torvalds “ship it” vs. Watts “does the problem exist?”)
• Auto-routing across Anthropic, OpenAI, Gemini, Ollama, and NVIDIA NIM to ensure model diversity rather than costume changes on one model
• Structured 3-round protocol: problem restatement, independent analysis, cross-examination, and enforcement scan
• Pre-defined domain triads (e.g., architecture, shipping, ai-safety) and leaner council profiles
• Groupthink prevention via dissent quotas, novelty gates, and forced steel-manning of minority views

Caveats

• Requires Claude Code or Codex CLI; it is a shell integration, not a standalone application
• Real model diversity depends on having multiple provider credentials installed locally
• Fundamentally prompt orchestration and subagent dispatch; the smarts are in the protocol, not a new model

Verdict Worth a look if you want structured disagreement for hard decisions instead of one LLM’s confident best guess. Skip it if you already find one AI voice noisy enough.