aaaronmiller

Deliberative Refinement — AI skill for Claude Code

AI community

How to install Deliberative Refinement

This entry records only its repository, not the path inside it, so there is no exact command to give. Open aaaronmiller/deliberative-refinement and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Deliberative Refinement does

Run your content through multiple rounds where the AI critiques itself from different expert angles (structured code review, strategic debate councils, elimination tournaments for binary choices), ground with multi-web search between passes, and revise until only ideas that survive adversarial pressure remain.

Alternatives in AI

Debate — AI Debate Hub - Structured four-way debates between Claude, Sonnet, Gemini, and Codex 2.8k ★
Mysti — AI coding dream team of agents for VS Code 1k ★
Agentty — AI pair programming in your terminal — one static binary, sub-ms startup, any model 597 ★

README

Deliberative Refinement

Stop asking AI for answers. Make it defend them.

A structured multi-agent deliberation framework for complex reasoning, uncertain situations, and consequential decisions.

What Is This?

**Deliberative Refinement** forces AI through multiple rounds of structured critique—different expert perspectives attack from different angles, web search grounds claims between passes, iteration continues until only ideas that survive adversarial pressure remain.

The Core Idea

Instead of prompt → output → polish, you architect critique rounds:

Expert councils attack the draft from different angles (code reviewers, strategy councils, elimination tournaments)
Web search grounds claims with actual evidence between rounds
Iteration continues until weak ideas die and strong ones survive

The Breakthrough

Six months ago, this required orchestrating a dozen different models—one for generation, another for critique, a third for fact-checking. **Tool-interspersed reasoning** collapsed all of that into a single model that switches roles on demand. Same quality, fraction of the complexity.

flowchart LR subgraph Input A[📄 Draft] end

subgraph "Phase 0: Intent Detection"
    B{Detect Mode}
    B -->|CREATE| C1[Generate First]
    B -->|REFINE| C2[Analyze Input]
    B -->|DECIDE| C3[Frame Options]
    B -->|DISCOVER| C4[Explore Space]
end

subgraph "Phase 1: Council Selection"
    D{Select Council}
    D -->|Code/Tech| E1[Structured Review]
    D -