LLM CLI Council

Get diverse AI perspectives on your implementation plans and code reviews

A Claude Code skill that orchestrates multiple LLM command-line tools (Claude, Copilot, Codex, Gemini, Ollama) to provide comprehensive feedback through a "council" approach. Claude acts as Chairman, synthesizing insights from multiple AI assistants into clear, actionable guidance.

Inspired by: karpathy/llm-council for the council orchestration concept and jarrodwatts/claude-delegator for multi-model delegation patterns.

Why Use a Council?

Single LLMs can be biased toward certain solutions. By consulting multiple AI assistants:

• 🎯 Diverse perspectives catch blind spots and assumptions
• ✅ Consensus builds confidence in recommendations
• ⚠️ Dissent highlights areas needing careful consideration
• 📊 Better coverage different models excel at different tasks

The Anti-Paralysis Principle

The council never overwhelms with conflicting advice. Instead, it:

• ✨ Synthesizes into max 5 prioritized recommendations
• 🎓 Resolves conflicts with clear reasoning
• ✅ Always provides a decisive verdict: APPROVE or REQUEST CHANGES

No endless debate. Just actionable guidance.

Features

• 🤝 Multi-LLM orchestration - Coordinates Claude, Copilot, Codex, Gemini, and Ollama
• 🎭 Independent reviews - Each provider reviews without seeing others' opinions
• ⚡ Parallel execution - All providers run simultaneously for speed
• 🪑 Chairman synthesis - Claude resolves conflicts and provides final verdict
• 🔒 Privacy mode - Local-only execution with Ollama (no data leaves your machine)
• 📈 Smart routing - Automatically selects best providers for each task type
• 🔧 Configurable modes - Quick (2 providers), Full (all providers), or Privacy (local only)
• 🌍 Cross-platform - Works on macOS, Linux, and Windows (WSL)

Quick Start

Get your first council review in 5 minutes:

# 1. Install the plugin
claude /plugin marketplace add GuitaristForEver/llm-cli-council
claude /plugin install llm-cli-council

# Restart Claude Code after installation

# 2. Setup and detect your LLM CLI tools
/llm-cli-council:setup

# 3. Review a plan
/llm-cli-council:review-plan PLAN.md

That's it! You'll get synthesized feedback from multiple AI perspectives.

Installation

npx llm-cli-council

This will detect your Claude Code directory and install the plugin files automatically.

Options:

• npx llm-cli-council --dry-run — preview what will be installed
• npx llm-cli-council --yes — skip the confirmation prompt

Manual install: If you prefer, copy files directly:

git clone https://github.com/GuitaristForEver/llm-cli-council
cd llm-cli-council
cp -r skills/* ~/.claude/skills/
cp -r lib ~/.claude/lib/
cp -r prompts ~/.claude/prompts/
cp -r rules ~/.claude/rules/
cp config/providers.json ~/.claude/config/

Available Commands

| Command | Description | |---------|-------------| | /llm-cli-council:setup | Detect and configure available LLM CLIs | | /llm-cli-council:review-plan | Get council review of an implementation plan | | /llm-cli-council:review-code | Get council review of code changes | | /llm-cli-council:status | Show council configuration and provider status | | /llm-cli-council:uninstall | Remove council configuration |

Basic Usage

Review an Implementation Plan

# Review with default (quick) mode - 2 best providers
/llm-cli-council:review-plan

# Review specific plan file
/llm-cli-council:review-plan path/to/PLAN.md

# Use all available providers (comprehensive review)
/llm-cli-council:review-plan --mode=full

# Privacy mode - only local Ollama (no data sent externally)
/llm-cli-council:review-plan --mode=privacy

Review Code Changes

# Review staged changes
/llm-cli-council:review-code

# Review last commit
/llm-cli-council:review-code HEAD

# Review specific file
/llm-cli-council:review-code src/app.js

# Review range of commits
/llm-cli-council:review-code main..HEAD --mode=full

Check Configuration

# Show current configuration
/llm-cli-council:status

# Re-check provider availability
/llm-cli-council:status --check

Council Modes

The council operates in three modes, each balancing speed, comprehensiveness, and privacy:

| Mode | Providers Used | Use Case | Speed | |------|----------------|----------|-------| | quick (default) | 2 best-match providers | Most reviews, fast feedback | ⚡⚡⚡ Fast | | full | All available providers | High-stakes decisions, comprehensive analysis | ⚡⚡ Slower | | privacy | Ollama only | Sensitive code, air-gapped environments | ⚡⚡⚡ Fast |

Mode selection example:

# Quick mode (default) - 2 providers, ~30-60 seconds
/llm-cli-council:review-plan

# Full mode - all providers, ~60-120 seconds
/llm-cli-council:review-plan --mode=full

# Privacy mode - local only, ~30-90 seconds
/llm-cli-council:review-plan --mode=privacy

Supported Providers

The council supports the following LLM CLI tools:

| Provider | CLI Command | Installation | Strengths | |----------|-------------|--------------|-----------| | Claude | claude | Claude Code | Reasoning, analysis, synthesis, planning | | Copilot | copilot | npm install -g @github/copilot-cli | Code generation, GitHub integration | | Codex | codex | OpenAI Codex CLI | Code analysis, debugging, implementation | | Gemini | gemini | Gemini CLI | Multimodal analysis, broad knowledge | | Ollama | ollama | Ollama | Privacy, offline, local execution |

Minimum requirement: At least 2 providers for meaningful council feedback. Recommendation: Install 3-5 providers for diverse perspectives.

Provider Setup

Each provider requires separate installation and authentication:

1. Install the CLI tool (see links above)
2. Authenticate (follow provider-specific instructions)
3. Run setup: /llm-cli-council:setup
4. Verify: /llm-cli-council:status

For detailed setup instructions, see Provider Setup Guide.

Configuration

The council uses environment variables for flexible configuration:

Environment Variables

| Variable | Default | Description | |----------|---------|-------------| | CLAUDE_COUNCIL_CONFIG_DIR | ~/.config/claude/council/ | Config directory location | | CLAUDE_COUNCIL_CONFIG_FILE | $CONFIG_DIR/config.json | Specific config file override | | CLAUDE_COUNCIL_LOG_DIR | $CONFIG_DIR/logs/ | Log directory | | CLAUDE_SKILLS_DIR | ~/.claude/skills/ | Skills installation directory | | CLAUDE_CONFIG_DIR | ~/.claude/ | Claude's main config directory |

Path Resolution Priority

When locating configuration, the council checks in this order:

1. $CLAUDE_COUNCIL_CONFIG_DIR (explicit override)
2. $CLAUDE_CONFIG_DIR/council/ (Claude config + council subdir)
3. $XDG_CONFIG_HOME/claude/council/ (XDG standard on Linux)
4. ~/.config/claude/council/ (XDG fallback)
5. ~/.claude/council/ (legacy compatibility)

Configuration File

After running setup, your configuration is stored at $COUNCIL_CONFIG_FILE:

{
  "version": "1.0.0",
  "providers": {
    "claude": {
      "available": true,
      "path": "/usr/local/bin/claude",
      "authenticated": true
    },
    "codex": {
      "available": true,
      "path": "/usr/local/bin/codex",
      "authenticated": true
    }
  },
  "defaultMode": "quick"
}

For detailed configuration options, see Configuration Guide.

How It Works

The council uses a three-stage deliberation process:

┌─────────────────────────────────────────────────────────┐
│ Stage 1: Independent Review                            │
├─────────────────────────────────────────────────────────┤
│ • Execute prompts in parallel across providers          │
│ • Each provider reviews independently                   │
│ • No cross-contamination of opinions                    │
└─────────────────────────────────────────────────────────┘
                         ↓
┌─────────────────────────────────────────────────────────┐
│ Stage 2: Collection & Analysis                         │
├─────────────────────────────────────────────────────────┤
│ • Gather all provider responses                         │
│ • Extract key recommendations                           │
│ • Identify consensus and dissent                        │
└─────────────────────────────────────────────────────────┘
                         ↓
┌─────────────────────────────────────────────────────────┐
│ Stage 3: Chairman Synthesis (Claude)                   │
├─────────────────────────────────────────────────────────┤
│ • Synthesize into max 5 prioritized recommendations     │
│ • Resolve conflicts with clear reasoning                │
│ • Provide decisive verdict: APPROVE / REQUEST CHANGES   │
└─────────────────────────────────────────────────────────┘

Why Claude as Chairman?

Claude serves as Chairman because:

• Context retention - Best at synthesizing multiple perspectives
• Reasoning depth - Excels at explaining trade-offs and resolutions
• Clarity - Provides decisive, actionable guidance without paralysis

Examples

Example: Plan Review Output

╔═══════════════════════════════════════════════════════╗
║ LLM CLI COUNCIL REVIEW                                ║
║ Plan: Implementation Plan for Feature X               ║
║ Mode: quick (2 providers)                            ║
╚═══════════════════════════════════════════════════════╝

VERDICT: REQUEST CHANGES

COUNCIL SYNTHESIS
─────────────────────────────────────────────────────────

Consensus (2/2 providers agree):
✓ Overall approach is sound and well-structured
✓ Clear separation of concerns
✗ Missing error handling strategy
✗ No consideration of database migration

TOP 5 RECOMMENDATIONS (Prioritized):

1. [CRITICAL] Add Database Migration Strategy
   • Plan lacks migration approach for schema changes
   • Risk: Data loss during deployment
   • Suggested: Add migration rollback plan in Phase 2

2. [HIGH] Define Error Handling Approach
   • No consistent error handling across services
   • Both reviewers flagged this independently
   • Suggested: Establish error handling patterns upfront

3. [MEDIUM] Add Performance Benchmarks
   • Consider load testing before production
   • Codex suggests: Define acceptable latency targets

4. [MEDIUM] Clarify Testing Strategy
   • Integration tests mentioned but not detailed
   • Suggested: Specify test coverage goals per phase

5. [LOW] Documentation Plan
   • Consider API documentation approach
   • Minor issue, can be addressed later

REASONING:
Database migration and error handling are architectural
decisions that affect all subsequent phases. Addressing
these upfront prevents costly refactoring later. The plan
is otherwise well-structured—these additions will make it
production-ready.

See more examples in the examples/ directory.

Troubleshooting

Common Issues

"Council not configured"

Solution: Run /llm-cli-council:setup to detect providers.

"No providers available"

Cause: No LLM CLI tools detected.

Solution:

1. Install at least 2 providers (see Supported Providers)
2. Ensure they're in your PATH
3. Run /llm-cli-council:setup again

"Provider authentication failed"

Cause: Provider CLI not authenticated.

Solution:

• Claude: claude auth login
• Copilot: copilot auth login
• Codex: codex auth
• Gemini: gemini auth login
• Ollama: No auth needed (ensure models downloaded: ollama pull model-name)

Timeout errors

Cause: Provider taking too long to respond.

Solution:

• Check network connection
• Try --mode=privacy if network is slow
• Increase timeout in config (advanced)

For more troubleshooting, see Troubleshooting Guide.

Documentation

Comprehensive documentation is available in the docs/ directory:

• Installation Guide - Detailed installation instructions
• Configuration Reference - All configuration options explained
• Provider Setup - Step-by-step setup for each provider
• Troubleshooting - Common issues and solutions
• Architecture - How the council works internally

Contributing

We welcome contributions! Here's how you can help:

Ways to Contribute

• 🐛 Report bugs - Open an issue with reproduction steps
• 💡 Suggest features - Share your ideas in discussions
• 📝 Improve docs - Fix typos, add examples, clarify instructions
• 🔧 Add providers - Support additional LLM CLI tools
• ✨ Submit PRs - Fix bugs or implement features

Development Setup

# Clone repository
git clone https://github.com/GuitaristForEver/llm-cli-council.git
cd llm-cli-council

# Install as local plugin
claude /plugin marketplace add .
claude /plugin install llm-cli-council@llm-cli-council-dev

# Restart Claude Code

# Make changes to skills/, prompts/, or rules/

# Test your changes
/llm-cli-council:setup
/llm-cli-council:review-plan test/fixtures/sample-plan.md

# To reload changes: uninstall and reinstall
claude /plugin uninstall llm-cli-council@llm-cli-council-dev
claude /plugin install llm-cli-council@llm-cli-council-dev

Adding a New Provider

See CONTRIBUTING.md for step-by-step instructions on adding support for new LLM CLI tools.

Roadmap

v1.1.0 (Next Release)

• [ ] GitHub Action for automated PR reviews
• [ ] Customizable provider weights and priorities
• [ ] Review history and analytics
• [ ] Support for custom prompt templates

Future

• [ ] Web UI for review visualization
• [ ] Docker container with pre-installed providers
• [ ] VS Code extension
• [ ] npm package distribution
• [ ] Multi-language support

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Anthropic for Claude and Claude Code CLI
• GitHub for Copilot CLI
• OpenAI for Codex
• Google for Gemini
• Ollama team for local LLM infrastructure

Special thanks to the open source community for feedback and contributions.

Support

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Documentation: docs/

Built with ❤️ for developers who value diverse perspectives