Multi-AI structured debate skill for Claude Code — orchestrate advocate, critic & external AI CLIs (codex, pi, gemini, qwen, kimi) in 3-round debates
SaferSkills independently audited claude-code-agent-team-debate (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Multi-AI structured debate skill for Claude Code. Orchestrates a team of Claude Code agents (advocate, critic, proxy) to debate any technical topic, optionally bringing in external AI perspectives from 5 different CLI tools.
Platform: Developed and tested on Windows + PowerShell 5.1+. Internal-only debates (no--withflag) work on all platforms. External CLI integration (--with) requires PowerShell (Windows native orpwshon Mac/Linux).
Claude Code is Anthropic's official CLI for Claude. You need it to run this skill.
Install Claude Code:
# Via npm (requires Node.js 18+)
npm install -g @anthropic-ai/claude-code
# Verify installation
claude --versionFirst-time setup:
claude # Opens Claude Code, follow login promptsYou need a Claude Pro/Max subscription for agent team features (multiple agents running simultaneously).
Agent teams are experimental and disabled by default. Add this to your ~/.claude/settings.json:
{
"env": {
"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
}
}Or set it temporarily via environment variable:
$env:CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS = "1"Display mode (optional): On macOS/Linux with tmux or iTerm2, agent teams can show in split panes. On Windows, teammates run in-process (use Shift+Up/Down to switch between them).# Check if installed
node --version # Need 18+
# If not installed — use winget (Windows)
winget install OpenJS.NodeJS.LTSgit --version # Need 2.x+
# If not installed
winget install Git.Gitpython --version
# If not installed
winget install Python.Python.3.12git clone https://github.com/meesp123/claude-code-multi-ai-debate.git
cd claude-code-multi-ai-debate# Replace $YOUR_PROJECT with your actual project path
Copy-Item .claude\commands\debate.md $YOUR_PROJECT\.claude\commands\ -Force
Copy-Item .claude\agents\debate-advocate.md $YOUR_PROJECT\.claude\agents\ -Force
Copy-Item .claude\agents\debate-critic.md $YOUR_PROJECT\.claude\agents\ -Force
Copy-Item .claude\agents\debate-proxy.md $YOUR_PROJECT\.claude\agents\ -Force# Create directory if it doesn't exist
New-Item -ItemType Directory -Path "$HOME\.claude\scripts" -Force
Copy-Item scripts\invoke-ai.ps1 "$HOME\.claude\scripts\invoke-ai.ps1" -Force/debate to verify.# Create directories
New-Item -ItemType Directory -Path "$HOME\.claude\commands" -Force
New-Item -ItemType Directory -Path "$HOME\.claude\agents" -Force
New-Item -ItemType Directory -Path "$HOME\.claude\scripts" -Force
# Copy all files
Copy-Item .claude\commands\debate.md "$HOME\.claude\commands\" -Force
Copy-Item .claude\agents\debate-*.md "$HOME\.claude\agents\" -Force
Copy-Item scripts\invoke-ai.ps1 "$HOME\.claude\scripts\" -ForceYou only need these if you want to use --with codex,gemini,... to bring in external AI perspectives. Without them, /debate still works with internal advocate + critic agents.
| CLI | Install Command | Auth Command | Verify |
|---|---|---|---|
| codex | npm i -g @openai/codex | codex auth | codex --version |
| pi | npm i -g @mariozechner/pi-coding-agent | pi auth | pi --version |
| gemini | npm i -g @google/gemini-cli@latest | gemini auth | gemini --version |
| qwen | npm i -g qwen-cli | qwen auth | qwen --version |
| kimi | pip install kimi-cli | kimi auth | kimi --version |
Note: CLI package names and auth methods may change. See docs/cli-setup.md for the latest detailed instructions.
# Test which CLIs are available
codex --version 2>$null && Write-Host "codex: OK" || Write-Host "codex: NOT INSTALLED"
pi --version 2>$null && Write-Host "pi: OK" || Write-Host "pi: NOT INSTALLED"
gemini --version 2>$null && Write-Host "gemini: OK" || Write-Host "gemini: NOT INSTALLED"
qwen --version 2>$null && Write-Host "qwen: OK" || Write-Host "qwen: NOT INSTALLED"
kimi --version 2>$null && Write-Host "kimi: OK" || Write-Host "kimi: NOT INSTALLED"This skill uses Claude Code's agent team feature. Here's what you need to know:
Claude Code can spawn multiple AI agents that work together. Each agent has:
| Agent | Role | Model | What it does |
|---|---|---|---|
| Lead (you) | Moderator | Your current model | Orchestrates phases, synthesizes DECISION |
| Advocate | Proposer | Sonnet | Proposes and defends a solution |
| Critic | Challenger | Sonnet | Finds weaknesses, suggests alternatives |
| Proxy ×N | Bridge | Haiku | Calls external AI CLIs, relays their opinions |
┌─────────────┐ ┌──────────────┐ ┌──────────────┐
│ Advocate │ │ Critic │ │ Proxy ×N │
│ (proposes) │ │ (challenges) │ │ (external AI)│
└──────┬───────┘ └──────┬───────┘ └──────┬───────┘
│ │ │
└────────────┬───────┘────────────────────┘
│
┌─────┴─────┐
│ Lead │
│(orchestrate│
│ + decide) │
└────────────┘/debate Should we use REST or GraphQL for our API --rounds 2/debate Database indexing strategy --files src/db/queries.ts --with codex,gemini/debate <topic> [--files <path1,path2>] [--with <cli1,cli2>] [--preset <balanced|deep>] [--rounds <2|3>] [--effort <low|medium|high|xhigh>]| Parameter | Required | Default | Description |
|---|---|---|---|
<topic> | Yes | — | The debate topic (free text) |
--files | No | — | Source files for agents to analyze |
--with | No | — | External CLIs to include (comma-separated) |
--preset | No | balanced | Timeout/depth preset |
--rounds | No | 2 | Number of debate rounds (2 or 3) |
--effort | No | high | Reasoning effort for external CLIs |
| Preset | Rounds | Timeout | Best for |
|---|---|---|---|
balanced | 2 | 15 min | Most technical debates |
deep | 3 | 30 min | Architecture decisions |
| Platform | Internal Debate | External CLI (--with) | Notes |
|---|---|---|---|
| Windows | ✅ Full | ✅ Full | Primary development platform |
| macOS | ✅ Full | ⚠️ Needs pwsh | brew install powershell |
| Linux | ✅ Full | ⚠️ Needs pwsh | Install PowerShell Core |
debate.md is in .claude/commands/ (project or $HOME\.claude\commands\)--preset deep for longer timeoutscodex auth / gemini authclaude --version$env:PYTHONIOENCODING = 'utf-8' in your PowerShell profileContributions welcome! Here's how:
git checkout -b feature/my-featuregit commit -m "feat: description"git push origin feature/my-featuregh pr create --fillfeat: — New featurefix: — Bug fixdocs: — Documentation onlyrefactor: — Code change that neither fixes a bug nor adds a featureBuilt with Claude Code by Anthropic.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.