Agora — Multi-Agent Debate System for OpenClaw

A structured multi-agent advisory system built on OpenClaw v2026.5.2.
Two AI agents — Socrates (GPT-5.5) and Plato (Gemini 3.1 Flash) — engage in a structured debate before delivering a synthesised advisory to the user.

Architecture

User Question
     │
     ▼
┌─────────────┐    sessions_spawn     ┌─────────────┐
│  Socrates   │ ────────────────────► │    Plato    │
│  (Master)   │                       │ (Disciple)  │
│  GPT-5.5    │ ◄──── auto-announce ──│ Gemini 3.1  │
│             │       (push-based)    │   Flash     │
└─────────────┘                       └─────────────┘
     │
     ▼
🏛️ AGORA ADVISORY

How It Works

1. User asks Socrates a question via the OpenClaw chat UI
2. Socrates frames an initial position internally
3. Socrates spawns a Plato subagent session via sessions_spawn with agentId: "plato"
4. Socrates yields his turn via sessions_yield and waits
5. Plato receives Socrates' framing and the user's question
6. Plato challenges Socrates' position (3–5 sentences, identifies weaknesses/risks/hidden assumptions)
7. Gateway pushes Plato's completed response back to Socrates
8. Socrates synthesises the debate into a final 🏛️ AGORA ADVISORY for the user

File Structure

agora-openclaw/
├── openclaw.json                     # Main gateway configuration
├── workspace-socrates/               # Socrates' workspace
│   ├── AGENTS.md                     # Debate protocol instructions
│   ├── IDENTITY.md                   # Agent identity
│   ├── SOUL.md                       # Personality & character
│   ├── TOOLS.md                      # Tool usage guidelines
│   ├── USER.md                       # User context
│   ├── HEARTBEAT.md                  # Heartbeat config
│   └── skills/agora/SKILL.md         # Agora skill definition
└── workspace-plato/                  # Plato's workspace
    ├── AGENTS.md                     # Standing orders (critical analysis)
    ├── IDENTITY.md                   # Agent identity
    ├── SOUL.md                       # Personality & character
    ├── TOOLS.md                      # Tool usage guidelines
    ├── USER.md                       # User context
    └── HEARTBEAT.md                  # Heartbeat config

Setup

Prerequisites

• OpenClaw v2026.5.2+ installed locally
• OpenAI API key (for Socrates / GPT-5.5)
• Google AI API key (for Plato / Gemini 3.1 Flash)

Installation

1. Clone this repo:

   git clone https://github.com/somewisecrack/agora-openclaw.git
   

2. Copy workspace files to your OpenClaw config directory:

   cp -r workspace-socrates ~/.openclaw/workspace-socrates
   cp -r workspace-plato ~/.openclaw/workspace-plato
   

3. Merge openclaw.json into your existing ~/.openclaw/openclaw.json, or replace it entirely if starting fresh. Replace <YOUR_TOKEN_HERE> with your gateway auth token.

4. Set your API keys in ~/.openclaw/.env:

   OPENAI_API_KEY=sk-...
   GOOGLE_AI_API_KEY=AI...
   

5. Restart the gateway:

   launchctl stop ai.openclaw.gateway
   launchctl start ai.openclaw.gateway
   

6. Open the OpenClaw Control UI at http://127.0.0.1:18789/chat

Key Configuration Details

Viewing the Debate

After Socrates delivers his advisory, you can read the full Plato debate:

1. Go to Sessions in the left sidebar
2. Check the Global checkbox
3. Click the agent:plato:subagent:... row (the newest one)
4. You'll see Socrates' framing and Plato's critical response

Known Limitations

• Single-round debate: OpenClaw v2026.5.2 injects a hardcoded "send that user-facing update now" instruction when a subagent completes, which forces Socrates to deliver immediately rather than continuing the debate for multiple rounds. Multi-round debates would require a framework-level change.
• No exec tool: Disabled for Socrates to prevent SSL certificate errors on macOS. Agents rely on internal knowledge only.

License

MIT