EJClaw

Tribunal multi-agent AI assistant (Owner + Reviewer + Arbiter) over Discord with autonomous paired review and Mixture of Agents.

Originally derived from qwibitai/nanoclaw, now independently maintained as EJClaw.
Prompt design inspired by Q00/ouroboros and garrytan/gstack, adapted for EJClaw's Discord and dual-service workflow.
Tribunal arbiter system inspired by multi-agent consensus architectures.

Overview

A single unified service (ejclaw) runs a Tribunal of three roles while
managing three Discord bots in one process:

The identity layer is role-fixed:

• Owner bot — Handles the owner turn output slot.
• Reviewer bot — Handles the reviewer turn output slot.
• Arbiter bot — Handles the arbiter turn output slot.

Each role's agent type and model are independently configurable via .env
(OWNER_AGENT_TYPE, REVIEWER_AGENT_TYPE, ARBITER_AGENT_TYPE, *_MODEL).
Three Discord bots provide the identity layer — which bot speaks is determined
by the active role, not hardcoded.

Room Assignment Model

Per-room routing now uses an explicit assignment model:

• room_settings is the room-level source of truth (SSOT)
• Each room stores:
  • room_mode: single or tribunal
  • owner_agent_type: codex or claude-code
• Public room assignment uses assign_room
• Legacy register_group public interfaces were removed
• registered_groups remains as a materialized capability/read-model layer and legacy fallback, not the authoritative room configuration

Operationally:

• single → one owner bot
• tribunal → per-room owner + globally configured reviewer + optional arbiter

This means tribunal is no longer inferred from “two bots registered on one room”; it is an explicit room setting.

Tribunal Flow

User message
  → Owner responds (implementation, answer, etc.)
    → Reviewer auto-triggered (critical review, design check)
      → Verdict:
          DONE              → Owner finalizes → Task completed → @user ✅
          DONE_WITH_CONCERNS → Owner addresses feedback → loop
          BLOCKED/NEEDS_CONTEXT
            ├─ Arbiter enabled → Arbiter judges → PROCEED/REVISE/RESET/ESCALATE
            └─ Arbiter disabled → Escalate to user → @user ⚠️
      → Owner BLOCKED/NEEDS_CONTEXT → Arbiter (same path as reviewer)
      → Deadlock (3+ round trips without progress)
          → Arbiter summoned → binding verdict → loop resumes

Mixture of Agents (MoA)

When enabled, the arbiter collects opinions from external models (Kimi, GLM, etc.) before rendering its verdict:

Deadlock detected → MoA reference queries (Kimi + GLM, parallel)
  → Opinions injected into arbiter's prompt
    → SDK arbiter (subscription-based) aggregates all perspectives
      → Final verdict: PROCEED / REVISE / RESET / ESCALATE

No extra SDK processes. External references use lightweight API calls (Anthropic-compatible).

Features

• Tribunal 3-agent system — Owner/reviewer/arbiter with on-demand deadlock resolution
• Discord-independent communication — Agent-to-agent data flows directly via DB, Discord is display-only
• Mixture of Agents — External model opinions (Kimi, GLM) enrich arbiter verdicts
• Per-role model selection — OWNER_MODEL, REVIEWER_MODEL, ARBITER_MODEL + effort + fallback toggle
• Container-isolated reviewer — Persistent Docker container with read-only source mount
• Global failover — Account-level Claude failure → all channels switch to codex, auto-recovers
• Post-approval change detection — Re-triggers review if owner modifies code after approval
• Auto user notification — @mention on task completion (✅ done, ⚠️ escalated)
• Loop protection — Deadlock threshold, merge_ready oscillation guard, arbiter re-invocation limit
• Voice transcription — Groq Whisper (primary) / OpenAI Whisper (fallback)
• Token rotation — Multi-account Claude/Codex rotation on rate limits
• Kimi usage dashboard — Coding plan 5h/7d usage displayed alongside Claude/Codex
• MCP integration — Memento (persistent memory) + EJClaw host tools
• Session persistence — Separate sessions per role (owner/reviewer/arbiter)
• Scheduled tasks — Cron/interval/once via MCP tool
• Mid-turn steering — Inject follow-up messages while agent is working
• Bun runtime — Native SQLite (bun:sqlite), fast startup, no native addon builds

Architecture

Discord ──► SQLite (WAL) ──► GroupQueue ──┬──► Owner (host process)
                                          │       │
                                          │       ▼ (auto-trigger)
                                          ├──► Reviewer (Docker container, :ro mount)
                                          │       │
                                          │   Verdict routing
                                          │       ├─ DONE → change detection → finalize or re-review
                                          │       ├─ BLOCKED → Arbiter (if enabled) or User
                                          │       └─ Feedback → Owner (loop)
                                          │
                                          ├──► Arbiter (on-demand, fresh session each time)
                                          │       │
                                          │   ┌───┴─── MoA (if enabled) ───┐
                                          │   │ Kimi API ──► opinion       │
                                          │   │ GLM API  ──► opinion       │
                                          │   │ → injected into prompt     │
                                          │   └────────────────────────────┘
                                          │       │
                                          │   PROCEED/REVISE/RESET/ESCALATE
                                          │
                                     IPC polling ◄── follow-up messages
                                          │
                              ┌────────── Router ──────────┐
                              ▼                            ▼
                   paired_turn_outputs           Discord (display only)
                   (agent ↔ agent data)      (user observation, @mention)

Setup

Prerequisites

• Linux (Ubuntu 22.04+) or macOS
• Bun 1.3+
• Docker (required for reviewer container isolation)
• Claude Code CLI
• Codex CLI (npm install -g @openai/codex)
• Discord bot tokens (3: owner, reviewer, arbiter)

Install

git clone https://github.com/phj1081/EJClaw.git
cd EJClaw
bun install
bun run build:runners
bun run build
bun run build:container          # Build reviewer Docker image

Documentation

• Architecture — Data flow, room model, container isolation, key files
• Configuration — Full .env reference, debugging paths

Environment

All configuration in a single .env file:

# Discord bots (canonical role-fixed names)
DISCORD_OWNER_BOT_TOKEN=         # Owner bot
DISCORD_REVIEWER_BOT_TOKEN=      # Reviewer bot
DISCORD_ARBITER_BOT_TOKEN=       # Arbiter bot

# Old service-based token names are no longer accepted.
# Rename existing values to the canonical role-based keys above.

# Agent types
OWNER_AGENT_TYPE=codex            # codex | claude-code
REVIEWER_AGENT_TYPE=claude-code   # claude-code | codex
ARBITER_AGENT_TYPE=codex          # codex | claude-code (optional, enables 3rd agent)

# Per-role model overrides
OWNER_MODEL=gpt-5.4
REVIEWER_MODEL=claude-opus-4-6
ARBITER_MODEL=gpt-5.4

# API keys
CLAUDE_CODE_OAUTH_TOKEN=          # Claude Code OAuth token
CLAUDE_CODE_OAUTH_TOKENS=         # Comma-separated for multi-account rotation
GROQ_API_KEY=                     # Voice transcription (Groq Whisper)

# Mixture of Agents (MoA)
MOA_ENABLED=true
MOA_REF_MODELS=kimi,glm
MOA_KIMI_MODEL=kimi-k2.5
MOA_KIMI_BASE_URL=https://api.kimi.com/coding
MOA_KIMI_API_KEY=sk-kimi-xxx
MOA_KIMI_API_FORMAT=anthropic
MOA_GLM_MODEL=glm-5.1
MOA_GLM_BASE_URL=https://open.bigmodel.cn/api/anthropic
MOA_GLM_API_KEY=xxx
MOA_GLM_API_FORMAT=anthropic

Deploy

bun run deploy

Development

bun run build                # Build main project
bun run build:runners        # Install + build both runners
bun run build:container      # Rebuild reviewer Docker image
bun run dev                  # Dev mode with hot reload
bun test                     # Run tests

License

MIT — Originally derived from qwibitai/nanoclaw