feat: support codex runner in reviewer container

- Build both agent-runner and codex-runner into container image
- Select runner path based on group.agentType at docker exec time
- Update README to role-centric overview, generic MoA references
- Add MoA reference opinions guidance to arbiter prompt

README.md

@@ -14,13 +14,15 @@ Tribunal arbiter system inspired by multi-agent consensus architectures.
 
 ## Overview
 
-A single unified service (`ejclaw`) manages three Discord bots in one process:
+A single unified service (`ejclaw`) runs a **Tribunal** of three roles:
 
-- **Codex-main** (`@codex`) — Owner agent. Handles user requests, writes code.
-- **Claude** (`@claude`) — Reviewer agent. Critically reviews owner's work, verifies design direction.
-- **Codex-review** (`@codex-review`) — Arbiter agent. Summoned on-demand to break deadlocks between owner and reviewer.
+| Role | Purpose | Default |
+|------|---------|---------|
+| **Owner** | Handles user requests, writes code | Codex (`@codex`) |
+| **Reviewer** | Critically reviews owner's work, verifies design direction | Claude Code (`@claude`) |
+| **Arbiter** | On-demand deadlock breaker between owner and reviewer | Codex (`@codex-review`) |
 
-All agent types and models are independently configurable per role via `.env`.
+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
 
@@ -60,22 +62,22 @@ User message
 
 ### Mixture of Agents (MoA)
 
-When enabled, the arbiter collects opinions from external models (Kimi, GLM, etc.) before rendering its verdict:
+When enabled, the arbiter collects opinions from configurable external reference models before rendering its verdict:
 
 ```
-Deadlock detected → MoA reference queries (Kimi + GLM, parallel)
+Deadlock detected → MoA reference queries (parallel, configurable via MOA_REF_MODELS)
   → Opinions injected into arbiter's prompt
-    → SDK arbiter (subscription-based) aggregates all perspectives
+    → Arbiter aggregates all perspectives
       → Final verdict: PROCEED / REVISE / RESET / ESCALATE
 ```
 
-No extra SDK processes. External references use lightweight API calls (Anthropic-compatible).
+No extra SDK processes. External references use lightweight OpenAI/Anthropic-compatible API calls.
 
 ## 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
+- **Mixture of Agents** — Configurable external reference models 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
@@ -107,8 +109,8 @@ Discord ──► SQLite (WAL) ──► GroupQueue ──┬──► Owner (ho
                                           ├──► Arbiter (on-demand, fresh session each time)
                                           │       │
                                           │   ┌───┴─── MoA (if enabled) ───┐
-                                          │   │ Kimi API ──► opinion       │
-                                          │   │ GLM API  ──► opinion       │
+                                          │   │ Ref model A ──► opinion    │
+                                          │   │ Ref model B ──► opinion    │
                                           │   │ → injected into prompt     │
                                           │   └────────────────────────────┘
                                           │       │

container/Dockerfile

@@ -52,18 +52,24 @@ RUN curl -fsSL https://bun.sh/install | bash
 # Create app directory
 WORKDIR /app
 
-# Copy host's agent-runner (same code, same SDK version)
-COPY runners/agent-runner/package*.json ./
-RUN bun install
-COPY runners/agent-runner/ ./
-RUN bun run build
+# Copy agent-runner (Claude Code)
+COPY runners/agent-runner/package*.json ./agent/
+RUN bun install --cwd ./agent
+COPY runners/agent-runner/ ./agent/
+RUN bun run --cwd ./agent build
+
+# Copy codex-runner
+COPY runners/codex-runner/package*.json ./codex/
+RUN bun install --cwd ./codex
+COPY runners/codex-runner/ ./codex/
+RUN bun run --cwd ./codex build
 
 # Create workspace directories
 RUN mkdir -p /workspace/project /workspace/group /workspace/global \
     /workspace/ipc/messages /workspace/ipc/input
 
-# Create entrypoint script
-RUN printf '#!/bin/bash\nset -e\ncat > /tmp/input.json\nbun /app/dist/index.js < /tmp/input.json\n' \
+# Entrypoint (fallback — persistent containers use docker exec with explicit runner path)
+RUN printf '#!/bin/bash\nset -e\ncat > /tmp/input.json\nbun /app/agent/dist/index.js < /tmp/input.json\n' \
     > /app/entrypoint.sh && chmod +x /app/entrypoint.sh
 
 # Set ownership to node user (non-root)

prompts/arbiter-paired-room.md

@@ -22,6 +22,15 @@ You have been summoned because the owner and reviewer reached a deadlock after m
   - The situation cannot be resolved without user input, regardless of technical agreement
   - The same NEEDS_CONTEXT or BLOCKED is repeated after a prior PROCEED — this means your PROCEED did not resolve the issue
 
+## MoA (Mixture of Agents) Reference Opinions
+
+You may receive reference opinions from external models appended to your prompt. When present:
+
+- **Cite them explicitly** in your verdict — e.g., "Reference model A agrees that...", "Reference model B raises a concern about..."
+- **Cross-reference** their opinions against the owner/reviewer conversation and code evidence
+- **Resolve conflicts** — if reference opinions disagree with each other or with owner/reviewer, state which view you adopt and why
+- Do NOT blindly follow reference opinions — they are inputs to your judgment, not authorities
+
 ## Rules
 
 - Base your verdict on evidence (code, test output, logs), not on who said what first

src/container-runner.ts

@@ -484,7 +484,11 @@ export async function runReviewerContainer(args: {
       }
     }
   }
-  execArgs.push(containerName, 'bun', '/app/dist/index.js');
+  const isCodexAgent = (group.agentType || 'claude-code') === 'codex';
+  const runnerPath = isCodexAgent
+    ? '/app/codex/dist/index.js'
+    : '/app/agent/dist/index.js';
+  execArgs.push(containerName, 'bun', runnerPath);
 
   return new Promise<AgentOutput>((resolve) => {
     const proc = spawn(CONTAINER_RUNTIME_BIN, execArgs, {