research-committee
v1.1.0
Published
Multi-agent research idea incubation system for OpenClaw. Oracle + committee of 4 sub-agents to turn vague research ambitions into sharp, falsifiable research questions.
Maintainers
Readme
Research Committee — OpenClaw Plugin
A multi-agent research-idea incubation system for OpenClaw.
Turn a vague research ambition into one concrete, falsifiable, top-tier-plausible research question — by convening a committee of specialized agents.
🎯 What it does
Researcher → Oracle → Committee
├── 🔍 literature-scout (prior work, novelty)
├── ⚙️ feasibility-analyst (path, bottlenecks)
├── 💀 red-team-critic (attack, kill/revise)
└── 🎯 venue-strategist (publication path)
→ FinalTopicSpecOracle convenes a committee through a blackboard. The researcher gets one sharp research question — not a list of directions.
✨ Features
- Oracle — researcher's primary interface; runs the full committee workflow
- Literature Scout — finds prior work, judges true vs. fake novelty
- Feasibility Analyst — critical path, bottlenecks, 6-month minimum deliverable
- Red Team Critic — structured attacks, kill/revise verdict
- Venue Strategist — minimum publishable unit, publication paths
- Blackboard system — agents communicate through structured JSON packs
- Evolvable thesis —
evolving_thesis.mdpersists across sessions
📦 Installation
Prerequisites
- OpenClaw
>= 2026.4.0 - A model provider configured (Minimax, OpenAI, Anthropic, etc.)
Install the plugin
openclaw plugins install research-committeeOr manually
git clone https://github.com/zleung9/openclaw-research-committee.git
cd research-committee-plugin
# Copy agent templates to your OpenClaw workspace
cp -r agents/* ~/.openclaw/
openclaw gateway restart🚀 Usage
After installation, describe your research ambition to Oracle:
"I'm working on [your field]. I want to publish something big but I only have a vague direction."
Oracle will guide you through clarification, then run the full committee workflow.
📁 Directory Structure
research-committee-plugin/
├── openclaw.plugin.json # Plugin manifest
├── package.json # npm package
├── README.md # This file
├── LICENSE # MIT
├── skills/
│ └── research-committee/ # Research committee workflow skill
│ └── SKILL.md
└── agents/
├── oracle/ # Oracle workspace template
│ ├── AGENTS.md
│ ├── IDENTITY.md
│ ├── SOUL.md
│ ├── TOOLS.md
│ └── blackboard/
│ └── SCHEMAS.md # All pack JSON schemas
├── literature-scout/
├── feasibility-analyst/
├── red-team-critic/
└── venue-strategist/🔧 Configuration
Agent config
Add to openclaw.json agents.list:
{
"id": "oracle",
"name": "oracle",
"workspace": "~/.openclaw/workspace-oracle",
"agentDir": "~/.openclaw/agents/oracle/agent",
"model": "minimax-portal/MiniMax-M2.7"
},
{
"id": "literature-scout",
"name": "literature-scout",
"workspace": "~/.openclaw/workspace-literature-scout",
"agentDir": "~/.openclaw/agents/literature-scout/agent",
"model": "minimax-portal/MiniMax-M2.7"
},
{
"id": "feasibility-analyst",
"name": "feasibility-analyst",
"workspace": "~/.openclaw/workspace-feasibility-analyst",
"agentDir": "~/.openclaw/agents/feasibility-analyst/agent",
"model": "minimax-portal/MiniMax-M2.7"
},
{
"id": "red-team-critic",
"name": "red-team-critic",
"workspace": "~/.openclaw/workspace-red-team-critic",
"agentDir": "~/.openclaw/agents/red-team-critic/agent",
"model": "minimax-portal/MiniMax-M2.7"
},
{
"id": "venue-strategist",
"name": "venue-strategist",
"workspace": "~/.openclaw/workspace-venue-strategist",
"agentDir": "~/.openclaw/agents/venue-strategist/agent",
"model": "minimax-portal/MiniMax-M2.7"
}🔒 Security Notes
- Never commit credentials,
sessions/, or personalmemory/to the repo - The plugin contains templates only — user fills in their own API keys
- Each agent has its own isolated workspace
📄 License
MIT
