🐺 OpenClaw Wolfpack

Deploy a collaborative AI agent team on OpenClaw in 60 seconds.

English • 中文 • 日本語 • Français • Deutsch • Español • Русский • Italiano

---

What Is This?

Wolfpack is a one-command multi-agent skill for OpenClaw. It deploys 8 specialized AI agents as a coordinated team — complete with identities, quality gates, and workflow templates.

Think of it as deploying an entire AI research lab with one script.

🐺 The Pack

Agent	Role	What It Does
🧠 Planner	Conductor	Decomposes tasks, tracks progress, coordinates the team
💡 Ideator	Creative Engine	Generates research ideas, assesses novelty
🎯 Critic	Taste Gate	Evaluates ideas with SHARP framework (score ≥18 to pass)
📚 Surveyor	Knowledge Base	Literature search, research gap identification
💻 Coder	Engineer	Algorithm implementation, experiment execution
✍️ Writer	Craftsman	Academic writing, LaTeX formatting
🔍 Reviewer	Gatekeeper	Simulated peer review, weakness diagnosis
📰 Scout	Radar	Daily paper digest, trend monitoring

🔥 Key Features

• One-command deploy — 8 agents provisioned with a single script
• SHARP quality gates — Critic enforces taste standards (no mediocre ideas proceed)
• Adversarial collaboration — Ideator↔Critic and Writer↔Reviewer creative tension
• 4 workflow templates — Paper Pipeline, Brainstorm, Rebuttal, Daily Digest
• Two deployment modes — Channel (Telegram/Discord/Feishu/etc.) or Local (CLI)
• Safe merge — Appends to existing config, never overwrites your main agent

---

Quick Start

Install

git clone https://github.com/LeoYeAI/openclaw-wolfpack.git
cd openclaw-wolfpack
chmod +x scripts/setup.sh

Deploy

# Interactive
./scripts/setup.sh

# Local mode (no channel needed)
./scripts/setup.sh --mode local --model anthropic/claude-sonnet-4-5

# Channel mode (e.g., Telegram)
./scripts/setup.sh --mode channel --channel telegram --group-id -1001234567890

Use

# Local mode
openclaw chat planner
# → "Start the paper-pipeline workflow for multi-agent reasoning efficiency"

# Channel mode
openclaw gateway restart
# → @planner in your group chat

---

Deployment Modes

Local Mode (agentToAgent)

Agents communicate directly via OpenClaw's internal messaging. No chat platform needed.

./scripts/setup.sh --mode local

Channel Mode

Each agent binds to a chat group. Interact via @mentions.

./scripts/setup.sh --mode channel --channel telegram --group-id -1001234567890

Supported channels: telegram | discord | feishu | whatsapp | slack

---

Workflows

📋 Paper Pipeline

Full research-to-submission flow: Survey → Ideation → SHARP Gate → Design → Code → Experiments → Write → Review → Submit

💡 Brainstorm

Rapid idea generation session with structured SHARP filtering.

🔄 Rebuttal

Coordinated response to reviewer comments with parallel task execution.

📰 Daily Digest

Automated paper monitoring with collision warnings and trend analysis.

---

SHARP Framework

The Critic's quality gate scoring:

Dimension	Question
Sharpness	Can you explain WHY it works in one sentence?
Horizon	Will this problem still matter in 5 years?
Asymmetry	Do you have a unique angle others lack?
Resistance	Can the toughest reviewer break your argument?
Parsimony	Is the method elegantly simple?

Score tiers: 🏆 Exquisite (23-25) · 🟢 Refined (18-22) · 🟡 Raw (13-17) · 🔴 Bland (≤12)

Ideas must score ≥18 to proceed past the idea phase.

---

Advanced Configuration

Per-Agent Models

./scripts/setup.sh --mode local \
  --model anthropic/claude-sonnet-4-5 \
  --model-map 'critic=anthropic/claude-sonnet-4-5,writer=anthropic/claude-sonnet-4-5'

Per-Agent Groups

./scripts/setup.sh --mode channel --channel telegram \
  --group-map 'planner=-100111,coder=-100222,scout=-100333'

Dry Run

./scripts/setup.sh --mode local --dry-run

---

Project Structure

openclaw-wolfpack/
├── SKILL.md                         # AgentSkill manifest
├── scripts/
│   └── setup.sh                     # One-command deployment
├── references/
│   ├── workflows.md                 # Workflow details
│   └── sharp-framework.md           # SHARP scoring guide
└── assets/
    ├── agents.yaml                  # Agent registry
    ├── agent-souls/                 # Per-agent identity files
    │   ├── planner/  (soul.md, agent.md, user.md)
    │   ├── ideator/
    │   ├── critic/
    │   ├── surveyor/
    │   ├── coder/
    │   ├── writer/
    │   ├── reviewer/
    │   └── scout/
    └── workflows/                   # Workflow templates
        ├── paper-pipeline.md
        ├── brainstorm.md
        ├── rebuttal.md
        └── daily-digest.md

---

License

MIT

---

Powered by MyClaw.ai — Your 24/7 Personal Agent