memory-hybrid

🦞 Clawdboss — lightsailvr fork

Pre-hardened, multi-agent OpenClaw setup — customized for LightSailVR.

Forked from NanoFlow-io/clawdboss and extended with a full five-agent squad, Notion integration, Gemini-powered research, LinkedIn content workflows, and a dedicated senior dev agent covering Unity, Swift, web, and Python.

One script to go from zero to a fully configured, secured, multi-agent AI assistant on Discord — with prompt injection defense, WAL Protocol context persistence, and best practices baked in.

What's Different in This Fork

| Feature | Upstream | This Fork | |---------|----------|-----------| | Full Squad | Main + Comms + Research + Security | + Dev agent (C#/Unity, Swift, Web, Python) | | Research agent model | Claude Sonnet 4.5 | Gemini 2.5 Pro (deep research, grounding) | | Comms agent | Generic communications | LinkedIn-first, XR/entertainment/storytelling/3D/AI topics, writing style baked in at setup | | Notion | Not included | Notion MCP — main agent reads/writes your workspace | | Dev MCP servers | Not included | Unity MCP + GitHub MCP for live project context | | TOOLS.md | Generic for all agents | Type-specific starter notes for Dev agent (Unity, Xcode, GitHub) |

What You Get

Five-agent squad — Main + Comms + Research (Gemini) + Security + Dev
Security-first — Prompt injection defense, anti-loop rules, content tagging, credential isolation
WAL Protocol — Write-Ahead Log for corrections, decisions, and details that survive context loss
Working Buffer — Danger zone logging to survive context compaction without losing work
Discord integration — Bot bound to your server with channel-per-agent routing
Telegram integration — Chat via Telegram bot with DM and group support
Web dashboard — ClawSuite Console with chat, file browser, terminal, cost analytics (optional)
Env-based secrets — All API keys in .env, never in config files
Notion workspace — Main agent connected to your Notion pages and databases
Gemini research — Research agent uses Gemini 2.5 Pro with native grounding for up-to-date information
LinkedIn content — Comms agent pre-loaded with your topics and writing style
Dev tooling — Senior engineer agent with Unity MCP + GitHub MCP for live project context
Knowledge graph — Graphthulhu for structured memory across agents (optional)
API discovery — ApiTap intercepts web traffic to teach agents how APIs work (optional)
Web scraping — Scrapling for anti-bot-bypassing data extraction (optional)
Browser automation — Playwright MCP for full GUI workflows (optional)
GitHub integration — Issues, PRs, CI/CD via gh CLI (optional)
Token compression — OCTAVE protocol for 3-20x compression in multi-agent handoffs (optional)
Observability — Clawmetry dashboard for token costs, sessions, live message flow (optional)
Security suite — ClawSec for file integrity, advisory feed, malicious skill detection (optional)
Continuous learning — Self-Improving Agent captures errors and lessons across sessions (optional)
Skill discovery — Find Skills helps agents install new capabilities on-the-fly (optional)

Quick Start

# 1. Clone this repo
git clone https://github.com/lightsailvr/clawdboss.git
cd clawdboss

# 2. Run the setup wizard
./setup.sh

The wizard auto-installs all dependencies (Node.js 22, Python, git, build-essential, OpenClaw) and walks you through everything else.

What You'll Need Ready

An LLM provider — one of:
GitHub Copilot subscription (cheapest — uses copilot-api proxy)
OpenAI API key (sk-...)
Anthropic API key (sk-ant-...)
A Gemini API key — free at aistudio.google.com — powers the Research agent
A Discord bot token — create one here (if using Discord)
A Telegram bot token — create one via @BotFather (if using Telegram)
A Notion integration token — create one here (for Notion MCP)
Optional: Brave Search API key, ElevenLabs API key, OpenAI key for image gen/whisper

The setup wizard will: 1. Ask about you — name, role, what you do, how you want to use your agent 2. Ask about your agent — name, personality, mission, domain expertise 3. Ask about your comms agent — topic focus and writing style (pre-filled with XR/entertainment/storytelling/3D/AI defaults) 4. Prompt for your API keys and Discord/Telegram/Notion credentials 5. Create your .env file (gitignored, never committed) 6. Generate openclaw.json with ${VAR} references to your .env 7. Create agent workspaces with security rules + WAL Protocol pre-baked 8. Offer optional tools: Graphthulhu, ApiTap, Scrapling, GitHub, Playwright, OCTAVE, Humanizer, Self-Improving Agent, Find Skills, Healthcheck, Clawmetry, ClawSec, Notion MCP, Unity MCP, GitHub MCP 9. Start the gateway

Your agent starts its first session knowing who you are, what you do, and what personality it should have — zero manual config file editing needed.

Configuration Tiers

| Tier | Agents | Best For | |------|--------|----------| | Solo | Main only | Personal assistant, simple setups | | Team | Main + 1-2 specialists | Small business, multiple workflows | | Full Squad | Main + Comms + Research + Security + Dev | Full operations center |

The Five-Agent Squad

| Agent | Model | Role | |-------|-------|------| | Main | Your choice | Personal assistant, orchestrator, Notion-connected | | Comms | Your choice | LinkedIn content, XR/entertainment/storytelling/3D/AI topics, your writing style | | Research | Gemini 2.5 Pro | Deep research with grounding — always up to date | | Security | Your choice | Infrastructure hardening, prompt injection defense, audits | | Dev (Forge) | Your choice | Senior engineer — C#/Unity, Swift/SwiftUI, Web, Python; Unity MCP + GitHub MCP |

File Structure

clawdboss/
├── README.md
├── setup.sh                    # Interactive setup wizard
├── .env.example                # Template showing required variables
├── .gitignore                  # Protects secrets
├── templates/
│   ├── openclaw.template.json  # Config with ${VAR} placeholders
│   ├── workspace/              # Main agent workspace files
│   │   ├── AGENTS.md           # Operating rules + WAL Protocol + security
│   │   ├── SOUL.md
│   │   ├── USER.md
│   │   ├── TOOLS.md
│   │   ├── IDENTITY.md
│   │   └── HEARTBEAT.md
│   └── agents/                 # Specialist agent templates
│       ├── comms/
│       │   ├── SOUL.md         # LinkedIn-first, topic + style placeholders
│       │   └── WRITING-STYLE.md # Brand voice bible, populated at setup
│       ├── research/
│       ├── security/
│       └── dev/                # Senior dev (C#/Unity, Swift, Web, Python)
│           ├── SOUL.md
│           └── TOOLS.md        # Starter notes: Unity, Xcode, GitHub
└── docs/
    ├── security.md             # Security architecture + WAL Protocol overview
    ├── customization.md        # How to customize your setup
    ├── recommended-tools.md    # Vetted ecosystem tools (Clawmetry, ClawSec)
    └── octave.md               # OCTAVE protocol guide

Notion Setup

After running the wizard:

1. Go to notion.so/my-integrations and create an integration 2. Your token is already in ~/.openclaw/.env as NOTION_API_KEY 3. Open each Notion page or database you want the agent to access → Share → invite your integration by name 4. The main agent can now read and write those pages via the Notion MCP server

Research Agent — Gemini 2.5 Pro

The Research agent is hardcoded to google/gemini-2.5-pro regardless of your main LLM provider. This gives it:

Native grounding — real-time web search built into the model
1M token context window — handles large documents and long research chains
Up-to-date information — not limited to training data cutoff

You need a Gemini API key (free at aistudio.google.com). The wizard will prompt for it and it's stored in GEMINI_API_KEY in your .env.

Dev Agent — Unity MCP + GitHub MCP

The Dev agent (default name: Forge) has two optional MCP servers that give it live project context:

Unity MCP — connects to a running Unity Editor:

Reads scenes, assets, console logs
Can run Editor commands
Install the companion Unity package: Window > Package Manager > + > Add from git URL > https://github.com/CoderGamester/mcp-unity.git

GitHub MCP — official GitHub API access:

Read issues, PRs, code, and CI results directly
Requires gh CLI authenticated (gh auth login)

Both are offered during setup and are optional.

Comms Agent — LinkedIn & Writing Style

The Comms agent is built for LinkedIn-first content across XR, entertainment, storytelling, 3D, and AI. During setup the wizard collects:

Topics (default: XR, entertainment, storytelling, 3D, AI)
Writing style description (default: conversational, storytelling-first, direct)
Sample posts (optional — paste 1-2 examples to anchor the voice)

These are baked into SOUL.md and WRITING-STYLE.md in the comms workspace at deploy time. The agent reads WRITING-STYLE.md before drafting any content. Add more sample posts to it at any time.

For LinkedIn posting automation, use the Playwright MCP (already in the ecosystem) — the agent can draft, review, and assist with posting via browser automation.

Context Persistence (WAL Protocol + 3-Layer Memory)

Clawdboss agents don't lose your corrections and decisions when context resets:

3-Layer Memory — L1 (Brain, loaded every turn) → L2 (Memory, searched semantically) → L3 (Reference, opened on demand)
SESSION-STATE.md — Agent writes important details here BEFORE responding (Write-Ahead Log)
Working Buffer — At ~60% context, every exchange is logged to survive compaction
Compaction Recovery — Agent reads buffer + state files after context loss, never asks "what were we doing?"

Security

All API keys are stored in ~/.openclaw/.env and referenced via ${VAR_NAME} syntax in the config. Keys never appear in JSON config files.

All agents come with prompt injection defense, anti-loop rules, and external content isolation pre-baked.

See docs/security.md for the full security architecture.

Recommended Ecosystem Tools

| Tool | Purpose | Install Method | |------|---------|---------------| | Graphthulhu | Knowledge graph memory | Binary / cargo | | ApiTap | API discovery from browser traffic | npm install -g @apitap/core | | Scrapling | Anti-bot web scraping | pip install scrapling | | Playwright MCP | Browser automation + LinkedIn workflows | clawhub install | | GitHub | Issues, PRs, CI/CD via gh CLI | clawhub install + gh CLI | | Notion MCP | Notion workspace integration | Auto-installed via wizard | | OCTAVE | 3-20x token compression | pip install octave-mcp | | Clawmetry | Observability dashboard | pip install clawmetry | | ClawSec | File integrity + advisory feed | Git clone | | Self-Improving Agent | Continuous learning from corrections | clawhub install | | Find Skills | Discover new skills on-the-fly | clawhub install |

After Install

# Start OpenClaw
openclaw gateway start
openclaw status

# Start Copilot proxy (if using GitHub Copilot)
tmux new-session -d -s copilot 'npx copilot-api start --port 4141'

Then open your Discord server, Telegram bot, or ClawSuite Console and start chatting.

Requirements

Ubuntu 22.04+ (or any Linux with bash)
Node.js 22+
2GB+ RAM recommended
LLM provider (GitHub Copilot, OpenAI, Anthropic, or others)
Gemini API key (free) — for Research agent
Discord and/or Telegram bot token

Credits

Based on Clawdboss by NanoFlow.io — MIT license.

License

MIT — see LICENSE for details.

Summary

Install to Claude Code