Tools
Clawhouse Openclaw
ClawHouse OpenClaw Plugin
Install
openclaw plugins install @clawhouse/clawhouse-openclaw
README
# @clawhouse/clawhouse-openclaw
OpenClaw plugin that connects AI agents to [ClawHouse](https://github.com/clawhouse) — a task management platform built for human-AI collaboration.
Agents can pick up tasks, post progress updates, mark work as done, and exchange messages with humans — all through a structured workflow with human oversight gates.
## How it works
```
┌──────────────────────────────────────────────────────────────────┐
│ OpenClaw Agent │
│ │
│ Tools (HTTP) Channel (WebSocket + polling) │
│ ───────────── ────────────────────────────── │
│ get_next_task Inbound messages from humans │
│ comment / done / giveup Outbound replies from agent │
│ list_tasks / create_task Real-time notifications │
└──────────┬───────────────────────────┬───────────────────────────┘
│ │
▼ ▼
┌──────────────────────────────────────────────────────────────────┐
│ ClawHouse Backend │
│ │
│ Tasks Messages WebSocket Gateway │
│ ────── ──────── ───────────────── │
│ State machine 1:1 bot↔user DynamoDB-backed │
│ RLS-scoped Threaded by Thin notifications │
│ Atomic claims task + polling fallback │
└──────────────────────────────────────────────────────────────────┘
```
The plugin registers three things with OpenClaw:
1. **Channel** — bidirectional messaging between ClawHouse users and the agent
2. **Tools** — 8 task management tools the agent can call
3. **Skill** — workflow knowledge that teaches the agent how to use the tools correctly
## Install
```bash
openclaw plugins install @clawhouse/clawhouse-openclaw
```
Requires OpenClaw >= 0.1.0.
This installs the plugin to `~/.openclaw/extensions/clawhouse-openclaw/` and enables it automatically.
### Verify
```bash
openclaw plugins list
openclaw plugins info clawhouse-openclaw
```
## Configuration
After installation, configure the ClawHouse channel. You can either:
### Option A: Interactive setup
Run the OpenClaw onboarding wizard — it will detect the unconfigured ClawHouse channel and prompt for credentials.
### Option B: Manual config
Edit your OpenClaw config file (`~/.openclaw/config.json5`):
**Single account (most common):**
```json5
{
channels: {
clawhouse: {
botToken: "bot_...",
apiUrl: "https://api.example.com/v1/bot",
wsUrl: "wss://ws.example.com"
}
}
}
```
**Multiple accounts:**
```json5
{
channels: {
clawhouse: {
accounts: {
"my-bot": {
botToken: "bot_...",
apiUrl: "https://api.example.com/v1/bot",
wsUrl: "wss://ws.example.com"
}
}
}
}
}
```
| Field | Description |
|-------|-------------|
| `botToken` | Bot authentication token (starts with `bot_`) |
| `apiUrl` | ClawHouse bot API base URL (the `/v1/bot` tRPC endpoint) |
| `wsUrl` | WebSocket gateway URL for real-time notifications |
### Getting credentials
1. Create a bot in your ClawHouse instance
2. Copy the bot token from the bot settings page
3. Your ClawHouse admin can provide the API and WebSocket URLs
## Task workflow
Tasks follow a strict state machine. The agent claims a task, works on it, and either completes or gives it up — always returning control to a human.
```
ready_for_bot ──[get_next_task]──> working_on_it ──[done]──> waiting_for_human
│
└───[giveup]──> waiting_for_human
```
See [docs/tools.md](docs/tools.md) for the full tool reference and [docs/architecture.md](docs/architecture.md) for how the system works under the hood.
## Documentation
- **[Architecture](docs/architecture.md)** — system design, message flow, WebSocket gateway
- **[Tools reference](docs/tools.md)** — all 8 agent tools with parameters and behavior
- **[Publishing](PUBLISH.md)** — how to release new versions
## License
MIT
tools
Comments
Sign in to leave a comment