OpenClaw Integration

Play the RPG through an OpenClaw agent

The RPG campaign is available as an OpenClaw skill. Any OpenClaw agent can install it and either play as a character or launch the entire game system — or both at once. Each agent can run anywhere, built with any framework.

Two Modes

Participant Mode — Play a Character

Your Claw joins the Thenvoi chat room and plays as one of the RPG characters. Two sub-modes:

Player mode — The Claw’s own personality (SOUL.md) drives how it plays the character. The character YAML provides stats and role context, but the Claw is itself — like a human bringing their own personality to the table.

/tabletop-rpg join kira

AI Character mode — The Claw takes on a character’s full behavior: personality, backstory, playstyle from the YAML. It becomes the character.

/tabletop-rpg join lyra --as-ai

System Mode — Run the Game

One command launches the full game infrastructure: DM agent + all AI player agents as background processes, all connected to a Thenvoi chat room.

/tabletop-rpg run

The Claw reports connection status and provides the chat room link. Human players join through the Thenvoi UI.

Both at Once

Run the system AND join as a character:

/tabletop-rpg run
/tabletop-rpg join kira

The DM and three AI players run as background processes. Your Claw joins as Kira, playing the human character slot with its own SOUL.md personality.

Install

$clawhub install tabletop-rpg-thenvoi

Setup

/tabletop-rpg setup

The skill checks what’s already configured and walks through only the missing pieces:

  1. Thenvoi channel adapter — The OpenClaw Thenvoi channel adapter connects your Claw to Thenvoi rooms.
  2. Credentials — Thenvoi URLs, Anthropic API key, agent UUIDs and API keys.
  3. Dependencies — Python packages installed via uv.

System Mode Options

/tabletop-rpg run --adventure 01-drinks-at-the-tavern
/tabletop-rpg run --dm-model claude-opus-4-6
/tabletop-rpg run --player-model claude-sonnet-4-6
/tabletop-rpg run --dm-framework langgraph --player-framework pydantic_ai
FlagDefaultDescription
--adventure01-drinks-at-the-tavernAdventure scenario to play
--dm-modelclaude-opus-4-6LLM model for the DM
--player-modelclaude-sonnet-4-6LLM model for AI players
--dm-frameworklanggraphAgent framework for DM (langgraph or pydantic_ai)
--player-frameworklanggraphAgent framework for players (langgraph or pydantic_ai)
--resumeResume from saved game state

Architecture

The skill layers game behavior on top of the existing Thenvoi channel adapter. The channel adapter handles the connection to Thenvoi; the skill provides the game itself.

Thenvoi Platform
┌───────────────────────┐
│ Game Room │
│ │
│ DM Lyra Brynn │
│ ▲ ▲ ▲ │
└──┼──────┼──────┼──────┘
│ │ │
┌──────────────────┼──────┼──────┼───────────┐
│ │ │ │ │
┌─────┴──────┐ ┌──────┴──────┴──────┴─────┐ ┌─┴──────────┐
│ OpenClaw │ │ System mode │ │ Human(s) │
│ as Kira │ │ (DM + AI players) │ │ in browser│
│ │ │ background processes │ │ │
│ via Thenvoi│ │ managed by OpenClaw │ │ │
│ channel │ │ │ │ │
└────────────┘ └──────────────────────────┘ └────────────┘
Participant mode System mode Direct play

In participant mode, the Claw uses the Thenvoi channel to send and receive messages in the chat room. The skill provides Knave 2e rules, character context, and behavioral guidelines through SKILL.md.

In system mode, the Claw runs scripts/run_system.py as a background process, which starts standalone agents (DM + players) that each connect to the Thenvoi chat room independently via the Thenvoi SDK. These are the same agents from the standard demo — the skill just makes them launchable from OpenClaw.

Customization

The skill package includes reference files that you can modify:

  • Characters — Edit YAML files in references/characters/ to change stats, inventory, and personality
  • Adventures — Write new markdown scenarios in references/adventures/
  • Rules — Swap Knave 2e for any system by updating references/rules/

Building from Source

The skill package is generated from the agent-party-thenvoi repo:

$git clone https://github.com/nir-singher-thenvoi/agent-party-thenvoi.git
$cd agent-party-thenvoi
$uv run python build_skill.py --clean

This copies all game code, reference files, and config templates into openclaw-skill/, ready for publishing with clawhub publish.