桌面游戏AI 是 AI Skill Hub 本期精选Agent工作流之一。综合评分 7.5 分,整体质量较高。我们推荐使用将其纳入你的 AI 工具库,帮助提升工作效率。
桌面游戏AI 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
桌面游戏AI 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
# 方式一:pip 安装(推荐)
pip install open-tabletop-gm
# 方式二:虚拟环境安装(推荐生产环境)
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install open-tabletop-gm
# 方式三:从源码安装(获取最新功能)
git clone https://github.com/Bobby-Gray/open-tabletop-gm
cd open-tabletop-gm
pip install -e .
# 验证安装
python -c "import open_tabletop_gm; print('安装成功')"
# 命令行使用
open-tabletop-gm --help
# 基本用法
open-tabletop-gm input_file -o output_file
# Python 代码中调用
import open_tabletop_gm
# 示例
result = open_tabletop_gm.process("input")
print(result)
# open-tabletop-gm 配置文件示例(config.yml) app: name: "open-tabletop-gm" debug: false log_level: "INFO" # 运行时指定配置文件 open-tabletop-gm --config config.yml # 或通过环境变量配置 export OPEN_TABLETOP_GM_API_KEY="your-key" export OPEN_TABLETOP_GM_OUTPUT_DIR="./output"
An LLM-agnostic Game Master framework for persistent tabletop RPG campaigns. Built to run on any model available through OpenCode, locally hosted models via LM Studio, or any other LLM service.
D&D 5e is included as the reference system. Any other tabletop RPG can be added by writing a system module — see SYSTEM-PORTING.md.
Using Claude? This framework was extracted from claude-dnd-skill, a Claude Code-specific version with deeper integration. If you're running Claude Code, that repo will give you a more optimised experience.

---
The core scripts have no dependencies. The optional display companion requires:
cd display
pip3 install -r requirements.txt
opencode.ai — supports Anthropic, OpenAI, Google, Ollama, LM Studio, and any OpenAI-compatible endpoint.
Improvised campaign — GM generates world and narrative arc:
/gm new my-campaign # generates world, factions, NPCs, optional story arc
/gm character new # create a character
/gm load my-campaign # start a session
Structured campaign — import a pre-written module:
/gm import module.pdf my-campaign # extract structure and build campaign files
/gm load my-campaign # start a session — GM enforces the arc beats
Once loaded, type naturally — no /gm prefix needed. The GM interprets all input as in-game action.
---
Point OpenCode at this skill by adding the following to your OpenCode config (~/.config/opencode/opencode.json):
{
"instructions": [
"/path/to/open-tabletop-gm/no_think.md",
"/path/to/open-tabletop-gm/paths.md",
"/path/to/open-tabletop-gm/SKILL-commands.md",
"/path/to/open-tabletop-gm/SKILL-branches.md"
]
}
SKILL.md (the GM persona) is not loaded at startup — it is read from disk the first time a session is loaded, keeping the standing system prompt lean (~2,300 tokens). See docs/LLM-GUIDE.md for why this matters on smaller models.
For a local model via LM Studio, add your provider config:
{
"provider": {
"lmstudio": {
"npm": "@ai-sdk/openai-compatible",
"name": "LM Studio (local)",
"options": {
"baseURL": "http://localhost:1234/v1"
},
"models": {
"your-model-id": {
"name": "Your Model Name"
}
}
}
}
}
| Option | How |
|---|---|
| **TV — Cast tab** | Chrome → three-dot menu → Cast → Cast tab; select your Chromecast or smart TV |
| **TV — Screen mirror** | macOS: Control Centre → Screen Mirroring → Apple TV / AirPlay receiver |
| **iPad / tablet** | Start with --lan, open http://<your-ip>:5001 in Safari or Chrome; works in landscape |
| **Second monitor** | Open http://localhost:5001 in a browser window and drag it to the second display |
HTTP is the default. Use --tls only on public or untrusted networks. When passed: - A self-signed cert is auto-generated if cert.pem is not present - A plain HTTP server starts on :8080 to serve cert.pem for download - Per-platform install instructions are printed to the terminal (iOS, Android, Mac)
For iOS: open http://<your-ip>:8080/cert.pem in Safari → tap Allow → Settings → General → VPN & Device Management → install profile → Certificate Trust Settings → enable full trust.
The framework is split into two layers:
SKILL.md ← GM core: pacing, NPCs, improvisation, world craft
Never changes. Works for any game.
systems/<your-system>/
system.md ← Your game's rules: dice, stats, health, resources
Loaded alongside SKILL.md at session start.
SKILL.md contains everything about being a good GM. system.md contains everything about your specific game. The GM model reads both at session start — it brings the craft, your system module brings the rules.
D&D 5e ships as the reference implementation. It demonstrates what a complete system module looks like: dice conventions, ability scores, spell slots, conditions, death saves, SRD lookup, and character scripts.
Building a new system module takes one file to start — a filled-in systems/TEMPLATE.md for your game. You can start with just dice resolution and health, play a session, then iterate. Full porting guide: SYSTEM-PORTING.md.
---
该工具使用 NOASSERTION 协议,商用场景请仔细阅读协议条款,必要时咨询法律意见。
AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。
📄 NOASSERTION — 请查阅原始协议条款了解具体使用限制。
经综合评估,桌面游戏AI 在Agent工作流赛道中表现稳健,质量良好。如果你已有明确的使用需求,可以直接上手体验;如果还在评估阶段,建议对比同类工具后再做决策。
| 原始名称 | open-tabletop-gm |
| 原始描述 | 开源AI工作流:LLM-agnostic tabletop RPG Game Master framework. Runs on OpenCode, LM Studio, or。⭐14 · Python |
| Topics | gamemasterllm-agentlm-studiopython |
| GitHub | https://github.com/Bobby-Gray/open-tabletop-gm |
| License | NOASSERTION |
| 语言 | Python |
收录时间:2026-05-26 · 更新时间:2026-05-30 · License:NOASSERTION · AI Skill Hub 不对第三方内容的准确性作法律背书。
选择 Agent 类型,复制安装指令后粘贴到对应客户端