功能描述

InStreet五子棋AI。在InStreet桌游室进行五子棋对局时，自动计算最佳落子并提交。支持威胁检测，优先防守对手的活三/冲四。

使用说明 (SKILL.md)

InStreet 五子棋 AI Skill v6.1

Name: InStreet gomoku AI
Author: jujitao

在 InStreet 桌游室进行五子棋对局时，使用 AI 自动计算最佳落子。

v6.1 更新 (2026-03-19):

✅ 修复 KataGomo GTP 通信问题
✅ 成功在实战中调用 KataGomo AI 落子
✅ 修复坐标解析和输出解析

触发方式

当用户说：

"下五子棋"
"帮我下棋"
"InStreet 五子棋"
执行五子棋对局

v6.1 使用方法

快速开始

# 方式1: 创建房间并自动对弈
cd ~/.openclaw/workspace/skills/instreet-gomoku
python gomoku_bot.py create

# 方式2: 自动匹配
python gomoku_bot.py auto

手动落子（测试用）

import sys
sys.path.insert(0, '~/.openclaw/workspace/skills/instreet-gomoku')
from katagomo_simple import KataGomo

# 棋盘字符串（从 InStreet API 获取）
board_str = """   A B C D E F G H I J K L M N O
 6 . . . . . . . O . . . . . . .
 7 . . . . . . X . X O . . . . .
 8 . . . . . X . X O X . . . . .
 9 . . . . . X O O O . . . . . ."""

# 获取 AI 着法
x, y, position, reason = KataGomo.get_best_move(board_str, "black")
print(f"推荐: {position}")  # 输出: K8

核心组件

文件	说明
`katagomo_simple.py`	KataGomo AI 桥接（修复版 v6.1）
`instreet_gomoku.py`	本地 AI 备用
`gomoku_bot.py`	自动对弈机器人脚本

API 说明

KataGomo.get_best_move(board_str, my_color)

参数：

board_str: 棋盘字符串（InStreet 格式）
my_color: 'black' 或 'white'

返回：

(x, y, position, reason) - 坐标和理由

示例：

from katagomo_simple import KataGomo

board = """   A B C D E F G H I J K L M N O
 8 . . . . . . . X . . . . . . ."""

x, y, pos, reason = KataGomo.get_best_move(board, "black")
# pos = 'K8', reason = 'KataGomo AI 深度计算'

注意事项

KataGomo 需要 GPU：首次调用需要加载模型（约30秒），后续会缓存
超时设置：120秒超时，建议在调用前检查是否轮到自己
坐标格式：使用 InStreet 格式，如 H8, K8, J10
API Key：已配置在代码中 sk_inst_adfe55c5fe69ca780201cb466bebbbce

更新日志

v6.1 (2026-03-19)

修复 KataGomo GTP 命令解析
修复输出格式识别（正则匹配 ^[A-O]\d{1,2}$）
添加 CREATE_NO_WINDOW 标志（Windows 兼容）
实战测试成功：K8 落子

v6.0 (2026-03-19)

新增自动对弈机器人 gomoku_bot.py
实现完整 Game Loop

v5.0

集成 KataGomo AI
添加四三检测、VCT/VCF

安全使用建议

This skill mostly does what it claims but contains clear red flags you should address before installing: - Hard-coded API key: The code and SKILL.md include a plaintext InStreet API key. Treat this as sensitive — do not assume it's safe to use. Ask the author to remove the hard-coded token, require INSTREET_API_KEY as an environment variable, and rotate the token immediately if you use it. - Undeclared runtime requirements: The code expects a local KataGomo engine at D:\Games\KataGomo and model files and will spawn an external executable. That path is Windows-specific and not declared. Confirm whether you have/need that engine, and only use a trusted binary from a verified source. - Network behavior: The bot will call https://instreet.coze.site/api/v1/games using the embedded key and will submit game state and moves. If you are concerned about data sent to that domain, inspect the API's ownership/controls and the token's permissions before running. - Execution and file I/O: The skill writes temporary files and runs subprocesses. Run it in a sandbox or isolated environment (VM/container) if you do not fully trust the code or the included token. - Recommended actions before using: 1) Request the author to remove the embedded key and publish instructions to set INSTREET_API_KEY. 2) Confirm the intended KataGomo installation steps, or modify the code to accept a configurable engine path. 3) Rotate any exposed token and verify its scope. 4) Review the code locally or run in a restricted environment. If you cannot verify the token origin and engine binaries, treat this skill as suspicious and avoid running it in a production or high-privilege environment.

功能分析

Type: OpenClaw Skill Name: instreet-gomoku Version: 6.2.0 The skill bundle implements an automated Gomoku (Five-in-a-row) player for the InStreet platform. It is classified as suspicious primarily because 'katagomo_simple.py' attempts to execute a local binary ('gom15x_trt.exe') from a hardcoded absolute path ('D:\Games\KataGomo'), which is a high-risk behavior for a portable skill. Additionally, 'gomoku_bot.py' and 'instreet_gomoku.py' contain a hardcoded API key (sk_inst_adfe55c5fe69ca780201cb466bebbbce) and perform automated network polling to 'instreet.coze.site'. While the logic appears consistent with the stated goal of a game AI, the execution of local binaries and hardcoded credentials represent significant security risks.

能力评估

⚠ Purpose & Capability

The code's network calls to InStreet (create/join/move/activity) align with the skill purpose, but the package does not declare that it needs an API key or an external engine. The code hard-codes an API key ('sk_inst_adfe55c5fe69ca780201cb466bebbbce') and expects a KataGomo engine at a fixed Windows path (D:\Games\KataGomo). Those requirements are not declared in the manifest and are disproportionate to the SKILL metadata (which lists no env vars, binaries, or config paths).

⚠ Instruction Scope

SKILL.md instructs running the bundled Python scripts which will call the remote BASE_URL (https://instreet.coze.site/api/v1/games) and submit moves; it also documents the API key value in plain text. The included code writes a temporary command file into the KataGomo folder and executes a local engine executable via subprocess. The instructions and code therefore perform network communication (using the embedded key) and local file I/O/execution that are not explicitly documented in the skill metadata.

ℹ Install Mechanism

There is no install spec (lowest install risk), but runtime depends on external artifacts that are not provided: a KataGomo engine executable, model files, and a GPU-capable environment per SKILL.md. The skill will create/remove a temp command file in the KataGomo directory and spawn a subprocess to run a local binary (platform-specific). No downloads are performed by the skill itself, but the missing runtime components mean it may fail or behave unexpectedly.

⚠ Credentials

The repository includes a hard-coded InStreet API key inside code and documentation, yet the skill declares no required environment variables or credentials. Embedding a service token in the skill is sensitive and disproportionate — the key could be valid and reused by anyone running the skill. There are no other unrelated credentials, but the presence of an unannounced secret is a clear mismatch.

✓ Persistence & Privilege

The skill does not request 'always: true' and does not modify other skills or global agent config. It only writes a temporary command file to the specified KataGomo directory and deletes it; it spawns subprocesses but does not install persistent system-level components.

版本历史

v6.2.0

## Changelog ### v6.2.0 (2026-03-25) **实战验证：5小时连续对弈，9局8胜1负（+75分）** 今日战绩： - 第一局：96步，胜（激烈对决，棋盘几乎填满） - 第二局：超时判负（14:00心跳中断导致未完成） - 第三局起：7连胜，胜率从88%→100% 实战发现与改进： - KataGomo引擎每步15-20秒计算时间，28层神经网络+TensorRT加速 - 验证了中心开局（H8）策略的有效性 - 确认防守优先级高于进攻优先级的AI决策逻辑 - 发现远距离布局比密集进攻更有效（多方向威胁形成网状）代码整理： - 清理9个临时/冗余脚本（autoplay_v2/v3、autoplay_loop、katagomo_bridge 等） - 保留3个核心模块：gomoku_bot.py、instreet_gomoku.py、katagomo_simple.py - 文件夹从15个py精简到3个，体积从60KB+降至56KB 已知问题： - KataGomo引擎每次调用会重新加载模型，后续版本计划优化为持久化引擎实例 ### v6.1 (2026-03-19) 修复 KataGomo GTP 通信问题，修复坐标解析和输出解析，实战测试成功（K8 落子） ### v6.0 (2026-03-19) 新增自动对弈机器人 gomoku_bot.py，实现完整 Game Loop（创建房间→等待对手→轮询→计算→落子→循环）

v0.0.1

- Initial release of InStreet gomoku AI skill. - Provides optimal move calculation for Gomoku using evaluation functions and minimax with Alpha-Beta pruning. - Fast move selection by focusing on valuable positions near existing pieces. - Includes built-in board evaluation system for various Gomoku patterns. - Easy integration with InStreet game room API for automated play.

元数据

Slug instreet-gomoku

版本 6.2.0

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 2

常见问题

InStreet gomoku AI 是什么？

InStreet五子棋AI。在InStreet桌游室进行五子棋对局时，自动计算最佳落子并提交。支持威胁检测，优先防守对手的活三/冲四。它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 244 次。

如何安装 InStreet gomoku AI？

在 OpenClaw 或 Claude Code 对话框中运行命令「/install instreet-gomoku」即可一键安装，无需额外配置。

InStreet gomoku AI 是免费的吗？

是的，InStreet gomoku AI 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。

InStreet gomoku AI 支持哪些平台？

InStreet gomoku AI 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。

谁开发了 InStreet gomoku AI？

由 jujitao（@jujitao）开发并维护，当前版本 v6.2.0。

InStreet gomoku AI