kiro-cli-openclaw-integration

通过本地 ACP-to-OpenAI Bridge 将 OpenClaw(或任何 OpenAI 兼容客户端)连接到 kiro-cli 的 ACP 后端,支持流式响应和工具调用。

Safety Notice

This listing is from the official public ClawHub registry. Review SKILL.md and referenced scripts before running.

Copy this and send it to your AI assistant to learn

Install skill "kiro-cli-openclaw-integration" with this command: npx skills add luoshixi/kiro-cli-openclaw-bridge

Kiro CLI 与 OpenClaw 集成

Local proxy that translates OpenAI/Anthropic API requests into ACP JSON-RPC calls, enabling OpenClaw and other OpenAI-compatible clients to use kiro-cli as the AI backend.

Architecture

OpenClaw / Any Client  ──HTTP──▶  Bridge (FastAPI)  ──stdio──▶  kiro-cli acp
                                  :18788/v1                     JSON-RPC 2.0

从零开始完整搭建指南

Step 1: 安装 kiro-cli

# 下载安装 kiro-cli(如已安装跳过)
# 参考 kiro 官方文档安装,确保 kiro-cli 在 PATH 中
which kiro-cli  # 验证安装

Step 2: 认证 kiro-cli

kiro-cli auth
# 按提示完成登录认证,确保能正常使用
# 验证:
kiro-cli acp --help  # 应显示帮助信息

Step 3: 获取 Bridge

推荐从 GitHub Releases 下载预编译二进制(无需 Python 环境):

# 下载预编译二进制(约 15MB,支持 Linux/WSL 和 macOS)
# https://github.com/LuoShiXi/kiro-cli-openclaw-bridge/releases

# 或从源码构建:
git clone https://github.com/LuoShiXi/kiro-cli-openclaw-bridge.git
cd kiro-cli-openclaw-bridge

Step 4: 安装 Python 环境(源码运行方式)

# 需要 Python 3.10+
python3 --version  # 验证版本

# 创建虚拟环境
python3 -m venv .venv
source .venv/bin/activate

# 安装依赖
pip install -r requirements.txt

依赖清单(requirements.txt):

fastapi==0.115.12
uvicorn==0.34.3
pytest==8.3.5
pytest-asyncio==0.25.3
pyinstaller==6.12.0

Step 5: 启动 Bridge

# 方式 A:源码运行
source .venv/bin/activate
python -m acp_openai_bridge.main --cwd /your/project

# 方式 B:使用预编译二进制(如已构建)
./dist/acp-bridge --cwd /your/project

# 方式 C:构建后运行
./build.sh
./dist/acp-bridge --cwd /your/project

指定模型(可选):

python -m acp_openai_bridge.main --cwd /your/project --model claude-sonnet-4-20250514

启动成功标志:

ACP-to-OpenAI Bridge running at http://127.0.0.1:18788

Step 6: 验证 Bridge 运行

# 健康检查
curl http://127.0.0.1:18788/health
# 应返回:{"status":"ok","acp_available":true}

# 模型列表
curl http://127.0.0.1:18788/v1/models
# 应返回包含 kiro-acp 的模型列表

# 快速测试对话
curl -X POST http://127.0.0.1:18788/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"kiro-acp","messages":[{"role":"user","content":"hello"}],"stream":false}'

Step 7: 配置 OpenClaw

编辑 ~/.openclaw/openclaw.json,在 models.providers 中添加:

{
  "models": {
    "mode": "merge",
    "providers": {
      "kiro-b": {
        "api": "openai-completions",
        "baseUrl": "http://127.0.0.1:18788/v1",
        "apiKey": "any-value",
        "models": [
          {
            "id": "kiro-acp",
            "name": "Kiro ACP",
            "input": ["text"],
            "contextWindow": 200000,
            "maxTokens": 65536
          }
        ]
      }
    }
  }
}

设置为默认模型(可选),在 agents.defaults 中添加:

{
  "agents": {
    "defaults": {
      "model": {
        "primary": "kiro-b/kiro-acp"
      },
      "models": {
        "kiro-b/kiro-acp": {
          "alias": "Kiro"
        }
      }
    }
  }
}

重启 OpenClaw 即可使用。

保存后在 OpenClaw 中选择该 Provider 和 kiro-acp 模型即可开始对话。


命令行参数

参数环境变量默认值说明
--hostACP_BRIDGE_HOST127.0.0.1监听地址
--portACP_BRIDGE_PORT18788监听端口
--kiro-cli-pathACP_BRIDGE_KIRO_CLI_PATH自动查找kiro-cli 路径
--cwdACP_BRIDGE_CWD当前目录ACP 会话工作目录
--timeoutACP_BRIDGE_TIMEOUT300请求超时(秒)
--modelACP_BRIDGE_MODELkiro-cli 默认模型 ID

API 端点

MethodPathDescription
POST/v1/chat/completionsOpenAI 聊天补全(支持 stream: true
POST/v1/messagesAnthropic Messages API(支持 stream: true
GET/v1/models模型列表
GET/health健康检查

流式响应

stream: true 时:

  1. Bridge 发送 session/prompt 到 ACP 子进程
  2. 持续读取 session/update 中的 agent_message_chunk
  3. 实时转换为 OpenAI SSE chat.completion.chunk 格式
  4. 完成后发送 finish_reason[DONE]

工具执行

Bridge 透传 kiro-cli 的内置能力,所有操作受限于 --cwd 指定的项目目录。建议仅在信任的项目目录中使用,并保持服务绑定在 localhost。

故障排查

问题解决方案
Bridge 启动但无响应确认 kiro-cli auth 已完成;检查 --cwd 指向有效目录
Agent 卡在 tool_callstdout 缓冲区已设为 10MB;检查日志是否有 readline buffer overflow
OpenClaw 连接被拒curl http://127.0.0.1:18788/health 验证 bridge 运行中
超时错误增大 --timeout 600;复杂任务需要更多时间
kiro-cli 找不到--kiro-cli-path /path/to/kiro-cli 显式指定

平台支持

平台说明
WSL (Linux)主要开发平台,自动检测 WSL
macOS ARM (M1/M2)支持,自动查找 Homebrew 安装的 kiro-cli
macOS Intel支持,查找 /usr/local/bin/kiro-cli

构建打包

./build.sh        # 构建单文件可执行程序 → dist/acp-bridge (~15MB)
./build.sh clean  # 清理构建产物

构建后的二进制无需 Python 环境即可运行。

项目地址

Source Transparency

This detail page is rendered from real SKILL.md content. Trust labels are metadata-based hints, not a safety guarantee.

Related Skills

Related by shared tags or category signals.

Coding

fboc

CLI tool to manage Facebook Pages: list, create, read, hide, delete posts; list, create, delete comments; and get page info via the Graph API.

Registry SourceRecently Updated
Coding

Claw4Claw Skills

Guide for using the Claw4Claw(虾连虾) CLI tool to interact with the AI Agent collaboration platform. Invoke when users need help with agent registration, task m...

Registry SourceRecently Updated
Coding

Black Diamond

Leading global brand specializing in premium climbing, backcountry skiing, and avalanche safety gear with heritage-based innovation and professional endorsem...

Registry SourceRecently Updated
Coding

Arista Networks

High-performance data center networking company known for its cloud-scale switches and programmable EOS operating system.

Registry SourceRecently Updated