clawgo
agent
Basarisiz
Health Gecti
- License — License: NOASSERTION
- Description — Repository has a description
- Active repo — Last push 0 days ago
- Community trust — 27 GitHub stars
Code Basarisiz
- rm -rf — Recursive force deletion command in .github/workflows/release.yml
- rm -rf — Recursive force deletion command in install.sh
- process.env — Environment variable access in workspace/skills/context7/query.ts
- network request — Outbound network request in workspace/skills/context7/query.ts
Permissions Gecti
- Permissions — No dangerous permissions requested
Purpose
ClawGo is a Go-based AI agent runtime that orchestrates multiple AI agents, manages memory/audit logs, and provides a WebUI dashboard. It is designed for long-running, recoverable agent workflows.
Security Assessment
Overall risk: Medium. The tool executes shell commands via its agent runtime and makes outbound network requests to LLM providers. Two files contain `rm -rf` commands: the install script and a GitHub Actions release workflow — both are standard for cleanup operations but warrant visual inspection. A TypeScript skill file accesses environment variables and makes network requests. No hardcoded secrets were detected and no dangerous system permissions are requested. The install command uses `curl | bash`, which is typical but inherently requires trusting the remote script.
Quality Assessment
The project is actively maintained (last push was today). It has 27 GitHub stars, indicating early-stage community adoption. The license is marked as NOASSERTION, meaning no clear open-source license is defined — this is a legal concern for production or derivative use. The repository includes a description and structured documentation in both Chinese and English.
Verdict
Use with caution — functional and actively maintained, but the lack of a declared license and the presence of `rm -rf` in install/release scripts mean you should review those files before deploying in production.
ClawGo is a Go-based AI agent runtime that orchestrates multiple AI agents, manages memory/audit logs, and provides a WebUI dashboard. It is designed for long-running, recoverable agent workflows.
Security Assessment
Overall risk: Medium. The tool executes shell commands via its agent runtime and makes outbound network requests to LLM providers. Two files contain `rm -rf` commands: the install script and a GitHub Actions release workflow — both are standard for cleanup operations but warrant visual inspection. A TypeScript skill file accesses environment variables and makes network requests. No hardcoded secrets were detected and no dangerous system permissions are requested. The install command uses `curl | bash`, which is typical but inherently requires trusting the remote script.
Quality Assessment
The project is actively maintained (last push was today). It has 27 GitHub stars, indicating early-stage community adoption. The license is marked as NOASSERTION, meaning no clear open-source license is defined — this is a legal concern for production or derivative use. The repository includes a description and structured documentation in both Chinese and English.
Verdict
Use with caution — functional and actively maintained, but the lack of a declared license and the presence of `rm -rf` in install/release scripts mean you should review those files before deploying in production.
Production-ready personal AI agent runtime in Go with WebUI, multi-agent orchestration, memory, audit, and distributed node support.
README.md
ClawGo
面向生产的 Go 原生 Agent Runtime。
ClawGo 不是“又一个聊天壳子”,而是一套可长期运行、可观测、可恢复、可编排的 Agent 运行时。
main agent负责入口、路由、派发、汇总subagent runtime负责本地分支执行runtime store持久化 run、event、thread、message、memory- WebUI 负责检查、状态展示和账号管理,不负责运行时配置写入
为什么是 ClawGo
大多数 Agent 项目停留在:
- 一个聊天界面
- 一组工具调用
- 一段 prompt
ClawGo 更关注真正的运行时能力:
- 多 Agent 拓扑和内部协作流
- 可恢复的 subagent run
- 本地主控加本地 subagent 分支
- 配置、审计、日志、记忆的工程化闭环
一句话:
ClawGo 是 Agent Runtime,而不只是 Agent Chat。
核心能力
1. 多 Agent 拓扑
- 统一展示
main / subagents - 内部协作流与用户对话分离
- 子代理执行过程可追踪,但不会污染用户主通道
2. 可恢复执行
subagent_runs.jsonlsubagent_events.jsonlthreads.jsonlagent_messages.jsonl- 重启后可恢复进行中的 subagent run
3. 工程化配置
config.json管理 agent registrysystem_prompt_file -> AGENT.md- WebUI 可查看状态、节点、账号与运行信息
- 运行时配置修改回到文件,不通过 WebUI 写入
4. 长期运行能力
- 本地优先
- Go 原生 runtime
- 多通道接入
- Task Audit / Logs / Memory / Skills / Config / Agents 全链路闭环
WebUI
WebUI 当前定位:
- Dashboard 与 Agent 拓扑查看
- 节点、日志、记忆、运行态检查
- OAuth 账号管理
WebUI 当前不负责:
- 修改 subagent/runtime 配置
- 查询或控制公开 task runtime
快速开始
1. 安装
curl -fsSL https://raw.githubusercontent.com/YspCoder/clawgo/main/install.sh | bash
2. 初始化
clawgo onboard
3. 选择服务商与模型
clawgo provider list
clawgo provider use openai/gpt-5.4
clawgo provider configure
如果服务商使用 OAuth,例如 codex、anthropic、gemini、kimi、qwen:
clawgo provider login codex
clawgo provider login codex --manual
如果同一个 provider 同时有 API key 和 OAuth 账号,推荐配置 auth: "hybrid":
- 优先使用
api_key - 额度或限流失败时自动切到 OAuth 账号池
- 仍保留多账号轮换和后台刷新
4. 启动
交互模式:
clawgo agent
clawgo agent -m "Hello"
网关模式:
clawgo gateway run
开发模式:
make dev
WebUI:
http://<host>:<port>/?token=<gateway.token>
架构概览
默认协作模式:
user -> main -> worker -> main -> user
当前系统包含四层:
main agent
负责用户入口、路由、派发、汇总local subagents
在config.json -> agents.subagents中声明,使用独立 session 和 memory namespaceruntime store
保存运行态、线程、消息、事件和审计数据
说明:
subagent_profile保留,用来创建和管理 subagent 定义spawn保留,用来触发 subagent 执行- 公开 task runtime、WebUI 配置写入、EKG、watchdog 已移除
配置结构
当前有两层配置视图:
- 落盘文件继续使用原始结构
- 只读接口可暴露标准化视图:
coreruntime
推荐结构:
{
"agents": {
"defaults": {
"context_compaction": {},
"execution": {},
"summary_policy": {}
},
"router": {
"enabled": true,
"main_agent_id": "main",
"strategy": "rules_first",
"policy": {
"intent_max_input_chars": 1200,
"max_rounds_without_user": 200
},
"rules": []
},
"communication": {},
"subagents": {
"main": {},
"coder": {},
"tester": {}
}
}
}
说明:
runtime_control已移除- WebUI 配置编辑已禁用
- 运行时配置修改请直接改
config.json - 启用中的本地 subagent 必须配置
system_prompt_file
完整示例见 config.example.json。
MCP 服务支持
ClawGo 支持通过 tools.mcp 接入 stdio、http、streamable_http、sse 型 MCP server。
- 在
config.json -> tools.mcp.servers中声明 server - 启动时自动发现远端 MCP tools,并注册为本地工具
- 工具命名格式为
mcp__<server>__<tool> permission=workspace时,working_dir必须留在 workspace 内permission=full时,working_dir可以是任意绝对路径
Prompt 文件约定
推荐把 agent prompt 独立成文件:
agents/main/AGENT.mdagents/coder/AGENT.mdagents/tester/AGENT.md
配置示例:
{
"system_prompt_file": "agents/coder/AGENT.md"
}
规则:
- 路径必须是 workspace 内相对路径
- 仓库不内置这些示例 prompt
- 用户或 agent workflow 需要自行创建实际文件
记忆与运行态
ClawGo 不是所有 agent 共用一份上下文。
main- 保存主记忆与协作摘要
subagent- 使用独立 session key
- 写入自己的 memory namespace
runtime store- 持久化 runs、events、threads、messages
- 内部仍可使用 task 建模,但不再暴露成公开控制面
这带来三件事:
- 更好恢复
- 更好追踪
- 更清晰的执行边界
适合谁
- 想用 Go 做 Agent Runtime 的开发者
- 想要可视化多 Agent 拓扑和内部协作流的团队
- 需要强观测、强审计、强恢复能力的系统
- 想把 agent 配置、prompt、工具权限工程化管理的团队
如果你想快速上手,先看 config.example.json,再跑一次 make dev。
Yorumlar (0)
Yorum birakmak icin giris yap.
Yorum birakSonuc bulunamadi