Skip to content

Self-hosted workflows for coding-agent delivery. Orchestrate CodeBuddy, Codex and Claude Code with scripts and human approval gates. Includes Local workflows for coding assistants.

docs(readme): list all ten executors and their credential rules (#496)

cocofhud53c576c93ad
diff --git a/README.md b/README.md
index 92322dbf..3ab5601e 100644
--- a/README.md
+++ b/README.md
@@ -37,7 +37,7 @@ https://github.com/user-attachments/assets/8af401d9-1aae-427c-b4bc-455c12d17beb
**LoopForge** is a self-hosted workspace where your team ships software with coding agents. Describe what you need to **Main Bot** in chat: it opens a Docker sandbox, checks out your repositories, delegates to project roles, and opens the PR. When a way of working proves itself, save it as an **SOP** (a standard workflow of agents, scripts and human gates) so the next requirement is delivered the same way, with review and approval built in.
-CodeBuddy, Codex and Claude Code do the work. LoopForge adds the roles, sandboxes, checkpoints and history around them, all on your own infrastructure under the MIT license.
+Coding-agent CLIs do the work: CodeBuddy, Codex, Claude Code, OpenCode, Cursor CLI, Antigravity, Kimi Code, Pi, DeepSeek Harness and ZCode. LoopForge adds the roles, sandboxes, checkpoints and history around them, all on your own infrastructure under the MIT license.
## Two ways to work
@@ -67,7 +67,7 @@ Ask a coding agent to "fix this bug" and the result depends on that run. It migh
- **Separated roles.** Planning, implementation, review and testing each get their own prompt, executor, model, rules, Skills and MCP tools. A role works as an SOP node or as a subagent in chat.
- **Human in the loop.** Approval gates where they matter. Read the report, approve, or send the work back to a chosen step.
- **Observable and resumable.** Node status, tool traces, logs and artifacts in one place. Pause, retry a failed step, or rerun from any node.
-- **Your stack, your servers.** CodeBuddy, Codex or Claude Code as executors. GitHub, Skills and MCP as tools. Docker sandboxes or your own machine as the runtime.
+- **Your stack, your servers.** Ten coding-agent CLIs as executors, from CodeBuddy, Codex and Claude Code to OpenCode, Cursor CLI and more. GitHub, Skills and MCP as tools. Docker sandboxes or your own machine as the runtime.
## Quick start
@@ -112,7 +112,19 @@ New accounts join the **LoopForge demo project**, which includes an SOP made onl
<br>
-- **Executors.** Pick **CodeBuddy, Codex or Claude Code** beside the home composer, then a model. Each executor uses its own credentials from personal settings, and every SOP role can choose its executor independently (built-in roles default to CodeBuddy). Requirement clarification and AI notification summaries currently use CodeBuddy; scheduled assistant tasks run on the CLI the task selects. The Docker image ships CodeBuddy and Codex; Claude Code installs from npm on first use and is cached, so that first run requires npm access; [native deployments](docs/deployment.md#workers-and-runners) need the CLI for the executor you choose.
+- **Executors.** Pick one of ten CLIs beside the home composer, then a model. Every SOP role can choose its executor independently (built-in roles default to CodeBuddy). Requirement clarification and AI notification summaries currently use CodeBuddy; scheduled assistant tasks run on the CLI the task selects.
+
+ | Executor | Credentials | Installation |
+ | --- | --- | --- |
+ | **CodeBuddy** | Personal key, or a shared key an administrator configures for chat and SOP runs | In the Docker image |
+ | **Codex** | Personal key; a connected execution machine can use its local login | In the Docker image |
+ | **Claude Code** | Personal key; a connected execution machine can use its local login | Installed from npm on first use, then cached |
+ | **OpenCode · Kimi Code · Pi** | Personal key only | Installed from npm on first use, then cached |
+ | **DeepSeek Harness** | Personal key only | In the Docker image |
+ | **Cursor CLI · Antigravity** | Personal key only | Provider installer on first chat use; not installed into SOP sandboxes |
+ | **ZCode** | Personal key only | Not bundled; install it on the service or execution machine PATH |
+
+ The seven agent CLIs after Claude Code keep their model and MCP configuration in a private home directory per run, so they always use the responsible user's own key and Base URL from personal settings, even on a connected execution machine. When rolling them out to a team, each member configures their own credentials. First-use installs need npm or provider access; [native deployments](docs/deployment.md#workers-and-runners) need the CLI for the executor you choose.
- **GitHub is optional.** All built-in generic SOPs run without a repository: product SOPs deliver documents, and new frontend/backend projects deliver source files in the task workspace. Bug fixes still need source code and reproduction details. Connect a repository when you work on existing code or want PRs.
- **Delivering to a repository.** Choose **Connect GitHub** in personal settings, approve the device code on GitHub and pick your repository (the LoopForge OAuth App is built in; no PAT required, manual PAT entry still works). Set executor credentials (or bind a local execution machine), then either run a project SOP or ask Main Bot to change the code and submit a PR/MR.
- **Sandboxes.** In the default Compose installation, SOP runs and conversation workspaces both use Docker sandbox containers started through the host Docker socket. Default sandboxes reuse the installed runtime image, so nothing extra is built after installation; custom Dockerfiles still build their own images. Docker socket access grants control of the host daemon, so this setup is for trusted teams, not untrusted multi-tenant isolation. See [deployment boundaries](docs/deployment.md#accounts). Re-running the installer keeps your data.
@@ -219,7 +231,7 @@ Use chat to explore and iterate; Main Bot can still open a PR. Once you find a w
**How is this different from Dify, n8n or Flowise?**
Those are general-purpose AI and automation workflow builders. LoopForge is built for **software delivery with coding agents**: implementer and reviewer roles, Docker sandboxes with real repositories, GitHub delivery, human acceptance gates and run evidence.
-**How is this different from using Claude Code, Codex or CodeBuddy directly?**
+**How is this different from using Claude Code, Codex, CodeBuddy or another coding CLI directly?**
LoopForge doesn't replace them; it orchestrates them. Each agent step and each chat turn runs one of those CLIs. LoopForge adds what surrounds it: who does what, in which sandbox, in which order, with which checks, and who signs off.
**Is it safe to expose on the internet?**
diff --git a/README.zh-CN.md b/README.zh-CN.md
index ecb2d9ca..b0684be2 100644
--- a/README.zh-CN.md
+++ b/README.zh-CN.md
@@ -37,7 +37,7 @@ https://github.com/user-attachments/assets/10b8ae2e-d7e7-4b78-8876-c45818d4a08f
**LoopForge** 是一个可自主部署的团队工作台,让你和编程 Agent 一起交付软件。在对话里把需求告诉 **Main Bot**:它会按需打开 Docker 沙箱、检出项目仓库、把任务分派给项目角色,最后提交 PR。某种做法被验证有效后,把它存成 **SOP**(由 Agent、脚本和人工关口组成的标准交付流程),下一个需求就按同样的方式交付,评审和验收都在流程里。
-干活的是 CodeBuddy、Codex 和 Claude Code;LoopForge 负责它们外面的角色、沙箱、检查点和留痕,全部运行在你自己的环境里,MIT 开源。
+干活的是编程 Agent CLI:CodeBuddy、Codex、Claude Code、OpenCode、Cursor CLI、Antigravity、Kimi Code、Pi、DeepSeek Harness 和 ZCode;LoopForge 负责它们外面的角色、沙箱、检查点和留痕,全部运行在你自己的环境里,MIT 开源。
## 两种工作方式
@@ -67,7 +67,7 @@ https://github.com/user-attachments/assets/10b8ae2e-d7e7-4b78-8876-c45818d4a08f
- **角色分离。** 方案、实现、评审、测试各有独立的角色提示词、执行器、模型、规则、Skills 和 MCP 工具;同一个角色既能做 SOP 节点,也能在对话中作为子 Agent 调用。
- **人在回路。** 在关键节点设置人工审批:看完报告再放行,或打回到指定环节返工。
- **可观测、可续跑。** 节点状态、工具轨迹、日志和产物集中呈现;支持暂停、重试失败步骤、从任意节点重跑。
-- **技术栈和服务器都是你的。** 执行器可选 CodeBuddy、Codex、Claude Code;工具接入 GitHub、Skills、MCP;运行环境可选 Docker 沙箱或你自己的机器。
+- **技术栈和服务器都是你的。** 执行器支持 10 种编程 Agent CLI,从 CodeBuddy、Codex、Claude Code 到 OpenCode、Cursor CLI 等;工具接入 GitHub、Skills、MCP;运行环境可选 Docker 沙箱或你自己的机器。
## 快速开始
@@ -112,7 +112,19 @@ GitLab、Gitea、Gogs 等托管可[配置 Git SSH](docs/deployment.md#git-reposi
<br>
-- **执行器。** 在首页输入框旁选择 **CodeBuddy、Codex 或 Claude Code**,再选择模型;不同执行器分别使用个人设置里的对应凭据。每个 SOP 角色都可以独立选择执行器(内置角色默认 CodeBuddy)。需求澄清和通知 AI 汇总目前使用 CodeBuddy;定时助手任务使用任务所选的执行器。Docker 镜像内置 CodeBuddy 和 Codex;Claude Code 在首次使用时从 npm 安装并缓存,需要能访问 npm;[原生部署](docs/deployment.md#workers-and-runners)需安装所选执行器的 CLI。
+- **执行器。** 在首页输入框旁从 10 种 CLI 中选择一种,再选择模型。每个 SOP 角色都可以独立选择执行器(内置角色默认 CodeBuddy)。需求澄清和通知 AI 汇总目前使用 CodeBuddy;定时助手任务使用任务所选的执行器。
+
+ | 执行器 | 凭据 | 安装方式 |
+ | --- | --- | --- |
+ | **CodeBuddy** | 个人 Key,或管理员配置的平台共享 Key(覆盖对话和 SOP 运行) | Docker 镜像内置 |
+ | **Codex** | 个人 Key;已连接的执行机器可使用本机登录 | Docker 镜像内置 |
+ | **Claude Code** | 个人 Key;已连接的执行机器可使用本机登录 | 首次使用时从 npm 安装并缓存 |
+ | **OpenCode · Kimi Code · Pi** | 仅个人 Key | 首次使用时从 npm 安装并缓存 |
+ | **DeepSeek Harness** | 仅个人 Key | Docker 镜像内置 |
+ | **Cursor CLI · Antigravity** | 仅个人 Key | 首次在对话中使用时通过厂商安装脚本安装;不会装进 SOP 沙箱 |
+ | **ZCode** | 仅个人 Key | 未内置,需安装到服务或执行机器的 PATH 中 |
+
+ Claude Code 之后的 7 种 Agent CLI 每次运行都在私有 HOME 目录里生成模型和 MCP 配置,因此始终使用需求责任人在个人设置中填写的 Key 和 Base URL,即使在已连接的执行机器上也一样。团队推广时,每位成员需要各自配置凭据。首次安装需要能访问 npm 或对应厂商;[原生部署](docs/deployment.md#workers-and-runners)需安装所选执行器的 CLI。
- **GitHub 可选。** 所有内置通用 SOP 都可以不绑定仓库运行:产品方案交付文档,前后端新项目在任务工作区交付源码;缺陷修复仍需提供待修代码和复现材料。需要修改已有代码或提交 PR 时再绑定仓库。
- **交付到仓库。** 在个人设置中点击「连接 GitHub」,用设备码授权后选择自己的仓库(已内置 LoopForge 授权应用,无需申请 PAT,仍可手动填写 PAT),再配置执行器凭据(或绑定本机执行机器),之后既可以运行一条项目 SOP,也可以直接让 Main Bot 修改代码并提交 PR/MR。
- **沙箱。** 默认 Compose 安装中,SOP 运行和对话工作区都使用 Docker 沙箱容器,由平台通过宿主机 Docker socket 启动;默认沙箱复用已安装的运行镜像,安装后无需再构建;自定义 Dockerfile 仍单独构建。Docker socket 赋予平台控制宿主机 Docker 的权限,因此只适用于可信团队,不构成不可信多租户隔离边界;详见[部署边界](docs/deployment.md#accounts)。重复执行安装脚本会保留已有数据。
@@ -217,7 +229,7 @@ npx -y loopforge-cli@latest install claude
**和 Dify、n8n、Flowise 有什么区别?**
它们是通用的 AI / 自动化工作流搭建平台。LoopForge 专注于 **用编程 Agent 做软件交付**:实现者与评审者等角色分工、带真实仓库的 Docker 沙箱、GitHub 交付、人工验收关口和运行证据。
-**和直接用 Claude Code / Codex / CodeBuddy 有什么区别?**
+**和直接用 Claude Code / Codex / CodeBuddy 等编程 CLI 有什么区别?**
LoopForge 不替代它们,而是编排它们。每个 Agent 步骤和每轮对话都运行这些 CLI 之一;LoopForge 补上外面的部分:谁来做、在哪个沙箱里、按什么顺序、经过哪些检查、由谁签字。
**可以直接暴露到公网吗?**