展示HN:Shannon – 本地桌面应用程序,用于协调Claude Code代理团队

1作者: lydionfinance2 个月前原帖
在使用Claude Code CLI处理较大项目时,我遇到了一个问题:我想将工作分配给不同的专门代理(一个负责写代码,另一个负责审查,还有一个负责运行测试),并通过依赖图进行协调。在多个终端会话中管理这一切非常麻烦。 Shannon解决了这个问题: - 创建具有不同模型(Opus/Sonnet/Haiku)和系统提示的定制代理 - 使用拖放式DAG编辑器构建团队工作流程(并行、顺序或完全自定义) - 用自然语言描述你的目标 → AI分析你的代码库并提出带有依赖关系的任务计划 - 实时监控一切:任务图、代理聊天、代码差异 此外,还有一个基于Monaco的提示编辑器,支持XML标签的语义语法高亮(Claude对此反应良好),自动补全,以及一个“一键AI改进”按钮,可以在一次点击中重写你的系统提示。 技术:Go后端,React前端,使用Wails v2作为桌面外壳,SQLite用于存储。它在后台调用Claude Code CLI,而不是直接调用API,因此你可以使用Claude Code的所有内置工具(文件编辑、bash等)。 如你所猜测的,这个项目是以Claude Shannon命名的。 限制:需要安装并认证Claude Code CLI。仅限本地桌面应用。作为一个爱好项目,可能会有一些粗糙之处。工作区副本在大型代码库上可能会占用磁盘空间。 提供Linux和Windows版本。采用MIT许可证。 [项目链接](https://github.com/yessGlory17/shannon)
查看原文
I ran into a problem when using Claude Code CLI on larger projects: I wanted to split work across specialized agents (one writes code, another reviews, another runs tests) and coordinate them with dependency graphs. Managing this across multiple terminal sessions was painful.<p>Shannon solves this:<p>- Create customized agents with different models (Opus&#x2F;Sonnet&#x2F;Haiku) and system prompts - Build team workflows with a drag-and-drop DAG editor (parallel, sequential, or fully custom) - Describe your goal in natural language → AI analyzes your codebase and proposes a task plan with dependencies - Watch everything in real-time: task graph, agent chat, code diffs<p>There&#x27;s also a Monaco-based prompt editor with semantic syntax highlighting for XML tags (the kind Claude responds well to), autocomplete, and an &quot;AI Improve&quot; button that rewrites your system prompt in one click.<p>Tech: Go backend, React frontend, Wails v2 for the desktop shell, SQLite for storage. It shells out to the Claude Code CLI under the hood — not calling the API directly — so you get all of Claude Code&#x27;s built-in tools (file editing, bash, etc.)<p>Named after Claude Shannon, as you might guess.<p>Limitations: requires Claude Code CLI installed and authenticated. Local-only desktop app. Hobby project — expect rough edges. Workspace copies can eat disk space on large repos.<p>Linux and Windows builds available. MIT licensed.<p><a href="https:&#x2F;&#x2F;github.com&#x2F;yessGlory17&#x2F;shannon" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;yessGlory17&#x2F;shannon</a>