DevMuse

A complete software development workflow for Claude Code — built on a four-layer architecture of rules, skills, agents, and knowledge.

View on GitHub → Install

How it works

It starts the moment you fire up Claude Code. When it sees you're building something, it doesn't jump into writing code — instead it steps back and scopes the work first: enumerating use cases, detecting conflicts, and assessing impact on existing code. Only then does it move to design.

Once it has teased a spec out of the conversation, it shows it to you in chunks short enough to actually read and digest. After you've signed off on the design, your agent puts together an implementation plan clear enough for an enthusiastic junior engineer with poor taste, no judgment, and an aversion to testing to follow. It emphasizes true red/green TDD, YAGNI, and DRY.

Then once you say "go", it launches a subagent-driven development process — agents work through each task, inspect and review their work, and continue. It's not uncommon for Claude to work autonomously for hours without deviating from the plan.

The pipeline

scope → arch → plan → code → review

Each stage is a skill that auto-invokes based on intent. Orthogonal skills (mu-explore, mu-debug, mu-retro) handle code comprehension, root cause analysis, and retrospectives. On-demand skills (mu-biz, mu-prd, mu-wiki) cover business analysis, product requirements, and architecture documentation — invoked explicitly with /slash commands.

Install

Inside Claude Code:

# Register the marketplace
/plugin marketplace add knotmark-ai/devmuse

# Install the plugin
/plugin install devmuse@devmuse

Then start a new session and ask for something that should trigger a skill — e.g. "help me plan this feature" or "let's debug this issue". The agent should automatically invoke the relevant skill.

What's inside

13 skills — full pipeline + orthogonal + on-demand workflows
2 agents — mu-coder (implementation) and mu-reviewer (six-mode review)
Bootstrap rule — always-on principles loaded at session start
Knowledge base — domain knowledge injected on demand (inversion test, NFR checklist, sign-off gate, …)

Credits

DevMuse is derived from Superpowers by Jesse Vincent. Released under the MIT license.

DevMuse

专为 Claude Code 设计的完整软件开发工作流，基于规则、技能、代理、知识四层架构构建。

GitHub 仓库 → 安装

工作原理

从你启动 Claude Code 的那一刻就开始了。当它发现你正在构建某个东西时，它不会直接跳进去写代码，而是先界定范围—— 枚举用例、检测冲突、评估对现有代码的影响。只有在这之后，才进入设计阶段。

在通过对话梳理出需求规格后，它会将内容分成足够短的片段展示给你，便于你实际阅读和消化。在你确认设计方案后，Agent 会制定一份实施计划—— 清晰到即使是一个热情但品味欠佳、缺乏判断力、不爱写测试的初级工程师也能遵循。计划强调真正的红/绿 TDD、YAGNI 和 DRY 原则。

接下来，当你说"开始"后，它会启动一个子 Agent 驱动开发流程，让多个 Agent 逐一完成每个任务、互相检查和审查，然后继续推进。Claude 连续自主工作几个小时而不偏离计划是很常见的。

管线

scope → arch → plan → code → review

每个阶段都是一个根据意图自动调用的技能。正交技能（mu-explore、mu-debug、mu-retro）负责代码理解、根因分析和回顾。按需技能（mu-biz、mu-prd、mu-wiki）覆盖商业分析、产品需求、架构文档——需要通过 /slash 命令显式调用。

安装

在 Claude Code 中：

# 注册市场
/plugin marketplace add knotmark-ai/devmuse

# 安装插件
/plugin install devmuse@devmuse

然后启动新会话，发起一些应该触发技能的请求——比如 "帮我规划这个功能" 或 "让我们调试这个问题"。 Agent 会自动调用相关技能。

包含哪些

13 个技能 — 完整管线 + 正交 + 按需工作流
2 个代理 — mu-coder（实现）和 mu-reviewer（六模式审查）
Bootstrap 规则 — 会话启动时始终加载的原则
知识库 — 按需注入的领域知识（反向测试、NFR 清单、签字门禁……）

致谢

DevMuse 基于 Jesse Vincent 的 Superpowers 派生。以 MIT 协议发布。