lvzhaobo/mumu-coding

Mumu Coding is a Chinese language playbook for rolling out AI coding tools inside a company. The author calls it an AI Coding Playbook with the tagline Any tool, any team, ship specs before code. The repository is positioned as a working set of engineering assets, not a tutorial: methodology documents, templates, checklists, and lab exercises that a team can pick up and apply directly to its own projects. The README diagnoses why AI coding adoption stalls inside companies, even though tools such as Qoder, Tongyi Lingma, Kiro, Trae, Claude Code, Codex, Copilot, and Cursor are already good enough. The failure modes listed are: teams treat the AI as a faster typist instead of aligning on requirements first, a change in one requirement is not reflected in the API, data, tests, and traceability, no shared methodology so new hires improvise, and management cannot get a clear ROI answer. This project fills the gap around the tools. The contents fall into five blocks. The first is an Enterprise AI Coding Maturity Model called E-ACMM, with five levels that describe a team from individual exploration up to AI coding as part of the organization's core. The second is a 14 document Spec matrix grouped under Meta, Proposal, Spec, Design, Plan, Test, and Trace, so that requirements, user stories, functional spec, non functional requirements, architecture, API contracts, data model, security, implementation plan, test strategy, and a traceability matrix all stay linked. The third block is the WAF, a Well Architected Framework for AI Coding with six pillars covering intent and standards, context and memory, human and AI collaboration, quality and safety, efficiency and ROI, and organization and culture. The fourth block is a lab library of 87 standardized experiments sorted across the full software life cycle, AI application development, vertical industry scenes, and a Spec specific track. The fifth block is an enterprise Skill Hub guide based on the Anthropic Skill format for internal sharing. The README compares the project to OpenSpec, positioning itself as an enterprise extension that adds the maturity model, cross document traceability, a bad smell catalog, and customer ready deliverables. Three quick start paths are described for new teams, teams already using AI with unstable code quality, and managers who need ROI numbers. The license is MIT, and the author asks for a Star as the main form of support.