MiniMax open-sources mcode-stack, the workflows and skills used to build MiniMax Code
MiniMax released mcode-stack, an open-source repository of 15 skills and checklists the MiniMax Code team uses daily. It is MIT-licensed and installs into MiniMax Code via /skills or as a plugin.
China context
- Original name
- 稀宇科技
- Outside China
- Open weights · github.com
- Claims
- Company-reported; not yet independently evaluated
- For builders
- Developers can integrate these skills into their own AI coding agents that support SKILL.md format, potentially improving code review and context management.
- For investors
- This release signals MiniMax's strategy to build a developer ecosystem around its coding tools, which could drive adoption and competitive positioning.
MiniMax published mcode-stack on GitHub, containing the engineering workflows and skills used internally to build MiniMax Code. The repository includes 15 skills covering planning, context management, long sessions, code review, and experiments, along with checklists and workflow documentation. It is open source under the MIT license and can be installed directly into MiniMax Code.
The skills use the common SKILL.md format, so most also work in other agents that support skills. Installation is via /skills command, copying the stacks/ronny directory as a plugin, or placing skills in .minimax/skills/, .claude/skills/, or .agents/skills/ directories.
Developers outside China can now adopt MiniMax's internal engineering practices for AI-assisted coding, potentially improving code quality and reducing context costs without access to MiniMax Code itself.
Open-sourcing internal workflows can attract developers to MiniMax Code and build an ecosystem around its skill format, potentially increasing adoption of MiniMax's coding tools.
The repository is expected to evolve as the team learns, with contributions encouraged. Watch for updates to the MCode Engineering Notes column and new skill versions.