The MiniMax Code team open-sources MCode Stack Agent workflows and Skills
English text is being prepared.
The MiniMax Code team has open-sourced MCode Stack, writing the Agent workflows and Skills the team uses daily into installable SKILL.md files and manifests, which can be integrated as plugins into MiniMax Code and any Agents that support Skills. The repository packages stacks by member, among which stacks/ronny/ is the most complete, containing 15 skills covering sub-Agent decomposition, context and tool budget auditing, long-session handoff, independent review, and hotfix judgment. The companion notes/ column MCode Engineering Notes explains the rationale; the six engineering-* entries are adapted from Google Engineering Practices, with the adapted portions under CC-BY 3.0 and the rest under MIT.
Why read this
AI score 65
The original article breaks down the team's daily Agent workflows into installable Skills and manifests, and provides concrete criteria for sub-Agent decomposition and context budgeting.
Topics
View original postWe just open-sourced MCode Stack: the actual workflows and skills the MiniMax Code team uses every day to build MiniMax Code. https://t.co/vREt7i8m2M
View original
We just open-sourced MCode Stack: the actual workflows and skills the MiniMax Code team uses every day to build MiniMax Code. https://t.co/vREt7i8m2M
Body
MiniMax Code @MiniMax_Code 团队把他们每天真正在用的 Agent 工作流和 Skills 开源出来了！
MCode Stack 中包含了团队何时拆子 Agent、长命令等多久等、长会话怎么保证不降级、每轮上下文塞什么 token 等等；把这些都写成可安装的 SKILL.md / 清单，并可作为插件进 MiniMax Code 和任意支持 Skills 的 Agents。
开源地址：https://t.co/tU3L7LxLF1
仓库里有什么 按人打包成 stack（每人一套插件）： · stacks/ronny/：最完整，15 个 skill · stacks/jiachen/：偏「跳出 Agent 循环、把笔记做成帖/视频」 · notes/：配套专栏 MCode Engineering Notes（文章讲道理，技能可执行） · community/、templates/stack-template/：欢迎外人贡献日常用法 · 三类物件：Skill（/技能名 或按描述自动加载）、Checklist（逐项过项目）、Workflow（WORKFLOW.md 描述日常怎么用）。Ronny 的 WORKFLOW.md 仍标 Coming soon，先以 skills 为准。
Ronny 那 15 个 Skills 1. 规划与拆分：subagent-split-check、engineering-small-prs、engineering-change-descriptions；要不要派子 Agent；大改拆成可审小 PR；写清变更与原因 2. 上下文与工具：tool-budget-audit、context-tax-audit；工具吃掉多少上下文；每轮注入了什么、对 prompt cache 的影响 3. 长会话与长命令：long-session-check、handoff-doc、long-command-check；摘要/落盘/新开会话；交接文档；长 shell/后台任务十点审计 4. 评审：independent-review、engineering-code-review 等；干净上下文只看 diff+需求+测试；代码健康评审；写/回评论 5. 实验与热修复：harness-experiment、trajectory-review、engineering-emergency-changes；一次改一点、看翻盘用例；看轨迹不只看过没过；热修是否真紧急
六条 engineering-* 改编自 Google Engineering Practices，改编部分仍属 CC-BY 3.0（见 stacks/ronny/NOTICE.md）；其余 MIT。
配套观点（Notes 01 为例） 子 Agent 首先是隔离上下文的工具，不是按「工种/问题类型」开分身；默认主循环 + 按需加载技能，只有量到上下文才是瓶颈再拆；代码评审给独立干净上下文。同系列还有工具过多吃上下文、compaction vs memory、bash 前台阻塞 vs 后台等（06-context-tax 标 upcoming）。