Source coding discipline
A² · Agentic AI × Anything:多 agent 编排方法论与治理 skill 合集(旗舰 cto-orchestration)
npx -y skills add martin1847/evolab --skill source-coding-disciplineAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 6 stars6 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Use when a task involves reading, editing, reviewing, testing, debugging, refactoring, or otherwise working with source code files. Source paths include **/*.{py,ts,tsx,js,jsx,mjs,go,rs,java,kt,swift,rb,c,cc,cpp,h,hpp,sh,bash,zsh,sql,vue,svelte}. Do not use for docs-only, planning-only, or non-code file tasks unless the task also touches source code.
SKILL.md
2.3 KB, as published. Nobody here has run it
写码纪律(碰源码文件时加载)
镜像注:四段纪律正文与本仓库
templates/rules/coding.md同源(路径相对 evolab 仓库根;本 skill 单独 cp 安装后该文件不随行,同步义务只落在仓库维护侧)—— rules 形态给支持paths:条件加载的 agent(Claude Code / omp),skill 形态给不支持的(codex 等)。 改任一边必须同步另一边(硬门:test/mirror-sync.test.sh四段逐字 diff,漂移即红); 同一 agent 勿两种形态同装(双重注入)。
先思考再编码:显式说出假设,不确定就问;多种解读全部列出、不静默选边;有更简单的路径就说出来;困惑就停下来问。
简单优先——交付物代码零投机(仅限交付物;风险枚举 / 预判 / 评审轴不受此限、该大胆发散):不做超出要求的功能;单次使用的代码不抽象;没要求的"灵活性/可配置性"不引入;不可能发生的场景不写错误处理。写了 200 行能压到 50 行的,重写。自问:"资深工程师会觉得这过度复杂吗?"
外科手术式改动:每一行改动都能直接追溯到用户请求。不"改进"邻近代码/注释/格式,顺应已有风格。自己的改动产生的孤儿(未用 import/变量/函数)要清;既有死代码指出来、不删。
目标驱动执行:把任务转成可验证目标——"加校验"→"为非法输入写测试并让它过";"修 bug"→"先写能复现的测试再修到过";"重构"→"改前改后测试都过"。多步任务先列简短计划:步骤 → 验证。强验证标准让你独立循环,弱标准("能跑")需要不停澄清。
验证诚实性:禁止声称测试通过、命令成功、文件存在,除非实际跑过 / 读过。未跑就说"未运行验证"。完成时简短说明:改了什么 / 跑了什么验证 / 什么没验证 / 剩余假设或风险。