Skip to content

xyaz1313/xyskill / skills/xy-workbench

xy-workbench

Verified

【多端Agent工作台搭建】把一个项目整理成 Claude Code / Codex / Grok / 通用 Agents(~/.agents/skills)多端一致、能长期维护的 Agent 工作台:审规则文件(CLAUDE.md / AGENTS.md)、认 skill 真源、统一命名、生成桥接、逐项验证。用户说「迁移到 Codex」「迁到 Claude Code」「Grok 也要能用」「统一 AGENTS.md」「我的 Agent 工作台很乱」「skill 散在好几个地方」「帮我把 Claude 和 Codex 和豆包打通」时使用。|小爷出品

FileKindSize
SKILL.mdskill md12.1 KB
agents/openai.yamlconfig315 B
evals/evals.jsonconfig1.3 KB
references/atoms.jsonlother4.3 KB

references/atoms.jsonl

View on GitHub
{"id": "XY-DY3-030", "type": "method", "confidence": "medium", "topics": ["AI与工具"], "knowledge": "九成人裸用 Claude Code 所以总翻车——不是模型不行,是让一个 AI 同时读项目、找 bug、写代码、做审查、补测试,它当然容易崩。正确用法是任务分工,装六个 Skill 各司其职:Explore 先画项目地图(入口在哪、核心文件在哪、功能调用链);Debug 拿到日志、复现步骤和最近改动后先定位根因,再给最小修复方案而不是乱修;Code Review 在改完一大段后挑逻辑漏洞、边界条件和回归风险;Test Engineer 判断哪里最容易回归、哪些测试最值得补;Code Simplifier 删重复逻辑、收命名收结构但不改变行为;Security Review 在涉及登录权限、密钥、支付、用户数据时最后扫一遍。高手不是让 AI 从头干到尾,而是把不同任务交给不同 Skill。", "source_type": "account_video", "source_label": "小爷账号-口播补挖", "chapter": null, "section": "Claude Code新手必装6个Skill", "blocks": ["12-4", "06-3"], "skills": ["xy-ai-workflow", "xy-ip", "xy-workbench"], "related": null, "q": 5}
{"id": "XY-DY3-006", "type": "method", "confidence": "high", "topics": ["AI与工具"], "knowledge": "装 OpenClaw(龙虾)时要同时装一个 Claude Code 给它做兜底,代码层面的事全让它干。原因:龙虾在自己体内升级很容易把自己搞挂,而 Claude Code 有底层代码能力,可以把龙虾的全部源码(1884 个文件)读完后边读边实施、一步步升级,双向打通后龙虾对升级过程还有体感、能反馈。分工原则就是\"专业的人干专业的事\"——除非你龙虾大脑接的模型本身接近 Claude 级别(代码与中文能力都顶级),否则把\"干活的 agent\"和\"管代码的 agent\"分开,是最稳的结构。", "source_type": "account_video", "source_label": "小爷账号-口播补挖", "chapter": null, "section": "用源码来升级openclaw龙虾", "blocks": ["12-4", "12-3"], "skills": ["xy-ai-workflow", "xy-workbench"], "related": ["XY-DY3-009"], "q": 4}
{"id": "XY-DY3-016", "type": "method", "confidence": "medium", "topics": ["AI与工具"], "knowledge": "给 Claude Code 装其他 Agent 框架(如 Hermes)不用自己折腾环境:把官方仓库下载的压缩包(约10.9MB)直接拖进 CC 对话里一丢、按回车,CC 会自动把安装全部做完,装完连安装包都可以删掉;接模型同理——要接 MiniMax、千问还是中转,把接口和 KEY 一股脑丢给 CC 让它自己配好,配完直接问它\"你现在用什么模型\"即可验证。核心思路是让 CC 代劳一切安装与配置工作,人只负责提供材料。", "source_type": "account_video", "source_label": "小爷账号-口播补挖", "chapter": null, "section": "用CC搞定Hermes", "blocks": ["12-4"], "skills": ["xy-ai-workflow", "xy-workbench"], "related": null, "q": 4}
{"id": "XY-DY3-035", "type": "method", "confidence": "medium", "topics": ["AI与工具"], "knowledge": "把知识库做成多个 Agent 的共用底座:Codex 能准确回答\"某模型做数字人不如另一家\"这种从没告诉过它的判断,因为它读得到这台电脑里的 agent 约定和全部知识库——另一边 Claude Code 一直在做的事它全知道,两边是打通的。做法就是把每一次工作的结果都沉淀进共用知识库,多 Agent 之间自然互通上下文,任何一个工具都能接着之前的积累继续干。", "source_type": "account_video", "source_label": "小爷账号-口播补挖", "chapter": null, "section": "做好你的知识库比什么都重要", "blocks": ["12-2", "12-4"], "skills": ["xy-ai-workflow", "xy-vault", "xy-workbench"], "related": null, "q": 3}
{"id": "XY-MD-033", "type": "method", "confidence": "high", "topics": ["AI与工具"], "knowledge": "技能文件五部分:触发条件、核心规则、工作流程、注意事项、更新日志。工具不是从零发明的,是从实战里沉淀的——每被问到同一个问题就变成一个标准答案,每发现 AI 犯同一个错就加一条规则。", "source_type": "adapted_book", "source_label": "外部方法论·内化改写", "chapter": null, "section": null, "blocks": ["12-3", "06-10"], "skills": ["xy-ai-workflow", "xy-workbench"], "related": null, "q": 1}
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall