[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6GK237FW2Q0KZ8DFNME6DB5":3,"skill-same:01M6GK237FW2Q0KZ8DFNME6DB5":70},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":22,"publishedAt":27,"linkOnly":28,"markdownContent":29,"markdownBody":30,"frontMatter":31,"refName":36,"commitSha":37,"skillDirectory":38,"forksCount":13,"repositoryHtmlUrl":39,"repositoryHomepageUrl":6,"ownerHtmlUrl":40,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":41,"install":42,"resources":55},"01M6GK237FW2Q0KZ8DFNME6DB5","子代理驱动开发",null,"在当前会话中执行包含独立任务的实现计划时使用","lihah111222333-cloud\u002FSuper-Dolphin-Agent-Cockpit","lihah111222333-cloud","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F254297992?v=4","01M6GJZ7ZK31PVRBS5XQWS56ET","01M6GJZ84Q0ANFYMPVK56PYSNW",0,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Flihah111222333-cloud\u002FSuper-Dolphin-Agent-Cockpit\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002F子代理驱动开发\u002FSKILL.md",[17],{"categoryNo":18,"categoryName":19,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000030","开发工程师","PROFESSION","适用职业",[5,23,24,25,26],"development","subagent","driven","子代理开发","2026-11-10 10:12:30",false,"---\nname: 子代理驱动开发\ndescription: 在当前会话中执行包含独立任务的实现计划时使用\naliases: [\"@子代理驱动开发\", \"@子代理开发\", \"@subagent-driven-development\"]\n---\n\n# 子代理驱动开发\n\n通过为每个任务派发新的子代理来执行计划，并在每个任务后进行两阶段审查：先审查规格符合性，再审查代码质量。\n\n**为什么使用子代理：** 你把任务委派给拥有隔离上下文的专门代理。通过精确构造它们的指令和上下文，你可以确保它们保持聚焦并完成任务。它们不应该继承你的会话上下文或历史；你要只构造它们需要的内容。这也能保留你自己的上下文，用于协调工作。\n\n**核心原则：** 每个任务使用新子代理 + 两阶段审查（先规格、后质量）= 高质量、快速迭代\n\n**super-agent-v3 编排选择：** 子代理生命周期不强制绑定 mcp-orch。优先使用当前平台可用的原生子代理能力；只有任务需要持久 DAG、重试、租约、cron\u002Fwakeup 或结构化交接记录时，才可选使用 `task_create_dag`、`task_start_dag`、`task_dispatch_node`、`task_update_node`。缺少 mcp-go-agent-orchestration 工具不是阻断条件；继续使用可用的子代理能力，或在不适合派发时改为当前会话执行并说明观测限制。\n\n## 何时使用\n\n```dot\ndigraph when_to_use {\n    \"Have implementation plan?\" [shape=diamond];\n    \"Tasks mostly independent?\" [shape=diamond];\n    \"Stay in this session?\" [shape=diamond];\n    \"子代理驱动开发\" [shape=box];\n    \"执行计划\" [shape=box];\n    \"Manual execution or brainstorm first\" [shape=box];\n\n    \"Have implementation plan?\" -> \"Tasks mostly independent?\" [label=\"yes\"];\n    \"Have implementation plan?\" -> \"Manual execution or brainstorm first\" [label=\"no\"];\n    \"Tasks mostly independent?\" -> \"Stay in this session?\" [label=\"yes\"];\n    \"Tasks mostly independent?\" -> \"Manual execution or brainstorm first\" [label=\"no - tightly coupled\"];\n    \"Stay in this session?\" -> \"子代理驱动开发\" [label=\"yes\"];\n    \"Stay in this session?\" -> \"执行计划\" [label=\"no - parallel session\"];\n}\n```\n\n**相比 Executing Plans（并行会话）：**\n- 同一会话（无需切换上下文）\n- 每个任务使用新子代理（避免上下文污染）\n- 每个任务后两阶段审查：先规格符合性，再代码质量\n- 迭代更快（任务之间不需要人工介入）\n\n## 流程\n\n```dot\ndigraph process {\n    rankdir=TB;\n\n    subgraph cluster_per_task {\n        label=\"Per Task\";\n        \"Dispatch implementer subagent (.\u002Fimplementer-prompt.md)\" [shape=box];\n        \"Implementer subagent asks questions?\" [shape=diamond];\n        \"Answer questions, provide context\" [shape=box];\n        \"Implementer subagent implements, tests, commits, self-reviews; record status\" [shape=box];\n        \"Dispatch spec reviewer subagent (.\u002Fspec-reviewer-prompt.md)\" [shape=box];\n        \"Spec reviewer subagent confirms code matches spec?\" [shape=diamond];\n        \"Implementer subagent fixes spec gaps\" [shape=box];\n        \"Dispatch code quality reviewer subagent (.\u002Fcode-quality-reviewer-prompt.md)\" [shape=box];\n        \"Code quality reviewer subagent approves?\" [shape=diamond];\n        \"Implementer subagent fixes quality issues\" [shape=box];\n        \"Mark task done and update_plan complete\" [shape=box];\n    }\n\n    \"Read plan, extract all tasks with full text, note context, choose dispatch path + update_plan\" [shape=box];\n    \"More tasks remain?\" [shape=diamond];\n    \"Dispatch final code reviewer subagent for entire implementation\" [shape=box];\n    \"Use superpowers:结束开发分支\" [shape=box style=filled fillcolor=lightgreen];\n\n    \"Read plan, extract all tasks with full text, note context, choose dispatch path + update_plan\" -> \"Dispatch implementer subagent (.\u002Fimplementer-prompt.md)\";\n    \"Dispatch implementer subagent (.\u002Fimplementer-prompt.md)\" -> \"Implementer subagent asks questions?\";\n    \"Implementer subagent asks questions?\" -> \"Answer questions, provide context\" [label=\"yes\"];\n    \"Answer questions, provide context\" -> \"Dispatch implementer subagent (.\u002Fimplementer-prompt.md)\";\n    \"Implementer subagent asks questions?\" -> \"Implementer subagent implements, tests, commits, self-reviews; record status\" [label=\"no\"];\n    \"Implementer subagent implements, tests, commits, self-reviews; record status\" -> \"Dispatch spec reviewer subagent (.\u002Fspec-reviewer-prompt.md)\";\n    \"Dispatch spec reviewer subagent (.\u002Fspec-reviewer-prompt.md)\" -> \"Spec reviewer subagent confirms code matches spec?\";\n    \"Spec reviewer subagent confirms code matches spec?\" -> \"Implementer subagent fixes spec gaps\" [label=\"no\"];\n    \"Implementer subagent fixes spec gaps\" -> \"Dispatch spec reviewer subagent (.\u002Fspec-reviewer-prompt.md)\" [label=\"re-review\"];\n    \"Spec reviewer subagent confirms code matches spec?\" -> \"Dispatch code quality reviewer subagent (.\u002Fcode-quality-reviewer-prompt.md)\" [label=\"yes\"];\n    \"Dispatch code quality reviewer subagent (.\u002Fcode-quality-reviewer-prompt.md)\" -> \"Code quality reviewer subagent approves?\";\n    \"Code quality reviewer subagent approves?\" -> \"Implementer subagent fixes quality issues\" [label=\"no\"];\n    \"Implementer subagent fixes quality issues\" -> \"Dispatch code quality reviewer subagent (.\u002Fcode-quality-reviewer-prompt.md)\" [label=\"re-review\"];\n    \"Code quality reviewer subagent approves?\" -> \"Mark task done and update_plan complete\" [label=\"yes\"];\n    \"Mark task done and update_plan complete\" -> \"More tasks remain?\";\n    \"More tasks remain?\" -> \"Dispatch implementer subagent (.\u002Fimplementer-prompt.md)\" [label=\"yes\"];\n    \"More tasks remain?\" -> \"Dispatch final code reviewer subagent for entire implementation\" [label=\"no\"];\n    \"Dispatch final code reviewer subagent for entire implementation\" -> \"Use superpowers:结束开发分支\";\n}\n```\n\n## 模型选择\n\n使用能胜任每个角色的最低能力模型，以节省成本并提高速度。\n\n**机械性实现任务**（隔离函数、明确规格、1-2 个文件）：使用快速、便宜的模型。当计划写得足够明确时，大多数实现任务都是机械性的。\n\n**集成和判断任务**（多文件协调、模式匹配、调试）：使用标准模型。\n\n**架构、设计和审查任务**：使用可用的最强模型。\n\n**任务复杂度信号：**\n- 触及 1-2 个文件且规格完整 → 便宜模型\n- 触及多个文件并有集成问题 → 标准模型\n- 需要设计判断或广泛理解代码库 → 最强模型\n\n## 处理实现者状态\n\n实现者子代理会报告四种状态之一。按下面方式处理：\n\n**DONE：** 进入规格符合性审查。\n\n**DONE_WITH_CONCERNS：** 实现者完成了工作，但标记了疑虑。继续前阅读疑虑。如果疑虑涉及正确性或范围，先处理再审查。如果只是观察（例如“这个文件变大了”），记录后继续审查。\n\n**NEEDS_CONTEXT：** 实现者需要未提供的信息。提供缺失上下文并重新派发。\n\n**BLOCKED：** 实现者无法完成任务。评估阻塞：\n1. 如果是上下文问题，提供更多上下文并用同一模型重新派发\n2. 如果任务需要更多推理，用更强模型重新派发\n3. 如果任务太大，拆成更小部分\n4. 如果计划本身错误，升级给用户\n\n**绝不要**忽略升级，也不要在没有变化的情况下强迫同一模型重试。如果实现者说卡住了，就需要改变一些东西。\n\n## 提示词模板\n\n- `.\u002Fimplementer-prompt.md`：派发实现者子代理\n- `.\u002Fspec-reviewer-prompt.md`：派发规格符合性审查子代理\n- `.\u002Fcode-quality-reviewer-prompt.md`：派发代码质量审查子代理\n\n## 示例工作流\n\n```\nYou: 我正在使用子代理驱动开发来执行这份计划。\n\n[Read plan file once: docs\u002Fsuperpowers\u002Fplans\u002Ffeature-plan.md]\n[Extract all 5 tasks with full text and context]\n[Choose native subagent dispatch or optional mcp-orch; create update_plan with all tasks]\n\nTask 1: Hook installation script\n\n[Get Task 1 text and context (already extracted)]\n[Dispatch implementation subagent with full task text + context]\n\nImplementer: \"Before I begin - should the hook be installed at user or system level?\"\n\nYou: \"User level (~\u002F.config\u002Fsuperpowers\u002Fhooks\u002F)\"\n\nImplementer: \"Got it. Implementing now...\"\n[Later] Implementer:\n  - Implemented install-hook command\n  - Added tests, 5\u002F5 passing\n  - Self-review: Found I missed --force flag, added it\n  - Committed\n\n[Dispatch spec compliance reviewer]\nSpec reviewer: ✅ Spec compliant - all requirements met, nothing extra\n\n[Get git SHAs, dispatch code quality reviewer]\nCode reviewer: Strengths: Good test coverage, clean. Issues: None. Approved.\n\n[Record status; if using mcp-orch, task_update_node status=done; mark Task 1 complete]\n\nTask 2: Recovery modes\n\n[Get Task 2 text and context (already extracted)]\n[Dispatch implementation subagent with full task text + context]\n\nImplementer: [No questions, proceeds]\nImplementer:\n  - Added verify\u002Frepair modes\n  - 8\u002F8 tests passing\n  - Self-review: All good\n  - Committed\n\n[Dispatch spec compliance reviewer]\nSpec reviewer: ❌ Issues:\n  - Missing: Progress reporting (spec says \"report every 100 items\")\n  - Extra: Added --json flag (not requested)\n\n[Implementer fixes issues]\nImplementer: Removed --json flag, added progress reporting\n\n[Spec reviewer reviews again]\nSpec reviewer: ✅ Spec compliant now\n\n[Dispatch code quality reviewer]\nCode reviewer: Strengths: Solid. Issues (Important): Magic number (100)\n\n[Implementer fixes]\nImplementer: Extracted PROGRESS_INTERVAL constant\n\n[Code reviewer reviews again]\nCode reviewer: ✅ Approved\n\n[Mark Task 2 complete]\n\n...\n\n[After all tasks]\n[Dispatch final code-reviewer]\nFinal reviewer: All requirements met, ready to merge\n\nDone!\n```\n\n## 优势\n\n**相比手动执行：**\n- 子代理会自然遵循 TDD\n- 每个任务都有新上下文（减少混淆）\n- 并行安全（子代理不会相互干扰）\n- 子代理可以提问（工作前和工作中都可以）\n\n**相比 Executing Plans：**\n- 同一会话（无需交接）\n- 持续推进（无需等待）\n- 自动审查检查点\n\n**效率提升：**\n- 没有文件读取开销（控制者提供完整文本）\n- 控制者精确筛选所需上下文\n- 子代理一开始就获得完整信息\n- 问题在工作开始前暴露（不是结束后）\n\n**质量门槛：**\n- 自审在交接前发现问题\n- 两阶段审查：规格符合性，然后代码质量\n- 审查循环确保修复确实有效\n- 规格符合性防止做多或做少\n- 代码质量确保实现构建得好\n\n**成本：**\n- 更多子代理调用（每个任务一个实现者 + 两个审查者）\n- 控制者需要更多准备工作（预先提取所有任务）\n- 审查循环增加迭代\n- 但能早发现问题（比之后调试更便宜）\n\n## 红旗\n\n**绝不要：**\n- 未经用户明确同意就在 main\u002Fmaster 分支上开始实现\n- 如果本轮已经选择 mcp-orch，却跳过 `task_create_dag` \u002F `task_start_dag` \u002F `task_dispatch_node` \u002F `task_update_node` 状态记录\n- 未使用 mcp-orch 时伪造 DAG\u002Fnode\u002Frun 证据\n- 跳过审查（规格符合性或代码质量）\n- 带着未修复问题继续\n- 并行派发多个实现子代理（会冲突）\n- 让子代理读取计划文件（改为提供完整文本）\n- 跳过场景上下文（子代理需要理解任务位置）\n- 忽略子代理问题（先回答再让它继续）\n- 在规格符合性上接受“差不多”（规格审查者发现问题 = 未完成）\n- 跳过审查循环（审查者发现问题 = 实现者修复 = 再审查）\n- 让实现者自审替代真正审查（两者都需要）\n- **在规格符合性 ✅ 前开始代码质量审查**（顺序错误）\n- 当任一审查还有未解决问题时进入下一个任务\n\n**如果子代理提问：**\n- 清晰完整地回答\n- 需要时提供额外上下文\n- 不要催它们直接实现\n\n**如果审查者发现问题：**\n- 实现者（同一子代理）修复\n- 审查者再次审查\n- 重复直到批准\n- 不要跳过复审\n\n**如果子代理任务失败：**\n- 派发带具体指令的修复子代理\n- 不要手动修复（会污染上下文）\n\n## 集成\n\n**必需工作流技能：**\n- **superpowers:使用git工作区**：必需，开始前设置隔离工作区\n- **superpowers:编写计划**：创建此技能执行的计划\n- **superpowers:请求代码审查**：审查者子代理使用的代码审查模板\n- **superpowers:结束开发分支**：所有任务完成后结束开发\n\n**子代理应使用：**\n- **superpowers:测试驱动开发**：子代理为每个任务遵循 TDD\n\n**替代工作流：**\n- **superpowers:执行计划**：用于并行会话，而不是同会话执行\n","# 子代理驱动开发\n\n通过为每个任务派发新的子代理来执行计划，并在每个任务后进行两阶段审查：先审查规格符合性，再审查代码质量。\n\n**为什么使用子代理：** 你把任务委派给拥有隔离上下文的专门代理。通过精确构造它们的指令和上下文，你可以确保它们保持聚焦并完成任务。它们不应该继承你的会话上下文或历史；你要只构造它们需要的内容。这也能保留你自己的上下文，用于协调工作。\n\n**核心原则：** 每个任务使用新子代理 + 两阶段审查（先规格、后质量）= 高质量、快速迭代\n\n**super-agent-v3 编排选择：** 子代理生命周期不强制绑定 mcp-orch。优先使用当前平台可用的原生子代理能力；只有任务需要持久 DAG、重试、租约、cron\u002Fwakeup 或结构化交接记录时，才可选使用 `task_create_dag`、`task_start_dag`、`task_dispatch_node`、`task_update_node`。缺少 mcp-go-agent-orchestration 工具不是阻断条件；继续使用可用的子代理能力，或在不适合派发时改为当前会话执行并说明观测限制。\n\n## 何时使用\n\n```dot\ndigraph when_to_use {\n    \"Have implementation plan?\" [shape=diamond];\n    \"Tasks mostly independent?\" [shape=diamond];\n    \"Stay in this session?\" [shape=diamond];\n    \"子代理驱动开发\" [shape=box];\n    \"执行计划\" [shape=box];\n    \"Manual execution or brainstorm first\" [shape=box];\n\n    \"Have implementation plan?\" -> \"Tasks mostly independent?\" [label=\"yes\"];\n    \"Have implementation plan?\" -> \"Manual execution or brainstorm first\" [label=\"no\"];\n    \"Tasks mostly independent?\" -> \"Stay in this session?\" [label=\"yes\"];\n    \"Tasks mostly independent?\" -> \"Manual execution or brainstorm first\" [label=\"no - tightly coupled\"];\n    \"Stay in this session?\" -> \"子代理驱动开发\" [label=\"yes\"];\n    \"Stay in this session?\" -> \"执行计划\" [label=\"no - parallel session\"];\n}\n```\n\n**相比 Executing Plans（并行会话）：**\n- 同一会话（无需切换上下文）\n- 每个任务使用新子代理（避免上下文污染）\n- 每个任务后两阶段审查：先规格符合性，再代码质量\n- 迭代更快（任务之间不需要人工介入）\n\n## 流程\n\n```dot\ndigraph process {\n    rankdir=TB;\n\n    subgraph cluster_per_task {\n        label=\"Per Task\";\n        \"Dispatch implementer subagent (.\u002Fimplementer-prompt.md)\" [shape=box];\n        \"Implementer subagent asks questions?\" [shape=diamond];\n        \"Answer questions, provide context\" [shape=box];\n        \"Implementer subagent implements, tests, commits, self-reviews; record status\" [shape=box];\n        \"Dispatch spec reviewer subagent (.\u002Fspec-reviewer-prompt.md)\" [shape=box];\n        \"Spec reviewer subagent confirms code matches spec?\" [shape=diamond];\n        \"Implementer subagent fixes spec gaps\" [shape=box];\n        \"Dispatch code quality reviewer subagent (.\u002Fcode-quality-reviewer-prompt.md)\" [shape=box];\n        \"Code quality reviewer subagent approves?\" [shape=diamond];\n        \"Implementer subagent fixes quality issues\" [shape=box];\n        \"Mark task done and update_plan complete\" [shape=box];\n    }\n\n    \"Read plan, extract all tasks with full text, note context, choose dispatch path + update_plan\" [shape=box];\n    \"More tasks remain?\" [shape=diamond];\n    \"Dispatch final code reviewer subagent for entire implementation\" [shape=box];\n    \"Use superpowers:结束开发分支\" [shape=box style=filled fillcolor=lightgreen];\n\n    \"Read plan, extract all tasks with full text, note context, choose dispatch path + update_plan\" -> \"Dispatch implementer subagent (.\u002Fimplementer-prompt.md)\";\n    \"Dispatch implementer subagent (.\u002Fimplementer-prompt.md)\" -> \"Implementer subagent asks questions?\";\n    \"Implementer subagent asks questions?\" -> \"Answer questions, provide context\" [label=\"yes\"];\n    \"Answer questions, provide context\" -> \"Dispatch implementer subagent (.\u002Fimplementer-prompt.md)\";\n    \"Implementer subagent asks questions?\" -> \"Implementer subagent implements, tests, commits, self-reviews; record status\" [label=\"no\"];\n    \"Implementer subagent implements, tests, commits, self-reviews; record status\" -> \"Dispatch spec reviewer subagent (.\u002Fspec-reviewer-prompt.md)\";\n    \"Dispatch spec reviewer subagent (.\u002Fspec-reviewer-prompt.md)\" -> \"Spec reviewer subagent confirms code matches spec?\";\n    \"Spec reviewer subagent confirms code matches spec?\" -> \"Implementer subagent fixes spec gaps\" [label=\"no\"];\n    \"Implementer subagent fixes spec gaps\" -> \"Dispatch spec reviewer subagent (.\u002Fspec-reviewer-prompt.md)\" [label=\"re-review\"];\n    \"Spec reviewer subagent confirms code matches spec?\" -> \"Dispatch code quality reviewer subagent (.\u002Fcode-quality-reviewer-prompt.md)\" [label=\"yes\"];\n    \"Dispatch code quality reviewer subagent (.\u002Fcode-quality-reviewer-prompt.md)\" -> \"Code quality reviewer subagent approves?\";\n    \"Code quality reviewer subagent approves?\" -> \"Implementer subagent fixes quality issues\" [label=\"no\"];\n    \"Implementer subagent fixes quality issues\" -> \"Dispatch code quality reviewer subagent (.\u002Fcode-quality-reviewer-prompt.md)\" [label=\"re-review\"];\n    \"Code quality reviewer subagent approves?\" -> \"Mark task done and update_plan complete\" [label=\"yes\"];\n    \"Mark task done and update_plan complete\" -> \"More tasks remain?\";\n    \"More tasks remain?\" -> \"Dispatch implementer subagent (.\u002Fimplementer-prompt.md)\" [label=\"yes\"];\n    \"More tasks remain?\" -> \"Dispatch final code reviewer subagent for entire implementation\" [label=\"no\"];\n    \"Dispatch final code reviewer subagent for entire implementation\" -> \"Use superpowers:结束开发分支\";\n}\n```\n\n## 模型选择\n\n使用能胜任每个角色的最低能力模型，以节省成本并提高速度。\n\n**机械性实现任务**（隔离函数、明确规格、1-2 个文件）：使用快速、便宜的模型。当计划写得足够明确时，大多数实现任务都是机械性的。\n\n**集成和判断任务**（多文件协调、模式匹配、调试）：使用标准模型。\n\n**架构、设计和审查任务**：使用可用的最强模型。\n\n**任务复杂度信号：**\n- 触及 1-2 个文件且规格完整 → 便宜模型\n- 触及多个文件并有集成问题 → 标准模型\n- 需要设计判断或广泛理解代码库 → 最强模型\n\n## 处理实现者状态\n\n实现者子代理会报告四种状态之一。按下面方式处理：\n\n**DONE：** 进入规格符合性审查。\n\n**DONE_WITH_CONCERNS：** 实现者完成了工作，但标记了疑虑。继续前阅读疑虑。如果疑虑涉及正确性或范围，先处理再审查。如果只是观察（例如“这个文件变大了”），记录后继续审查。\n\n**NEEDS_CONTEXT：** 实现者需要未提供的信息。提供缺失上下文并重新派发。\n\n**BLOCKED：** 实现者无法完成任务。评估阻塞：\n1. 如果是上下文问题，提供更多上下文并用同一模型重新派发\n2. 如果任务需要更多推理，用更强模型重新派发\n3. 如果任务太大，拆成更小部分\n4. 如果计划本身错误，升级给用户\n\n**绝不要**忽略升级，也不要在没有变化的情况下强迫同一模型重试。如果实现者说卡住了，就需要改变一些东西。\n\n## 提示词模板\n\n- `.\u002Fimplementer-prompt.md`：派发实现者子代理\n- `.\u002Fspec-reviewer-prompt.md`：派发规格符合性审查子代理\n- `.\u002Fcode-quality-reviewer-prompt.md`：派发代码质量审查子代理\n\n## 示例工作流\n\n```\nYou: 我正在使用子代理驱动开发来执行这份计划。\n\n[Read plan file once: docs\u002Fsuperpowers\u002Fplans\u002Ffeature-plan.md]\n[Extract all 5 tasks with full text and context]\n[Choose native subagent dispatch or optional mcp-orch; create update_plan with all tasks]\n\nTask 1: Hook installation script\n\n[Get Task 1 text and context (already extracted)]\n[Dispatch implementation subagent with full task text + context]\n\nImplementer: \"Before I begin - should the hook be installed at user or system level?\"\n\nYou: \"User level (~\u002F.config\u002Fsuperpowers\u002Fhooks\u002F)\"\n\nImplementer: \"Got it. Implementing now...\"\n[Later] Implementer:\n  - Implemented install-hook command\n  - Added tests, 5\u002F5 passing\n  - Self-review: Found I missed --force flag, added it\n  - Committed\n\n[Dispatch spec compliance reviewer]\nSpec reviewer: ✅ Spec compliant - all requirements met, nothing extra\n\n[Get git SHAs, dispatch code quality reviewer]\nCode reviewer: Strengths: Good test coverage, clean. Issues: None. Approved.\n\n[Record status; if using mcp-orch, task_update_node status=done; mark Task 1 complete]\n\nTask 2: Recovery modes\n\n[Get Task 2 text and context (already extracted)]\n[Dispatch implementation subagent with full task text + context]\n\nImplementer: [No questions, proceeds]\nImplementer:\n  - Added verify\u002Frepair modes\n  - 8\u002F8 tests passing\n  - Self-review: All good\n  - Committed\n\n[Dispatch spec compliance reviewer]\nSpec reviewer: ❌ Issues:\n  - Missing: Progress reporting (spec says \"report every 100 items\")\n  - Extra: Added --json flag (not requested)\n\n[Implementer fixes issues]\nImplementer: Removed --json flag, added progress reporting\n\n[Spec reviewer reviews again]\nSpec reviewer: ✅ Spec compliant now\n\n[Dispatch code quality reviewer]\nCode reviewer: Strengths: Solid. Issues (Important): Magic number (100)\n\n[Implementer fixes]\nImplementer: Extracted PROGRESS_INTERVAL constant\n\n[Code reviewer reviews again]\nCode reviewer: ✅ Approved\n\n[Mark Task 2 complete]\n\n...\n\n[After all tasks]\n[Dispatch final code-reviewer]\nFinal reviewer: All requirements met, ready to merge\n\nDone!\n```\n\n## 优势\n\n**相比手动执行：**\n- 子代理会自然遵循 TDD\n- 每个任务都有新上下文（减少混淆）\n- 并行安全（子代理不会相互干扰）\n- 子代理可以提问（工作前和工作中都可以）\n\n**相比 Executing Plans：**\n- 同一会话（无需交接）\n- 持续推进（无需等待）\n- 自动审查检查点\n\n**效率提升：**\n- 没有文件读取开销（控制者提供完整文本）\n- 控制者精确筛选所需上下文\n- 子代理一开始就获得完整信息\n- 问题在工作开始前暴露（不是结束后）\n\n**质量门槛：**\n- 自审在交接前发现问题\n- 两阶段审查：规格符合性，然后代码质量\n- 审查循环确保修复确实有效\n- 规格符合性防止做多或做少\n- 代码质量确保实现构建得好\n\n**成本：**\n- 更多子代理调用（每个任务一个实现者 + 两个审查者）\n- 控制者需要更多准备工作（预先提取所有任务）\n- 审查循环增加迭代\n- 但能早发现问题（比之后调试更便宜）\n\n## 红旗\n\n**绝不要：**\n- 未经用户明确同意就在 main\u002Fmaster 分支上开始实现\n- 如果本轮已经选择 mcp-orch，却跳过 `task_create_dag` \u002F `task_start_dag` \u002F `task_dispatch_node` \u002F `task_update_node` 状态记录\n- 未使用 mcp-orch 时伪造 DAG\u002Fnode\u002Frun 证据\n- 跳过审查（规格符合性或代码质量）\n- 带着未修复问题继续\n- 并行派发多个实现子代理（会冲突）\n- 让子代理读取计划文件（改为提供完整文本）\n- 跳过场景上下文（子代理需要理解任务位置）\n- 忽略子代理问题（先回答再让它继续）\n- 在规格符合性上接受“差不多”（规格审查者发现问题 = 未完成）\n- 跳过审查循环（审查者发现问题 = 实现者修复 = 再审查）\n- 让实现者自审替代真正审查（两者都需要）\n- **在规格符合性 ✅ 前开始代码质量审查**（顺序错误）\n- 当任一审查还有未解决问题时进入下一个任务\n\n**如果子代理提问：**\n- 清晰完整地回答\n- 需要时提供额外上下文\n- 不要催它们直接实现\n\n**如果审查者发现问题：**\n- 实现者（同一子代理）修复\n- 审查者再次审查\n- 重复直到批准\n- 不要跳过复审\n\n**如果子代理任务失败：**\n- 派发带具体指令的修复子代理\n- 不要手动修复（会污染上下文）\n\n## 集成\n\n**必需工作流技能：**\n- **superpowers:使用git工作区**：必需，开始前设置隔离工作区\n- **superpowers:编写计划**：创建此技能执行的计划\n- **superpowers:请求代码审查**：审查者子代理使用的代码审查模板\n- **superpowers:结束开发分支**：所有任务完成后结束开发\n\n**子代理应使用：**\n- **superpowers:测试驱动开发**：子代理为每个任务遵循 TDD\n\n**替代工作流：**\n- **superpowers:执行计划**：用于并行会话，而不是同会话执行\n",{"name":5,"aliases":32,"description":7},[33,34,35],"@子代理驱动开发","@子代理开发","@subagent-driven-development","main","287849f4ce06aaaa7e8f517126f73b016043d332",".agent\u002Fskills\u002F子代理驱动开发","https:\u002F\u002Fgithub.com\u002Flihah111222333-cloud\u002FSuper-Dolphin-Agent-Cockpit","https:\u002F\u002Fgithub.com\u002Flihah111222333-cloud","2026-10-11 02:13:17",{"source":8,"skillName":5,"commands":43,"prompt":53,"zipAvailable":54},[44,47,50],{"tool":45,"command":46},"npx","npx skills add lihah111222333-cloud\u002FSuper-Dolphin-Agent-Cockpit --skill \"子代理驱动开发\"",{"tool":48,"command":49},"bunx","bunx skills add lihah111222333-cloud\u002FSuper-Dolphin-Agent-Cockpit --skill \"子代理驱动开发\"",{"tool":51,"command":52},"pnpm","pnpm dlx skills add lihah111222333-cloud\u002FSuper-Dolphin-Agent-Cockpit --skill \"子代理驱动开发\"","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Flihah111222333-cloud\u002FSuper-Dolphin-Agent-Cockpit\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002F子代理驱动开发\u002FSKILL.md",true,[56,62,66],{"path":57,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":60,"sourceHtmlUrl":61},"code-quality-reviewer-prompt.md","REFERENCE","参考文档",1454,"https:\u002F\u002Fgithub.com\u002Flihah111222333-cloud\u002FSuper-Dolphin-Agent-Cockpit\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002F子代理驱动开发\u002Fcode-quality-reviewer-prompt.md",{"path":63,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":64,"sourceHtmlUrl":65},"implementer-prompt.md",4787,"https:\u002F\u002Fgithub.com\u002Flihah111222333-cloud\u002FSuper-Dolphin-Agent-Cockpit\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002F子代理驱动开发\u002Fimplementer-prompt.md",{"path":67,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":68,"sourceHtmlUrl":69},"spec-reviewer-prompt.md",2530,"https:\u002F\u002Fgithub.com\u002Flihah111222333-cloud\u002FSuper-Dolphin-Agent-Cockpit\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002F子代理驱动开发\u002Fspec-reviewer-prompt.md",[71,101,117,140,156,170],{"skillNo":72,"name":73,"nameZh":6,"description":74,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":75,"categories":76,"keywords":82,"publishedAt":27},"01M6GK59PDCBRZQ6D03BGXM8XM","mcp-lsp-project-config","配置、安装、检查、修复或迁移项目级 mcp-lsp stdio MCP，覆盖 Codex、Claude Code 和 Google Antigravity，以及 Windows 原生、WSL 原生和 Windows 宿主通过 wsl.exe 桥接 Linux sidecar 的自动选型。用户提到项目 MCP、LSP MCP、.codex\u002Fconfig.toml、.mcp.json、.agents\u002Fmcp_config.json、Codex 中 LSP 工具有时直接暴露有时只在延迟工具目录、ALL_TOOLS 过滤为空、工具已显示但调用报 client is closing、EOF、Transport closed，或希望三个客户端共用仓库内 mcp-lsp 时使用。","https:\u002F\u002Fgithub.com\u002Flihah111222333-cloud\u002FSuper-Dolphin-Agent-Cockpit\u002Fblob\u002Fmain\u002Fbin\u002FLSP\u002Fmcp-lsp-project-config-skill\u002FSKILL.md",[77],{"categoryNo":78,"categoryName":79,"taxonomyType":80,"taxonomyTypeDesc":81},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[83,84,85,86,87,88,89,90,91,92,93,94,95,96,97,98,99,100],"mcp lsp project config","MCP","config","lsp","project","mcp lsp","google","linux","antigravity","codex","windows","code","Tools","json","toml","client","工具","closing",{"skillNo":102,"name":103,"nameZh":6,"description":104,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":105,"categories":106,"keywords":107,"publishedAt":27},"01M6GK57H2A7XK3QZ00HCV9XM0","注释规范","当在 super-agent-v3 中编写或审查 Go 函数级中文注释、React hooks\u002Fservice\u002Fcontroller 注释，或处理注释守卫失败时使用。","https:\u002F\u002Fgithub.com\u002Flihah111222333-cloud\u002FSuper-Dolphin-Agent-Cockpit\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002F注释规范\u002FSKILL.md",[],[103,108,109,110,111,112,113,114,115,116],"go","react","controller","中文","hooks","service","v3","react hooks","comments",{"skillNo":118,"name":119,"nameZh":6,"description":120,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":121,"categories":122,"keywords":126,"publishedAt":27},"01M6GK545MTTE5X1Q4NF92PNRN","字段守卫","当新增、删除、重命名或跨层传递结构化字段，以及修改 DTO、mapper、schema、API\u002FRPC、事件、配置、store、序列化或前后端契约时自动使用；追踪生产源到全部消费端并建立 fail-fast 字段守卫。","https:\u002F\u002Fgithub.com\u002Flihah111222333-cloud\u002FSuper-Dolphin-Agent-Cockpit\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002F字段守卫\u002FSKILL.md",[123],{"categoryNo":124,"categoryName":125,"taxonomyType":80,"taxonomyTypeDesc":81},"01JBSEEDCAT000000000000010","后端与接口",[119,127,128,129,130,131,85,132,133,134,135,136,137,138,139],"后端","event","payload","store","mapping","rpc","fast","api","guard","消费","schema","drift","contract",{"skillNo":141,"name":127,"nameZh":6,"description":142,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":143,"categories":144,"keywords":147,"publishedAt":27},"01M6GK50VVC89B3MMSTKBG76ZS","在 super-agent-v3 中编写、审查或重构 Go 后端代码时使用。","https:\u002F\u002Fgithub.com\u002Flihah111222333-cloud\u002FSuper-Dolphin-Agent-Cockpit\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002F后端\u002FSKILL.md",[145,146],{"categoryNo":124,"categoryName":125,"taxonomyType":80,"taxonomyTypeDesc":81},{"categoryNo":78,"categoryName":79,"taxonomyType":80,"taxonomyTypeDesc":81},[127,148,149,150,151,108,84,152,153,154,114,155,130],"sqlc","backend","cmd","golang","runner","platform","module","fx",{"skillNo":157,"name":158,"nameZh":6,"description":159,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":160,"categories":161,"keywords":165,"publishedAt":27},"01M6GK4YP5DTSBD02AF8YSRNME","前端","当在 super-agent-v3 中开发、审查或修复当前 React\u002FVite 前端 UI 时使用；前端源码只在 frontend-app。","https:\u002F\u002Fgithub.com\u002Flihah111222333-cloud\u002FSuper-Dolphin-Agent-Cockpit\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002F前端\u002FSKILL.md",[162],{"categoryNo":163,"categoryName":164,"taxonomyType":80,"taxonomyTypeDesc":81},"01JBSEEDCAT000000000000009","前端开发",[166,167,168,114,169,109],"frontend","vite","app","ui",{"skillNo":171,"name":172,"nameZh":6,"description":173,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":174,"categories":175,"keywords":176,"publishedAt":27},"01M6GK4WWB9A6N51VPCZHC3NB1","全量项目地图生成","当在 super-agent-v3 中生成、更新或校验 docs\u002Fdoc\u002Fcodemap 项目地图、AI index 或能力清单时使用。","https:\u002F\u002Fgithub.com\u002Flihah111222333-cloud\u002FSuper-Dolphin-Agent-Cockpit\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002F全量项目地图生成\u002FSKILL.md",[],[172,177,178,114,179,180,181,182,183],"地图","生成","codemap","校验","doc","index","docs"]