jnMetaCode/superpowers-zh / skills/subagent-driven-development
subagent-driven-development
Verified当在当前会话中执行包含独立任务的实现计划时使用
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 17.7 KB |
| implementer-prompt.md | markdown | 6.0 KB |
| re-review-prompt.md | markdown | 5.0 KB |
| references/example-workflow.md | markdown | 2.5 KB |
| references/final-review.md | markdown | 1.6 KB |
| references/finishing.md | markdown | 868 B |
| references/fix-loop.md | markdown | 4.3 KB |
| references/preflight-and-resuming.md | markdown | 3.8 KB |
| references/report-handling.md | markdown | 1.6 KB |
| scripts/review-packageexecutable | script | 2.4 KB |
| scripts/sdd-repoexecutable | script | 2.3 KB |
| scripts/sdd-workspaceexecutable | script | 4.7 KB |
| scripts/task-briefexecutable | script | 4.6 KB |
| task-reviewer-prompt.md | markdown | 9.0 KB |
references/fix-loop.md
View on GitHub### 4. 修复循环
当审查报告规格 ❌、任何 Critical 或 Important 发现、或者你确认为真实缺口的 ⚠️ 条目时,循环触发。
循环开始之前,有两条路会立刻离开它:
- **Minor 发现**随手记进进度账本(`Task <N>: minor (deferred): <一句话>`),并把最终的整分支审查指向那份清单,让它去甄别哪些必须在合并前修掉。**没人读的汇总等于静默丢弃。** Minor 发现永远不进入循环。
- 被标为"计划要求的"发现——或任何与计划原文所要求的内容冲突的发现——**由你来裁决**:把这条发现放到计划原文旁边掂量,以规格为有约束力的权威作出决定,并在据此行动之前把裁决记进账本。不要因为计划要求就驳回这条发现,也不要在没有一条记录在案的裁决的情况下分派一个与计划相违的修复。
其他一切都进入循环。**一轮修复 = 一次修复分派 + 一次定向复审。每个任务最多五轮。**
**第 1-3 轮——唤回原来那个实现者(resume)。** 把未解决的发现**逐字**发给它。它的上下文是完整的:它知道任务、知道代码、知道自己做过的选择。如果你的运行环境无法给一个活着的子智能体再发消息,就分派一个全新实现者,带上简报路径、报告文件路径和那些发现——无论走哪条路,报告文件都是那份持久化记忆。
**第 4-5 轮——用更强的模型分派一个全新实现者**(按"模型选择"),带上简报路径、报告文件路径、未解决的发现,以及这样的框定语:"某个此前的实现者尝试过这个任务 [N] 次;现在它归你了。读报告文件了解已经试过什么。"一个熬过三次唤回的循环,通常意味着实现者看不见自己的问题——换新眼睛加提升能力,一步到位。
**每一轮,无论走哪条路:** 实现者修复、重跑覆盖被改动代码的测试、把修复报告追加到**同一个**报告文件、回传那个简短契约。重新分派审查者之前,先确认修复报告里含有覆盖用的测试、跑过的命令、以及输出;三者齐备才分派复审。在修复消息里点名覆盖用的测试文件——一行的修复不需要整包套件。
**复审是定向的。** 运行 `bash scripts/review-package PLAN_FILE FIX_BASE HEAD`(从正在执行的项目工作树中,用脚本的绝对路径运行),其中 FIX_BASE 是上一次审查所看到的那个 head,然后用 [re-review-prompt.md](../re-review-prompt.md) 分派,附上发现清单、简报、报告文件和打印出的 diff 路径。复审者对每条发现给出 ADDRESSED 或 NOT ADDRESSED 的结论,并且**只**标记修复 diff 里的新破坏。修复 diff 里新出现的 Critical/Important 破坏加入未解决发现清单。范围外的观察作为延后的 Minor 进账本——它们永远不延长循环。
**每轮结束后**往账本追加:
`Task <N>: fix round <R>/5 (<X> addressed, <Y> open — <发现的一句话概括>; commits <a7>..<b7>)`
**绝不在控制者会话里自己修发现**——你的上下文要保持干净以供协调,而且控制者的修复会跳过审查。
**熔断。** 当第 5 轮的复审仍然留下未解决的发现时,**停止分派**。你自己逐条裁定这些未解决的发现——你掌握着审查者所缺的计划和跨任务上下文:
- **审查者错了,或者这一点是可争议的:** 搁置它——`Task <N>: parked — <发现> — Ruling: <为什么代码可以维持原样>`。最终审查会看到双方说法。
- **是真实的,但下游没有任何东西建立在它之上:** 同样搁置,裁定里写明它是真的、被延后了。
- **真实且承重**——后面的任务建立在它之上,或者它揭示了一个计划缺陷:对**能解开后续工作的最小改动**作出裁决,以 `Task <N>: Ruling: <发现> — <你决定了什么,以及为什么>` 记进账本,并把它带进下一个任务的分派里。把一个结构性失败悄悄搁置掉,会让每个依赖它的任务都建立在它之上。只有当这个缺陷让每条前进路径都只能靠猜时,才停下来。
**只在触及上限时才裁定。** 为了结束循环而提早裁定,只是换了个名字的"预先定性"。每一次裁定都是一条账本记录——**静默丢弃是禁止的**。
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| 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: 463k | Updated | Claude CodeCodex | Verified | Install |
| 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: 463k | Updated | Claude CodeCodex | Verified | Install |
| discernment-nudgeanthropics/skillsAfter you give a substantive answer or draft that the user may act on — advice or recommendations, drafted artifacts such as goals, plans, pitches, proposals, or emails, estimates or projections, analysis or interpretation of data, factual claims they may rely on, or a multi-step argument — invoke this skill BEFORE finalizing your reply and then, if it applies, append 2-3 short follow-up questions, each tied to something specific in what you just produced, that help the user check key facts, probe the reasoning or assumptions, and notice missing context. Do this at most once per conversation. Skip it when the user asked a trivial how-to or simple lookup, wants a purely educational explanation, asked you only to format, convert, or assemble a file from content they provided, is writing code they will run, is doing creative writing or casual chat, or already asked you to double-check, cite, or review — the skill file explains these boundaries and the exact output format. | Stars: 195k | Updated | Claude CodeCodex | Verified | Install |
| internal-commsanthropics/skillsA set of resources to help me write all kinds of internal communications, using the formats that my company likes to use. Claude should use this skill whenever asked to write some sort of internal communications (status reports, leadership updates, 3P updates, company newsletters, FAQs, incident reports, project updates, etc.). | Stars: 195k | Updated | Claude CodeCodex | Verified | Install |