jnmetacode avatar

Subagent Driven Development

This is a systematic approach to executing implementation plans by spawning fresh subagents for each task, then running two-phase reviews: spec compliance first, code quality second. You stay in one session coordinating while each subagent gets isolated context you carefully construct. The workflow is rigid by design: implementer builds and self-reviews, spec reviewer catches scope creep or gaps, quality reviewer checks patterns, then you mark it done in a TodoWrite and move to the next task. It's heavier than manual execution but catches problems early through the review loops. Works best when your plan is detailed enough that most tasks become mechanical, letting you use cheaper models for implementation and save the expensive ones for architecture decisions.

skills.sh installs
1K
GitHub stars
8.5K
last updated
Aug 21, 2026

Install

$ npx -y skills add jnmetacode/superpowers-zh --skill subagent-driven-development --agent claude-code

Installs into .claude/skills of the current project.

Files

子智能体驱动开发

通过为每个任务分派一个全新的实现子智能体来执行计划:每个任务完成后做一次任务审查(规格合规性 + 代码质量),全部任务结束后再做一次覆盖整个分支的宽范围审查。

为什么用子智能体: 你把任务委派给具有隔离上下文的专用智能体。通过精心设计它们的指令和上下文,确保它们专注并成功完成任务。它们绝不应继承你会话的上下文或历史记录——你要精确构造它们所需的一切。这样也能为你自己保留用于协调工作的上下文。

核心原则: 每个任务一个全新子智能体 + 任务审查(规格 + 质量)+ 结尾宽范围审查 = 高质量、快速迭代

旁白: 工具调用之间最多说一句简短的旁白——进度账本和工具结果本身就是记录。

持续执行: 不要在任务之间停下来向你的人类伙伴确认。不间断地执行计划里的所有任务。唯一该停下的理由是:你无法解决的 BLOCKED 状态、确实妨碍推进的歧义,或所有任务已完成。"我该继续吗?"之类的询问和进度小结都在浪费他们的时间——他们让你执行计划,那就执行。

做裁决,不要停摆。 一个正在跑的计划不等人。冲突、歧义、计划缺陷、你本来想申请突破的上限——你自己定。规格是有约束力的权威,计划是它的论证,两者都答不上来的部分由你的判断来定。每个决定都以 Ruling: <你决定了什么> — <为什么> — <如果错了代价是什么> 记进账本,然后继续。一个错误的裁决,代价是你人类伙伴看得见、也撤得掉的返工;一个停在问题上的会话,代价是他们的一整天,而且什么也换不来。

只有四件事会让你停下,也只有这四件:不可逆或破坏性的操作;涉及安全的动作;这个工作树之外、按惯例应当先问一声的副作用(合并、推送到共享分支、发布);以及一个坏到每条前进路径都只能靠猜的计划。遇到这四类,停下来问。

何时使用

digraph when_to_use {
    "有实现计划?" [shape=diamond];
    "任务基本独立?" [shape=diamond];
    "伙伴选了内联执行,或没有子智能体工具?" [shape=diamond];
    "subagent-driven-development" [shape=box];
    "executing-plans" [shape=box];
    "手动执行或先头脑风暴" [shape=box];

Sponsors

Details

Repository
github.com/jnmetacode/superpowers-zh
Category
AI & Agent Building
First seen
Jun 3, 2026

More from jnmetacode/superpowers-zh

All 20
1aradotso avatarclaude-peers-mcparadotso/trending-skillsclaude peers mcp1Kskills.sh installs
2aradotso avatarmy-brain-is-full-crewaradotso/trending-skillsmy brain is full crew1Kskills.sh installs
3daymade avatarllm-icon-finderdaymade/claude-code-skillsFinding and accessing AI/LLM model brand icons from lobe-icons library. Use when users need icon URLs, want to download brand logos for AI models/providers/appl989skills.sh installs
4heygen-com avatarai-video-genheygen-com/skillsGenerate AI videos from text prompts using the HeyGen API. Use when: (1) Generating videos from text descriptions, (2) Creating AI-generated video clips for con987skills.sh installs
5myzy-ai avatardokie-ai-pptmyzy-ai/dokie-ai-pptdokie ai ppt986skills.sh installs
6aradotso avatarai-marketing-skills-automationaradotso/marketing-skillsai marketing skills automation976skills.sh installs