subagent-driven-development

Dispatch fresh subagents for isolated task execution with two-stage reviews.

1|Updated Mar 18, 2026
One-click install
npx skills add https://github.com/xxih/ai-harness-zh --skill subagent-driven-development-xxih
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: subagent-driven-development
Source: https://github.com/xxih/ai-harness-zh/tree/main/references/translations/superpowers/skills/subagent-driven-development
Command: npx skills add https://github.com/xxih/ai-harness-zh --skill subagent-driven-development-xxih

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

为每个任务派发一个全新 subagent 来执行计划,并在任务完成后进行两阶段评审(先进行规格合规性评审,再进行代码质量评审),以提升任务并行性、降低上下文污染并确保输出质量。

Core Features & Use Cases

  • 隔离执行:为每个任务启动独立的 subagent,避免上下文污染。
  • 两阶段评审:在实现完成后先做规格合规性评审,再做代码质量评审,确保产出符合预期。
  • 工作流协调:在同一会话内协调多个独立任务的派发、实现、评审与合并。
  • 快速迭代:通过并行执行和快速反馈缩短开发周期。

Quick Start

请在当前会话中为每个任务派发一个全新 subagent,并在任务完成后执行两阶段评审。

Frequently Asked Questions about subagent-driven-development

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I prevent context pollution when running parallel AI agent tasks?

To prevent context pollution during parallel execution, dispatch a fresh subagent for each task to execute the plan with isolated contexts. This per-task subagent orchestration ensures independent operation without cross-contaminating working memory.

What is the two-stage review process for AI-driven code generation?

The two-stage review process for AI code generation first performs a spec compliance review to verify requirements, then executes a code quality review. This automated sequence ensures output meets both functional specifications and technical standards.

How do I orchestrate multiple independent development tasks in a single session?

You orchestrate multiple independent tasks in a single session by using per-task subagent orchestration to dispatch, implement, review, and merge work. This workflow applies state management for task completion to coordinate parallel execution effectively.

Can I use subagent task dispatch for non-independent development tasks?

Subagent task dispatch is designed for tasks that are mostly independent within a single session. If your development tasks have heavy dependencies or require continuous shared context, this isolated execution approach may not be the best fit.

What is the best way to automate code reviews during parallel task execution?

The best way to automate code reviews during parallel execution is applying a two-stage review process after each subagent finishes. Implementing spec compliance and code quality reviews ensures faster feedback and cleaner task execution.