What is the advantage of using fresh subagents per task?

Using a fresh subagent for each task prevents 'context pollution,' ensuring the AI focuses only on the specific task requirements without being distracted by previous implementation details or irrelevant session history.

How does the two-stage review process work?

Every task is first audited by a spec-reviewer to ensure all requirements are met. Once approved, it passes to a code-quality reviewer to check for best practices, performance, and maintainability.

How does this differ from the standard 'Executing Plans' skill?

Unlike standard plan execution which may use parallel sessions, this skill operates within the same session for faster iteration while enforcing stricter quality gates through specialized reviewer agents.

Can I use this skill without an implementation plan?

This skill is specifically designed to execute existing plans. It works best when paired with planning skills that generate structured tasks with clear acceptance criteria.

Subagent-Driven Development

Name: Subagent-Driven Development
Author: tianyuanztm

bytianyuanztm

0•

Productivity & Workflow

Executes complex implementation plans by dispatching fresh subagents for independent tasks with a rigorous two-stage review process.

Subagent-Driven Development is a specialized framework for Claude Code designed to execute multi-task implementation plans with high precision. By breaking down a plan into independent units and assigning a fresh subagent to each, the skill prevents context pollution and maintains focused iteration. The hallmark of this workflow is its mandatory two-stage quality gate: every task must first pass a specification compliance review to ensure it meets requirements, followed by a code quality review to ensure architectural integrity. This approach is ideal for developers who want to maintain the speed of AI-driven coding without sacrificing the rigor of professional software engineering standards.

Key Features

01Mandatory two-stage review (Spec Compliance then Code Quality)

02Seamless integration with TDD and Git worktree workflows

030 GitHub stars

04Built-in Q&A loop for resolving ambiguities before implementation

05Fresh subagent isolation per task to prevent context drift

06Automated task tracking using TodoWrite integration

Use Cases

01Executing independent software tasks within a single, continuous session

02Ensuring high-quality code output through automated multi-agent peer review

03Implementing complex features from a pre-defined multi-step plan

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add tianyuanztm/myplan subagent-driven-development

For use in Claude.ai and ChatGPT

Key Features

01Mandatory two-stage review (Spec Compliance then Code Quality)

02Seamless integration with TDD and Git worktree workflows

030 GitHub stars

04Built-in Q&A loop for resolving ambiguities before implementation

05Fresh subagent isolation per task to prevent context drift

06Automated task tracking using TodoWrite integration

Use Cases

01Executing independent software tasks within a single, continuous session

02Ensuring high-quality code output through automated multi-agent peer review

03Implementing complex features from a pre-defined multi-step plan

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add tianyuanztm/myplan subagent-driven-development

For use in Claude.ai and ChatGPT