subagent-driven-development

Subagent-Driven Development

Skill Overview

Subagent-Driven Development is an AI-assisted programming workflow that assigns an independent subagent to implement each task, together with a two-stage review mechanism (standards and compliance review + code quality review), enabling high-quality, high-efficiency iterative development.

Applicable Scenarios

Executing implementation plans with multiple independent tasks — When you have a detailed implementation plan and the tasks are independent of one another, you can complete them one by one. This skill assigns a fresh subagent to each task to avoid context contamination, while also ensuring code quality through an automated review process.

Need rapid iteration but require quality assurance — Compared with manual execution, subagents naturally follow Test-Driven Development (TDD) principles. After each task is completed, the workflow automatically performs two-stage reviews—standards and compliance, then code quality—so you maintain fast iteration without sacrificing code quality.

Complete complex development workflows within the same session — Unlike cross-session Executing Plans, Subagent-Driven Development finishes all tasks within the same session. There’s no need for context switching, and developers can monitor progress in real time and answer questions raised by subagents.

Core Features

Independent subagent task assignment — Start a new subagent for each implementation task to ensure a clean context environment. Subagents can ask the developer questions before or during the work to get clear implementation direction.

Two-stage automatic review — After each task is completed, a standards review subagent first verifies that the code meets the requirements of the implementation plan (neither doing too much nor too little). Then a code quality review subagent evaluates the implementation quality. If any issues are found, the code is fixed and re-reviewed until it passes.

Integrated full development workflow — Seamlessly integrates with skills such as writing-plans, requesting-code-review, and finishing-a-development-branch. It forms a complete development loop—from plan creation, task execution, code review, to branch completion.

Common Questions

What is Subagent-Driven Development?

Subagent-Driven Development is an AI-assisted programming methodology. The core idea is to assign a dedicated subagent to each implementation task, rather than using the same AI assistant throughout the entire development process. Each subagent has a clean context and focuses on completing a single task. After the task is done, it undergoes a two-stage review: first a standards and compliance review (confirming the code meets the plan requirements), and then a code quality review (evaluating the quality of the implementation). This approach not only preserves task independence, but also ensures overall quality through automated review.

How is Subagent-Driven Development different from manual execution?

Compared with manual execution, Subagent-Driven Development has these advantages: (1) subagents naturally follow TDD principles by writing tests before implementing functionality; (2) each subagent uses a fresh context, preventing information mix-ups across tasks; (3) two-stage reviews are automated, requiring no manual intervention; (4) subagents can proactively ask questions during implementation to ensure the direction is correct. Manual execution requires developers to manage context themselves and proactively perform code reviews, making it easier to miss quality check points.

How does this skill differ from Executing Plans?

Both are used to execute an implementation plan, but they work differently: Executing Plans start a separate parallel session to execute all tasks at once, which is suitable for scenarios requiring fully automated execution. Subagent-Driven Development completes tasks one by one within the current session. Developers can monitor progress in real time, answer subagent questions, and have review checkpoints after each task. If you need to retain control over the development process and respond quickly to issues, choose Subagent-Driven Development. If you want fully unattended automated execution, choose Executing Plans.