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

While standard execution often happens in parallel sessions, this skill stays in the current session, providing faster iteration without handoff delays while using fresh subagents to prevent context pollution.

Why use a fresh subagent for every task?

This prevents 'context pollution,' where the AI might get confused by previous attempts or unrelated code, ensuring each task is implemented with a clean state and specific focus.

What happens if a subagent fails a task?

The workflow includes a feedback loop where a follow-up subagent is dispatched specifically to address the review findings or failures, rather than requiring the user to fix it manually.

Can I use this skill without a pre-written plan?

No, this skill requires a plan generated by the 'writing-plans' skill to effectively track progress and dispatch tasks accurately.

Is code review mandatory in this workflow?

Yes, the skill is designed around quality gates. Every task implementation is followed by a code review to catch issues early before they compound.

Subagent-Driven Development

Name: Subagent-Driven Development
Author: martin-janci

bymartin-janci

•

Productivity & Workflow

Executes complex development plans by dispatching fresh subagents for individual tasks with automated quality gates and code reviews.

This skill optimizes the software development lifecycle within a single Claude session by breaking down implementation plans into discrete, independent tasks. By dispatching a fresh subagent for each task, it prevents context pollution and maintains high focus. The workflow includes mandatory code reviews between tasks and a final architectural audit, ensuring that fast iteration does not compromise code quality. It is ideal for developers who want to maintain continuous progress on a branch while ensuring every line of code meets project standards through automated iteration and feedback loops.

Key Features

01Fresh subagent dispatch per task to prevent context drift and pollution

021 GitHub stars

03Integrated Test-Driven Development (TDD) enforcement for all subagents

04Progress tracking with TodoWrite to manage multi-step implementation plans

05Final architectural review to validate entire feature sets before merging

06Automated code review checkpoints after every individual task implementation

Use Cases

01Performing complex refactors where tasks are independent but require strict quality gates

02Scaling development velocity in a single session without losing oversight or testing rigor

03Executing multi-step feature implementations defined in a structured plan

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add martin-janci/claude-marketplace subagent-driven-development

For use in Claude.ai and ChatGPT

Download Skill