subagent-driven-development

Execute multi-task implementation plans with fresh subagents and two-stage reviews.

38|12|Updated Apr 24, 2026
One-click install
npx skills add https://github.com/databrickslabs/coding-agents-databricks-apps --skill subagent-driven-development-databrickslabs
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: subagent-driven-development
Source: https://github.com/databrickslabs/coding-agents-databricks-apps/tree/main/.claude/skills/subagent-driven-development
Command: npx skills add https://github.com/databrickslabs/coding-agents-databricks-apps --skill subagent-driven-development-databrickslabs

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Manually executing multi-task implementation plans leads to context pollution, missed requirements, inconsistent code quality, and slow iteration as developers switch between tasks and forget details or skip critical review steps.

Core Features & Use Cases

  • Fresh subagent per task: Dispatches a new subagent for each independent task to eliminate context carryover and confusion between work items.
  • Enforced two-stage review: Requires spec compliance review first, then code quality review for every task to catch missing requirements and implementation issues before moving forward.
  • Automated review loops: Automatically re-dispatches reviewers after fixes are applied to ensure issues are fully resolved before marking tasks complete.
  • Use Case: For a plan with 4 independent tasks like adding user authentication, updating API endpoints, writing integration tests, and updating documentation, this Skill ensures each task is built to spec, reviewed for quality, and committed without manual handoffs or skipped checks.

Quick Start

Use the subagent-driven-development skill to execute the implementation plan in docs/plans/feature-plan.md, dispatching a fresh subagent for each independent task with two-stage review after every implementation.

Frequently Asked Questions about subagent-driven-development

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

FAQPage Schema
How do I automate execution of multi-task implementation plans without context pollution?

Automate multi-task implementation plans by dispatching a fresh subagent per independent task, eliminating context carryover. This ensures each task is built to spec without developers forgetting details or switching focus between work items.

What is a two-stage code review workflow for software development?

A two-stage code review workflow enforces spec compliance review first, then code quality review for every task. This sequential process catches missing requirements and implementation issues before moving forward to the next independent task.

How do I enforce spec compliance and quality gates for each task in a development workflow?

Enforce spec compliance and quality gates by applying automated review loops after each task implementation. The system automatically re-dispatches reviewers after fixes are applied to ensure issues are fully resolved before marking tasks complete.

Can I run multiple loosely coupled software tasks in a single development session?

Yes, you can run loosely coupled software tasks in a single session without parallel handoffs. This approach suits software development scenarios where implementation plans are broken into independent tasks executable sequentially.

Why does manual execution of implementation plans lead to missed requirements?

Manual execution leads to missed requirements because developers switch between tasks and forget details or skip critical review steps. Context pollution occurs without isolated subagent context, causing inconsistent code quality and slow iteration.

When should I not use subagent dispatch for implementation plans?

Subagent dispatch is not suitable for implementation plans requiring parallel handoffs or highly interdependent tasks. It is designed for independent, loosely coupled tasks that can be executed sequentially in a single session.