subagent-driven-development

Use when executing implementation plans with independent tasks or facing multiple independent issues that can be investigated without shared state - dispatches fresh subagent for each task with code review between tasks.

  • Purpose - Execute plans through coordinated subagents with quality checkpoints

  • Output - Completed implementation with all tasks verified and reviewed

When to Use SADD

Use SADD when:

  • You have an implementation plan with 3+ distinct tasks

  • Tasks can be executed independently (or in clear sequence)

  • You need quality gates between implementation steps

  • Context would accumulate over a long implementation session

  • Multiple unrelated failures need parallel investigation

  • Different subsystems need changes that do not conflict

Use regular development when:

  • Single task or simple change

  • Tasks are tightly coupled and need shared understanding

  • Exploratory work where scope is undefined

  • You need human-in-the-loop feedback between every step

Usage

How It Works

SADD supports four execution strategies based on task characteristics:

Sequential Execution

For dependent tasks that must be executed in order:

Parallel Execution

For independent tasks that can run concurrently:

Parallel Investigation

Special case for fixing multiple unrelated failures:

Last updated