When should I use the decomposing-plans skill?

Use it immediately after generating a monolithic implementation plan (typically via the writing-plans skill) but before starting the actual code implementation.

How does the skill identify parallel tasks?

The script analyzes the plan for tasks that modify different files and lack explicit sequence dependencies, grouping them into batches that can safely run at the same time.

What is the benefit of context reduction in this workflow?

By splitting a large plan into small task files, subagents only read the specific instructions they need, preventing 'lost in the middle' errors and saving significant token costs.

What happens if the script incorrectly identifies a dependency?

The skill generates a manifest.json file that you can manually edit to adjust parallel batches or task dependencies before proceeding to the execution phase.

Why is there a limit of two parallel subagents?

This constraint is designed to maintain high code review quality and prevent architectural drift, ensuring that parallel changes remain manageable and integrated.

Plan Decomposition & Parallelization

Name: Plan Decomposition & Parallelization
Author: seanGSISG

byseanGSISG

0•

生産性とワークフロー

Decomposes monolithic implementation plans into granular task files and identifies opportunities for parallel execution.

This skill streamlines complex coding projects by breaking down large, monolithic implementation plans into smaller, focused task files. By analyzing file-based and explicit dependencies, it creates a structured execution manifest that identifies which tasks can safely run in parallel using up to two simultaneous subagents. This process significantly reduces context window usage for AI subagents by 90% and accelerates development timelines by up to 40% through efficient task orchestration.

主な機能

01Automatic decomposition of monolithic Markdown plans into individual task files

02Automated identification of parallel execution batches to optimize subagent usage

03Generation of standardized JSON execution manifests for subagent orchestration

04Drastic context token reduction by providing subagents with focused, bounded scopes

050 GitHub stars

06Intelligent dependency analysis based on shared files and explicit task mentions

ユースケース

01Optimizing AI subagent performance by minimizing the context tokens required for specific tasks

02Accelerating project velocity by identifying and running independent coding tasks simultaneously

03Scaling complex feature implementations that involve multiple files and distinct logic steps

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add seangsisg/crispy-claude decomposing-plans

For use in Claude.ai and ChatGPT

Download Skill