Can this skill handle large non-code documents like PDFs?

Yes. It includes specific protocols for large documents, focusing on Table of Contents and metadata first, then splitting content into logical chunks for independent processing by sub-agents.

Does it improve Claude's autonomous behavior?

Absolutely. The skill encourages 'Agentic Behavior' by empowering Claude to self-correct, plan multi-step searches, and manage its own context window without requiring user permission for every sub-step.

How does this skill prevent AI 'context rot'?

It treats inputs as external environmental variables. Instead of loading everything into memory, it uses targeted tools like grep and line-specific file viewing to process data in small, high-focus windows.

When should I use the recursive-decomposition skill?

You should apply this skill when tasks involve more than 10 files, contexts exceeding 50,000 tokens, or complex multi-hop questions that require gathering evidence from many scattered sources.

Recursive Decomposition

Name: Recursive Decomposition
Author: massimodeluisa

bymassimodeluisa

•

Productivity & Workflow

Manages large-scale coding tasks and massive documents using recursive strategies to bypass context limits and prevent hallucinations.

Based on Recursive Language Models (RLM) research (2025), this skill enables Claude to handle tasks exceeding standard context windows by programmatically decomposing complex problems into manageable sub-tasks. It treats files as environmental variables rather than direct input, utilizing progressive disclosure and recursive self-invocation to analyze entire codebases, aggregate information across dozens of documents, and perform multi-hop queries without losing accuracy or suffering from 'context rot.' This approach ensures high-fidelity results for tasks involving 10+ files or contexts exceeding 50,000 tokens.

Key Features

01Smart chunking and batch processing for lists exceeding 10 items

02Recursive problem decomposition for tasks exceeding 50k tokens

0318 GitHub stars

04Evidence-based synthesized answer verification to ensure high-fidelity results

05Autonomous sub-agent orchestration for parallelized information gathering

06Programmatic codebase-wide searching using grep and glob optimization

Use Cases

01Mapping architectural patterns and error handling across a multi-repo codebase

02Extracting and summarizing feature requirements from dozens of PRDs or technical documents

03Large-scale code migrations and technical debt audits involving hundreds of source files

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add massimodeluisa/recursive-decomposition-skill recursive-decomposition

For use in Claude.ai and ChatGPT

Download Skill