What is the 'Escalation Ladder'?

It is a strategy for efficiency: Level 1 uses Python for structural tasks, Level 2 uses LLM queries within the REPL for judgment, and Level 3 uses subagents for final synthesis.

Does RLM support parallel processing?

Yes, RLM includes an llm_query_batch function that allows for parallel processing of multiple file chunks or prompts to speed up analysis.

What file sizes can RLM handle?

RLM is optimized for files exceeding 100KB or 2000 lines, efficiently handling multi-megabyte datasets that would typically exceed standard LLM context windows.

Can I use RLM for code analysis?

Yes, it is ideal for exploring large repositories. You can use Level 1 REPL analysis for structural searches and Level 2 or 3 for semantic reasoning about code logic.

How does RLM save context tokens?

It keeps the full file content in a Python REPL environment. Only your specific Python print outputs or synthesized summaries are returned to the main Claude context, keeping the chat history lean.

RLM: Recursive Large File Processor

Name: RLM: Recursive Large File Processor
Author: Whamp

byWhamp

•

生産性とワークフロー

Processes massive files and codebases that exceed standard LLM context windows by utilizing a REPL-based recursive analysis system.

RLM (Recursive Language Model) is a specialized skill designed to handle files larger than 100KB or 2000 lines, such as extensive logs, long transcripts, or entire code repositories. By keeping large content within a Python REPL environment rather than the active LLM context, it allows for efficient structural analysis, semantic querying via subprocesses, and final synthesis through subagents. This architecture effectively bypasses token limits while maintaining high reasoning accuracy, enabling Claude to navigate and summarize data that would otherwise be impossible to process in a single session.

主な機能

01REPL-based memory management to handle multi-megabyte files without context overflow

0213 GitHub stars

03Context protection budget to prevent token bloat during complex synthesis tasks

04Three-level escalation ladder from structural Python analysis to semantic LLM reasoning

05Parallel batch processing for chunked file analysis using isolated subagents

06Built-in utilities for regex searching, JSON parsing, and content aggregation

ユースケース

01Performing structural and semantic audits across entire software repositories

02Summarizing and querying lengthy legal documents, transcripts, or research papers

03Analyzing massive application logs to identify and categorize intermittent system errors

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add whamp/pi-rlm rlm

For use in Claude.ai and ChatGPT

Download Skill