Recursive Language Model (RLM) Skill

"Context is an external resource, not a local variable."

This skill equips Claude Code (and compatible agents) with the Recursive Language Model (RLM) pattern described in the research paper: Recursive Language Modeling (ArXiv:2512.24601).

It enables the agent to handle massive codebases (100+ files, millions of lines) by treating the filesystem as a database and using parallel background agents to process information recursively, eliminating "context rot".

📦 Installation

Run this one-liner in your terminal:

curl -fsSL https://raw.githubusercontent.com/BowTiedSwan/rlm-skill/main/install.sh | bash

Auto-detects Claude Code and installs the skill.

🚀 Usage

Once installed, simply ask Claude to handle a large task:

"Use RLM to analyze the entire codebase for security vulnerabilities." "Scan all 500 files and find where UserID is defined."

The skill triggers automatically on keywords like:

• "analyze codebase"
• "scan all files"
• "large repository"
• "RLM"

🧠 How It Works

The skill operates in two distinct modes to eliminate "context rot":

1. Native Mode (Default): Optimized for Zero-Shot Filtering. It uses high-speed filesystem tools like grep and find for rapid codebase traversal and pattern discovery. Best for mapping project structure and locating specific definitions.
2. Strict Mode (Paper Implementation): Optimized for Dense Data Processing. It uses the rlm.py engine to perform Programmatic Slicing (Chunking). By loading data into memory and serving it in atomic chunks, it allows precise analysis of massive logs, monorepos, and CSVs that exceed standard context limits.

The Pipeline

1. Index: The agent scans your file structure using find or ls.
2. Filter: It uses grep / ripgrep to narrow down candidate files (Zero-Shot Filtering).
3. Map: It spawns multiple parallel background agents. Each sub-agent reads one file and answers one question.
4. Reduce: The main agent collects the structured outputs and synthesizes the final answer.

📜 Credits

• Research Paper: Recursive Language Modeling
• Skill Author: Bowtiedswan