massive-context-mcp: Skill for Claude Code

RLM (Recursive Language Model) Skill

Overview

The RLM pattern enables processing massive contexts (10M+ tokens) that exceed Claude's context window by recursively chunking, processing, and aggregating results. Instead of failing on large files, use RLM to break them into manageable pieces.

When to Use RLM

Use RLM when you encounter:

• Large files: Any file >100KB or >2000 lines
• Multi-file analysis: Processing multiple files together (combined size matters)
• Context exceeded: User asks to analyze content that won't fit in context window
• Aggregation tasks: Summarizing logs, finding patterns across large datasets, counting/filtering operations
• Deep codebase analysis: Understanding architecture across many files
• Document processing: Analyzing reports, research papers, documentation sets

Don't use RLM for:

• Small files (<100KB)
• Single-pass tasks that fit in context
• Interactive editing (use standard tools)

The RLM Pattern

The core workflow is: Load → Inspect → Chunk → Sub-Query → Aggregate

Step 1: Load Context

# Load large content into RLM memory
rlm_load_context(
  name="codebase",
  content=file_contents  # Full file content
)

Returns: {name, size_bytes, size_chars, line_count, loaded: true}

Step 2: Inspect Context

# Understand structure without loading into prompt
rlm_inspect_context(
  name="codebase",
  preview_chars=500  # Optional preview
)

Returns: Metadata + preview (first N chars)

Step 3: Chunk Context

# Break into manageable pieces
rlm_chunk_context(
  name="codebase",
  strategy="lines",  # or "chars" or "paragraphs"
  size=100  # Lines per chunk (or chars if strategy=chars)
)

Chunking Strategies:

• lines (default): Split by line count - best for code, logs, structured data
• chars: Split by character count - best for prose, unstructured text
• paragraphs: Split by blank lines - best for documents, markdown

Returns: {name, chunk_count, strategy, size_per_chunk}