What is the three-tier memory strategy?

It is a framework that divides context into Tier 1 (minimal global info in CLAUDE.md), Tier 2 (task-specific priming commands), and Tier 3 (just-in-time file reads during execution).

What are priming commands?

Priming commands are specific task-oriented instructions (like /prime-bug) that load relevant files and context only when a specific type of work begins.

How does this skill improve Claude's accuracy?

By reducing 'noise' and irrelevant background information, it ensures Claude's attention is focused strictly on the files and rules pertinent to the current task.

Why should CLAUDE.md be kept under 2KB?

Keeping the primary context file small ensures that core project rules are always loaded without consuming excessive tokens or crowding the model's active memory.

Context Hierarchy Design

Name: Context Hierarchy Design
Author: melodic-software

bymelodic-software

•

生産性とワークフロー

Optimizes Claude's memory by implementing a progressive, three-tier context loading strategy for efficient codebase management.

概要

Context Hierarchy Design is a specialized skill for Claude Code designed to eliminate 'context bloat' and improve agent performance in large repositories. It guides developers through organizing project metadata into a structured three-tier system: a minimal permanent layer (CLAUDE.md), task-specific 'priming' commands for targeted context loading, and on-demand file reads for specific implementations. By implementing this tiered approach, the skill reduces token consumption, minimizes model confusion, and ensures Claude has exactly the right information for every task without overwhelming its context window.

主な機能

Custom priming command design for specialized workflows like bug fixing and feature development
Context auditing framework to categorize project information by necessity
CLAUDE.md optimization to maintain global context under 2KB
1 GitHub stars
Progressive disclosure patterns to scale context for massive codebases
Three-tier memory strategy implementation (Global, Task-Specific, On-Demand)

ユースケース

Optimizing agent response times and accuracy in large-scale enterprise projects
Refactoring bloated CLAUDE.md files that cause AI performance degradation
Designing specialized context layers for complex, multi-directory monorepos

概要

主な機能

Custom priming command design for specialized workflows like bug fixing and feature development
Context auditing framework to categorize project information by necessity
CLAUDE.md optimization to maintain global context under 2KB
1 GitHub stars
Progressive disclosure patterns to scale context for massive codebases
Three-tier memory strategy implementation (Global, Task-Specific, On-Demand)

ユースケース

Optimizing agent response times and accuracy in large-scale enterprise projects
Refactoring bloated CLAUDE.md files that cause AI performance degradation
Designing specialized context layers for complex, multi-directory monorepos