Why are style and linting rules removed during optimization?

Claude learns coding styles better through in-context examples in your existing code. Offloading formatting to tools like Prettier or Biome via 'Stop Hooks' is more effective than text-based instructions.

What happens to the detailed information removed from CLAUDE.md?

This skill uses 'Progressive Disclosure' to move detailed API patterns, architecture deep-dives, and version histories into a structured agent_docs/ directory, which the model can still reference when needed.

Why should I limit my CLAUDE.md to 150 lines?

Research indicates that as instruction count increases, an LLM's ability to comply with any single instruction drops uniformly. Keeping the file lean ensures the most important rules are actually followed.

When is the best time to use this skill?

It is best applied immediately after running the /init command, as /init often generates verbose content that can clutter the model's primary context window.

CLAUDE.md Optimization

Name: CLAUDE.md Optimization
Author: tianzecn

bytianzecn

内容管理

Refactors and prunes project context files to maximize model compliance and minimize token waste.

关于

This skill automates the optimization of CLAUDE.md files based on research-backed best practices to prevent instruction degradation. By applying the principle of progressive disclosure, it relocates verbose technical specifications into a dedicated agent_docs directory while maintaining a lean, high-leverage core instruction set (60-150 lines). This ensures Claude Code remains focused on critical project rules and architecture, improving the reliability of implementation tasks and reducing token overhead in every conversation.

主要功能

Organizes project context into structured WHAT, WHY, and HOW dimensions
Implements 'Progressive Disclosure' by migrating detailed specs to an agent_docs/ folder
Prioritizes critical rules at the top to exploit LLM attention distribution biases
Prunes instruction count to the high-compliance sweet spot of 60-150 lines
0 GitHub stars
Identifies and removes redundant style/linting rules better handled by external tools

使用场景

Refactoring verbose /init output into a focused, high-leverage project context
Improving rule compliance in projects where the AI has started ignoring instructions
Organizing complex monorepo documentation for better agent navigation

关于

主要功能

Organizes project context into structured WHAT, WHY, and HOW dimensions
Implements 'Progressive Disclosure' by migrating detailed specs to an agent_docs/ folder
Prioritizes critical rules at the top to exploit LLM attention distribution biases
Prunes instruction count to the high-compliance sweet spot of 60-150 lines
0 GitHub stars
Identifies and removes redundant style/linting rules better handled by external tools

使用场景

Refactoring verbose /init output into a focused, high-leverage project context
Improving rule compliance in projects where the AI has started ignoring instructions
Organizing complex monorepo documentation for better agent navigation