auditing-instruction-hierarchy

Audits CLAUDE.md hierarchies for size, cache hygiene, and scope drift.

2|Updated May 23, 2026
One-click install
npx skills add https://github.com/rocklambros/rcs --skill auditing-instruction-hierarchy
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: auditing-instruction-hierarchy
Source: https://github.com/rocklambros/rcs/tree/main/skills/claude-code-meta/auditing-instruction-hierarchy
Command: npx skills add https://github.com/rocklambros/rcs --skill auditing-instruction-hierarchy

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill helps you audit a Claude Code instruction hierarchy when CLAUDE.md files start to feel bloated, slow, or mis-scoped. It identifies where rules belong, what should be removed, and what should move into a skill, hook, or system reminder.

Core Features & Use Cases

  • Size budgeting: Measures the total instruction footprint across user, project, and plugin layers against the 250-line target and 400-line hard cap.
  • Cache hygiene: Flags timestamps, run IDs, and other dynamic content that would break prompt-cache reuse.
  • Scope triage: Separates always-on guidance from project-only rules and workflow logic that should live elsewhere.
  • Prioritized trimming: Produces a keep, move, or drop list so the hierarchy becomes smaller and more stable.
  • Use case: Audit a growing CLAUDE.md setup before adding another instruction file, or after noticing higher token usage at session start.

Quick Start

Ask for an audit of your CLAUDE.md hierarchy and have the Skill measure size, flag cache-breaking lines, and recommend what to keep or move.

Frequently Asked Questions about auditing-instruction-hierarchy

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I audit my CLAUDE.md instruction hierarchy for bloat and cache issues?

Auditing a CLAUDE.md instruction hierarchy involves measuring per-file line counts against a 250-line target, detecting dynamic content that breaks prompt caching, and checking hierarchy precedence across user, project, and plugin layers to produce a prioritized keep, move, or drop trim list.

Why does my prompt cache hit rate drop when loading Claude Code instruction files?

Prompt cache hit rates drop when CLAUDE.md files contain dynamic content like timestamps or run IDs. Auditing these instruction files identifies and flags cache-breaking lines so you can remove or relocate them to stabilize the prompt cache.

What is the line count limit for CLAUDE.md files in Claude Code?

The line count limit for CLAUDE.md files targets 250 lines, with a 400-line hard cap. Auditing measures the total instruction footprint across user, project, and plugin layers to ensure files stay within these size budgets.

When should I move project rules out of my CLAUDE.md files?

You should move project rules out of CLAUDE.md files when scope triage identifies workflow logic or project-only rules that belong in a skill, hook, or system reminder rather than always-on guidance, ensuring the instruction hierarchy remains lean.

Does Claude Code support checking hierarchy precedence across multiple CLAUDE.md instruction files?

Claude Code supports checking hierarchy precedence across user, project, and plugin CLAUDE.md instruction files. An audit verifies that rules are correctly scoped and separated into always-on guidance versus project-specific configurations.

What is the best way to reduce token usage at session start when using Claude Code?

The best way to reduce token usage at session start is to audit your CLAUDE.md instruction hierarchy for bloat, trim dynamic content that breaks prompt caching, and move workflow logic into skills or hooks to shrink the total instruction footprint.