cache-audit

What problem does it solve?

The cache-audit Skill detects prompt caching anti-patterns and dynamic-content placement that cause cache misses, unnecessary token usage, and fragile session behavior in Claude Code setups. It consolidates many manual checks into a single scored report so teams can quickly identify and fix the issues that most impact performance and cost.

Core Features & Use Cases

• Ordering checks: Verifies that static system prompts load before dynamic content and that CLAUDE.md files do not contain session-specific data.
• Hook and message validation: Ensures hooks emit dynamic data via additionalContext/system-reminder messages instead of editing system prompts.
• Tool, model, and size audits: Detects mid-session tool or model changes, checks MCP tool defer patterns, and estimates per-turn injected content size (git status, streaks, meeting syncs).
• Fork and compaction review: Reviews compaction and subagent handoffs to ensure consistent prefixes and tool schemas.
• Use case: Run during onboarding, CI checks, or debugging when sessions show frequent cache misses, unexpected behavior, or high token costs.

Quick Start

Run /cache-audit to analyze your Claude Code config and receive a scored report with prioritized fixes.