What problem does it solve? Claude Code sessions accumulate silent token bloat from proliferating rules files, skill descriptions, and always-on context layers, inflating costs and degrading prompt cache efficiency without any systematic way to audit or control it. ## Core Features & Use Cases - Eleven-Category Reduction Framework (A~K): Binding standards covering session static tax, workflow compression, memory DB strategy, RAG pre-injection, pipeline orchestration, and skill DMI policies. - Budget Ceilings & Regression Monitoring: Enforces a 6,000-token always-on ceiling with an audit checklist, anti-pattern catalog, and regression actions when limits are exceeded. - Prompt Caching Economics & Decision Tree: Codifies cache write/read cost rules and provides a decision tree for routing new content to rules, skills, or the memory DB. - Use Case: When your project's always-on context grows from 4,910 to 19,090 tokens after adding rules and skills, run the audit mode to identify overlapping rule files, skills missing DMI flags, and merge candidates with quantified savings. ## Quick Start Ask the AI to run a token budget audit of the current Claude Code configuration using the claude-token-decrease checklist and report which rules and skills exceed the ceilings.
