Text Optimizer

This applies 41 optimization rules across six categories to cut token counts by 20 to 40 percent in prompts and documentation without losing meaning. You get three modes: light for basic cleanup, medium for balanced restructuring, and deep for maximum compression. The rules cover Claude-specific behavior patterns (like avoiding "think" phrasing and literal instruction following), token efficiency tricks (tables beat prose, bullets beat numbered lists), and structural improvements using XML tags and progressive disclosure. It checks reference integrity and fixes broken file paths as it goes. Useful when you're hitting context limits, paying per token, or just trying to make agent instructions clearer and less prone to hallucinations.