ContextSlim — Slash 70k Token Waste: AI Agent That Trims, Summarizes, and Streams Project Context

The poster reports that SuperClaude appears to consume about 70k tokens from .md and .yml files used to initialize Claude, causing the model's available context to drop to roughly 30% early in a project. They argue that if this is how Claude loads files, SuperClaude offers little benefit and may be wasting tokens and cost.