GitHub - drona23/claude-token-efficient: Universal CLAUDE.md - cut Claude output tokens by 63%. Drop-in. No code changes.

A drop-in CLAUDE.md configuration file that reduces Claude Code's output verbosity by approximately 63% by banning sycophantic openers, hollow closings, prompt restatements, unsolicited suggestions, and over-engineered code. The file is placed in the project root and read automatically by Claude Code. Benchmarks show ~384 output tokens saved per 4 prompts, though the file itself adds input tokens on every message, so net savings only apply at high output volume. Multiple profiles are available for coding, agent pipelines, and analysis workflows. The project also documents when this approach is not worth using, such as single short queries, casual use, or when guaranteed parseable output is needed.