One CLAUDE.md file. Keeps Claude responses terse. Reduces output verbosity on heavy workflows. Drop-in, no code changes.
Browse cluster: Cluster 125 →Claude-token-efficient is a single-file instruction template designed to reduce token consumption when working with Claude by keeping responses concise and eliminating verbose default behaviors. The project addresses the problem domain of token efficiency in AI workflows, particularly for automation pipelines and repeated structured tasks where Claude's default verbosity compounds across many calls. The repository includes a base CLAUDE.md file containing rules for reading files before writing, avoiding sycophantic language, skipping large files, and testing before finishing. It also provides a benchmark harness written in Python 3 that measures token cost and semantic correctness through reproducible testing against Claude's local CLI. The material shows token reduction benchmarks ranging from 4-12% on actual model outputs (haiku, sonnet, opus) and profiles tailored to different workflows: multi-file structured approaches, one-shot execution, and ultra-lean configurations. The project was built by drona23 and accepts community contributions targeting specific failure modes in Claude's default behavior.