# claude-token-efficient

> A single CLAUDE.md file you drop into your project root to reduce Claude's output verbosity and cut token usage on output-heavy workflows.

claude-token-efficient is a single-file configuration tool that reduces Claude's default verbosity by dropping a `CLAUDE.md` file into your project root. Claude Code reads it automatically, eliminating sycophantic preambles, unsolicited suggestions, and over-engineered responses. Benchmarks show up to 63% reduction in output word count with no signal loss, making it especially useful for automation pipelines and repeated structured tasks.

- **Zero-setup drop-in**: *Copy `CLAUDE.md` to your project root via `curl` or `git clone` — no code changes, no configuration required.*
- **Eliminates sycophancy**: *Removes "Sure!", "Great question!", and closing fluff that adds tokens without value.*
- **Targeted edits over rewrites**: *Instructs Claude to prefer minimal, surgical code changes rather than rewriting entire files.*
- **Multiple profiles**: *Choose from `CLAUDE.coding.md`, `CLAUDE.agents.md`, `CLAUDE.analysis.md`, and `CLAUDE.benchmark.md` for different workflows.*
- **Versioned config sets**: *Three optimization strategies (v5 multi-file structured, v6 one-shot execution, v8 ultra-lean) let you tune cost vs. capability.*
- **Composable rules**: *Works with Claude's global (`~/.claude/CLAUDE.md`), project-level, and subdirectory-level file hierarchy so rules stay close to where they apply.*
- **Override-friendly**: *User instructions always win — ask for verbose output any time and the file won't fight you.*
- **Model-agnostic rules**: *Written as plain instructions; should work on any model that reads context, though benchmarks were run on Claude.*
- **Community-driven**: *Open to PRs and issue submissions; new fix rules are versioned and credited to contributors.*
- **Honest trade-offs documented**: *README explicitly states when the file costs more than it saves (low-volume, single-query, or fresh-session workflows).*

## Features
- Single CLAUDE.md file, no code changes required
- Eliminates sycophantic preambles and closing fluff
- Reduces output verbosity by up to 63% in benchmarks
- Multiple task-specific profiles (coding, agents, analysis, benchmark)
- Versioned configuration sets (v5, v6, v8) for different cost/capability trade-offs
- Composable with global, project-level, and subdirectory CLAUDE.md files
- User instructions always override file rules
- Model-agnostic prompt rules
- Community-contributed fix rules with versioning and credit

## Integrations
Claude Code, Anthropic Claude, llama.cpp, Mistral

## Platforms
WINDOWS, API, CLI

## Pricing
Open Source

## Version
M-drona23-v8

## Links
- Website: https://github.com/drona23/claude-token-efficient
- Documentation: https://github.com/drona23/claude-token-efficient/blob/main/BENCHMARK.md
- Repository: https://github.com/drona23/claude-token-efficient
- EveryDev.ai: https://www.everydev.ai/tools/claude-token-efficient