Claude Skill
win4r/lossless-claw-enhanced
lossless-claw-enhanced 是 lossless-claw 的分支,增加了中日韩文字令牌估算,提升 Claude Skill 和 OpenClaw 工作流中中文、日文和韩文文本的处理精度。
概览
仓库信息
安装这个 Skill
git clone https://github.com/win4r/lossless-claw-enhanced.gitRegistry 信息
git clone https://github.com/win4r/lossless-claw-enhanced.gitnpm installnpx vitest run --dir testnpx vitest run test/estimate-tokens.test.ts
项目简介
基于 Martian-Engineering/lossless-claw 的分支,增加了对中日韩文字(CJK)的令牌估算支持,提升 Claude Skill 工作流中中文、日文和韩文文本的处理精度。
Fork of Martian-Engineering/lossless-claw with CJK-aware token estimation
要点
- 支持中日韩文字的令牌估算,提升文本处理精度
- 基于 lossless-claw 的分支,增强语言支持
- 兼容 OpenClaw 插件和技能
- 基于 TypeScript 实现,确保可靠性
使用场景
- 提升 Claude Skill 对中日韩文字的处理精度
- 多语言 AI 工作流中的令牌估算
- 增强 OpenClaw 插件对亚洲语言的性能
README 摘要
# 🦞lossless-claw-enhanced > [!WARNING] > **DEPRECATED — please use upstream instead** > > The two original differentiators of this fork — **CJK-aware token estimation** and **CJK trigram FTS search** — have both been merged into upstream [`Martian-Engineering/lossless-claw`](https://github.com/Martian-Engineering/lossless-claw) ([PR #344](https://github.com/Martian-Engineering/lossless-claw/pull/344), [PR #219](https://github.com/Martian-Engineering/lossless-claw/pull/219)). As of upstream `v0.11.x` there is no functional gap between this fork and upstream. > > This fork is at `0.5.3` and last adapted to OpenClaw `2026.4.14` — about 6 minor versions behind upstream. It will not be updated further. > > - **New users**: use [`Martian-Engineering/lossless-claw`](https://github.com/Martian-Engineering/lossless-claw) directly. It builds clean against current OpenClaw and includes all the CJK fixes this fork pioneered. > - **Existing users on OpenClaw 2026.5.x+**: switch to upstream. This fork no longer compiles against current OpenClaw (see [#32](https://github.com/win4r/lossless-claw-enhanced/issues/32)). Lossless Context Management plugin for [OpenClaw](https://github.com/openclaw/openclaw), based on the [LCM paper](https://papers.voltropy.com/LCM) from [Voltropy](https://x.com/Voltropy). Replaces OpenClaw's built-in sliding-window compaction with a DAG-based summarization system that preserves every message while keeping active context within model token limits. 中文配置文档: [docs/configuration.zh-CN.md](docs/configuration.zh-CN.md) ## Video Tutorial > Full walkthrough: installation, configuration, and hybrid retrieval internals. [](https://youtu.be/m21PNaIW3N4) **https