Overview
dsh-headroom
Headroom-inspired automatic context compression plugin for DeepSeek Harness (dsh): compresses tool outputs before they reach the model and keeps every lossy compression reversible via CCR retrieval tools.
README / EN
Package documentation
Registry summary
Headroom-inspired automatic context compression plugin for DeepSeek Harness (dsh): compresses tool outputs before they reach the model and keeps every lossy compression reversible via CCR retrieval tools.
dsh.pub verifies the pinned bundle contract, runtime facts, and distribution semantics. The complete README remains in the source repository.
Read the full README on GitHubLIMITATIONS
Known limitations
- **默认评分器是启发式模拟，不是真实模型推理**：`lib/kompress.js` 忠实移植了 Headroom Kompress-v2-base 的管线结构与评分公式，但默认打分来自确定性纯 JS 启发式； 需要真实模型语义打分时，可库级注入 ONNX/PyTorch scorer。 - **Kompress 输出是保留词碎片**：为追求高压缩率，默认阈值会删除大量普通词，输出可读性 有限，适合给模型看要点；精确原文始终可通过 `headroom_retrieve` 逐字节取回。 - **保守不压缩场景**：全部由 must-keep 类词组成的文本、纯重复中文文本会直接走 head/tail 回退或原样返回，防止把内容删光。 - **模板文本不去重**：高频重复的数字/标识符会逐个保留（must-keep 语义），模板化输出 可能保留较多重复事实 token，压缩率低于普通散文。