← 返回 Skills 市场
largetool

4d Compression Core

作者 largetool · GitHub ↗ · v1.0.3 · MIT-0
cross-platform ⚠ suspicious
530
总下载
0
收藏
2
当前安装
4
版本数
在 OpenClaw 中安装
/install 4d-compression-core
功能描述
把长内容压缩成结构化 4D 向量——节省约 60-80% Token,保留核心信息
使用说明 (SKILL.md)

🌀 4D Compression Core - 4D 向量压缩

把长内容压缩成结构化 4D 向量——节省约 60-80% Token,保留核心信息

全部本地运行 · 无隐私收集 · 危险操作二次确认 · 可随时卸载


🎯 核心功能

  • 🧠 结构化压缩 - 空间/能量/信息/时间四维分离
  • 🎯 语义保留 - 保留核心语义,压缩率约 60-80%
  • 📊 智能路由 - 自动选择最优压缩版本(A/B/C)
  • 零额外消耗 - 规则匹配,不调用大模型

🚀 快速开始

你说:"压缩这段内容"

Neo 显示:
🌀 4D 压缩完成

原始:12,345 tokens
压缩后:3,456 tokens
节省:约 72%

【结构化要点】核心信息...

💡 保留核心语义,可安全使用

🎯 触发词

你想做什么 你说
压缩文本 "压缩"、"4d"、"精简"、"compress"

不用配置,说"压缩"就启动。


⚠️ 安全与边界

压缩率说明

  • 文本:约 60-80%(因内容而异)
  • 知识类:约 70-80%(深度无损)
  • 对话类:约 40-60%(保留情感)

语义保留

  • 压缩版:保留核心语义(97%+)
  • 不支持:EPUB/PDF 书籍处理(需手动转换文本)

本地运行

  • 所有压缩在本地完成
  • 无云端依赖,无隐私泄露

可随时卸载

  • clawhub uninstall 4d-compression-core
  • 无残留,无影响

📦 安装

clawhub install 4d-compression-core

🔗 可验证证据


版本:1.0.3
创建时间:2026-02-24
许可证:MIT
开发者:Neo(宇宙神经系统)

安全使用建议
This skill claims sophisticated local compression but contains no implementation artifacts — treat it as incomplete or unverified. Before installing or using it, ask the developer for: (1) the actual implementation (scripts, binaries, or deterministic algorithm) showing how compression is done locally; (2) example inputs and outputs (same input → compressed + decompressed verification) and unit tests proving the claimed semantic retention; (3) a network-activity audit or assertion (and ideally a short network-capture run) proving no external calls are made; (4) clarification about trigger behavior (disable automatic trigger words or require explicit user consent before processing); and (5) removal of developer-local paths or other leftover artifacts. Until you get verifiable code/tests and a declaration of no network I/O, avoid running this on sensitive data and consider testing in an isolated sandbox with non-sensitive examples.
功能分析
Type: OpenClaw Skill Name: 4d-compression-core Version: 1.0.3 The skill bundle '4d-compression-core' (v1.0.3) consists entirely of metadata and documentation (SKILL.md, VERSION_PROTOCOL.md) defining a logic for the AI agent to summarize and compress text. There is no executable code, scripts, or evidence of malicious prompt injection; the instructions are strictly aligned with the stated purpose of token optimization and semantic retention.
能力评估
Purpose & Capability
The skill claims a nontrivial local compression algorithm (three-version A/B/C selection, semantic-preserving 4D vectors, specific retention rates) yet the package is instruction-only with no code, no scripts, and no described binaries beyond 'bash'. It's unclear how the agent will actually perform the compression locally or whether it will rely on the platform LLM. The single required binary 'bash' is insufficient evidence for the claimed capabilities.
Instruction Scope
SKILL.md is high-level and prescriptive but lacks concrete runtime commands or file/path interactions. It also advertises trigger-word auto-start ('说"压缩"就启动'), which could cause unexpected invocation; '危险操作二次确认' is promised but not defined. VERSION_PROTOCOL.md includes a developer-local filesystem path (/Users/abc/...), indicating leftover dev artifacts. The instructions do not explicitly direct reading unrelated system files or env vars, but the absence of implementation details leaves scope unclear.
Install Mechanism
No install spec and no code files — instruction-only — so nothing will be downloaded or installed by the skill bundle itself. This is the lowest install risk, but it increases the importance of understanding how the agent will implement the described behavior.
Credentials
The skill declares no required environment variables, credentials, or config paths. That is proportional to an instruction-only skill; there are no apparent demands for unrelated secrets or cloud credentials.
Persistence & Privilege
Skill flags are default (always:false, user-invocable:true). It does not request persistent or elevated privileges. Autonomous invocation is allowed by platform default but not unusual here.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install 4d-compression-core
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /4d-compression-core 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.3
4d-compression-core v1.0.3 - 更新文档说明,突出全部本地运行、无隐私收集、支持零额外消耗、不再支持 EPUB/PDF 直接处理 - 核心功能新增“结构化 4D 向量”、智能路由、多压缩版本选择与本地规则匹配 - 明确压缩率范围和适用类型,新增对对话类、知识类文本压缩率说明 - 简化和更新触发词与安装/卸载说明 - 增加可验证证据入口和测试报告链接
v1.0.2
外循环验证版 - 自动化发布脚本就绪
v1.0.1
- Simplified and streamlined the SKILL.md documentation for easier reading. - Updated feature highlights and usage instructions for clarity and brevity. - Added version number and self-review note at the top. - Focused on main use cases, quick start examples, and essential trigger words. - Provided clear compression rate expectations and outlined security boundaries. - Removed in-depth technical and implementation details for a more concise overview.
v1.0.0
4d-compression-core v1.0.0 release: - Implements a 4D vector-based token compression engine using the UPTEF framework. - Compresses lengthy content into structured 4D vectors, saving 60-80% tokens while retaining core information and semantics. - Supports automatic and manual triggers for text, EPUB, PDF, and Markdown inputs. - Provides dual output formats (JSON & Markdown) for both concise and full reports. - Features real-time token-saving statistics, quality evaluation, and UPTEF force logging. - Integrates with Token Water Meter for smart token usage monitoring and guidance.
元数据
Slug 4d-compression-core
版本 1.0.3
许可证 MIT-0
累计安装 2
当前安装数 2
历史版本数 4
常见问题

4d Compression Core 是什么?

把长内容压缩成结构化 4D 向量——节省约 60-80% Token,保留核心信息. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 530 次。

如何安装 4d Compression Core?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install 4d-compression-core」即可一键安装,无需额外配置。

4d Compression Core 是免费的吗?

是的,4d Compression Core 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

4d Compression Core 支持哪些平台?

4d Compression Core 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 4d Compression Core?

由 largetool(@largetool)开发并维护,当前版本 v1.0.3。

💬 留言讨论