← 返回 Skills 市场
skill-compressor 精简技能*减少token
作者
David0Ming
· GitHub ↗
· v1.0.0
· MIT-0
92
总下载
1
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install skill-compressor
功能描述
压缩 skill 降 token 成本。用户说 /skill-compressor 或要求优化/压缩/瘦身 skill 时触发。
安全使用建议
This skill appears to do what it says: it reads a skill's SKILL.md and reference files, classifies and compresses content, then writes a compressed copy to <skill>/.reduced/. Before using it: (1) verify the skill source you install from (README suggests a GitHub repo but the registry entry lacks an install spec—confirm the origin); (2) only run the compressor on skill directories that don't contain secrets, private keys, or production credentials, because it will read all referenced files; (3) review the generated .reduced/ output and REDUCTION_REPORT.md before replacing any originals; (4) if Gate 2 runtime validation is enabled, confirm what commands/tests will be executed (to avoid unintended code execution). If you need higher assurance, request the author's repository and review the tool's implementation before granting it filesystem access.
功能分析
Type: OpenClaw Skill
Name: skill-compressor
Version: 1.0.0
The skill-compressor is a utility designed to optimize OpenClaw skills by reducing token consumption through text compression and content categorization. The logic in SKILL.md and the supporting documentation (background.md, examples.md) focuses entirely on text processing, classification (core rules vs. examples/background), and file organization. It includes explicit safety constraints to ensure that critical operational data—such as environment variables, API names, and file paths—are never deleted or modified during the compression process. The skill operates non-destructively by writing output to a '.reduced/' subdirectory rather than overwriting original files, and it shows no signs of data exfiltration, malicious execution, or harmful prompt injection.
能力评估
Purpose & Capability
The name/description claim to compress other skills to reduce token cost, and the runtime instructions explicitly read a skill's SKILL.md and its reference files and write a compressed copy into <skill>/.reduced/. That file I/O is exactly what a compressor needs. There are no unrelated env vars or binaries requested.
Instruction Scope
Instructions direct the agent to read the target skill's description, body, and all reference files, classify and split content, run faithfulness checks, and write outputs under the target skill's .reduced/ directory. That scope is expected for this task, but it means the skill will read arbitrary files under whatever skill path you point it at (which may contain secrets or sensitive configuration). Gate 2 mentions optional runtime validation—how that validation is run is not specified, so actual execution/side effects during validation could occur if the agent is allowed to run testcases; the spec leaves that behavior underspecified.
Install Mechanism
This is instruction-only with no install spec and no binaries to install, which is the lowest-risk model. README suggests optional manual git clone or npx/clawhub install commands, but the registry entry contains no install block; that discrepancy is informational rather than an execution risk in the package itself.
Credentials
The skill requests no environment variables or credentials. The instructions explicitly preserve API names/env var names from the original skill rather than reading other secrets. However, because it reads arbitrary files in the specified skill folder, it can encounter any environment-like strings stored there—so you should avoid running it on skill directories that contain private keys or tokens unless you trust the environment and the skill's author.
Persistence & Privilege
Flags are default: always:false and user-invocable:true. The skill writes output into the target skill's .reduced/ directory but does not claim to modify other skills or system-wide configuration. That write-to-target behavior is consistent with its purpose and scoped to the target skill.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install skill-compressor - 安装完成后，直接呼叫该 Skill 的名称或使用
/skill-compressor触发 - 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
v1.0.0
Initial release of skill-compressor.
- Introduces a step-by-step process to compress skills and reduce token usage.
- Supports user prompts for compression via trigger phrases or feedback about token size.
- Details specific procedures for analyzing, labeling, and condensing skill descriptions and bodies.
- Ensures operational details (numbers, thresholds, paths, APIs, env variables) are preserved.
- Outputs all compressed files to a .reduced directory, keeping originals unchanged.
- Provides clear criteria for when compression should not be performed.
元数据
常见问题
skill-compressor 精简技能*减少token 是什么？
压缩 skill 降 token 成本。用户说 /skill-compressor 或要求优化/压缩/瘦身 skill 时触发。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 92 次。
如何安装 skill-compressor 精简技能*减少token？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install skill-compressor」即可一键安装，无需额外配置。
skill-compressor 精简技能*减少token 是免费的吗？
是的，skill-compressor 精简技能*减少token 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
skill-compressor 精简技能*减少token 支持哪些平台？
skill-compressor 精简技能*减少token 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 skill-compressor 精简技能*减少token？
由 David0Ming（@david0ming）开发并维护，当前版本 v1.0.0。
推荐 Skills