独立运营 · 不出售排名位置  · 我们怎么评估工具 →
Anclum 提交收录
R

RusticStack/skills

本站为独立目录,非该工具官网;下方信息取自其公开仓库。

Agent skills that prune what a repository doesn't need: stale comments, dead docs and bloated agent instruction files. (原文为英文)

装之前:命令归属未核实,仅作参考展示 · 许可证 MIT

第一步(仅参考,归属未核实):npx skills add rusticstack/skills

更新 2026-11-18
在 GitHub 查看 ↗直链 · 无中转页

装完之后怎么算装好了

  1. 装完该看到什么 — 命令正常结束时退出码为 0,终端里没有 "command not found" 或 "permission denied"。
  2. 一行自检 — 按该仓库 README 里给出的最小用法跑一次;能按 README 的描述输出就是通了。
  3. 答不出来怎么办 — 这个项目没有写排错说明,去仓库的 Issues 搜 install。

安装命令取自该仓库;以上判据只描述可观察到的结果,不替代项目自己的文档。

★ 数取值
0 ★ · RusticStack/skills · 取自 GitHub 搜索 API · (2026-11-19 20:00:48 核对)
来源明细
从该仓库的 GitHub 页面直接读取 · 条目类型:开源项目
收录与活跃
收录于 2026-11-19 · 最近见到 2026-11-20 · 共见到 2 次 · 被 4 个来源列出
评分构成
综合分 403 · 4 个来源 · repo
安装
安装命令见上方「安装与配置」。
常见问题(答案都来自我们的记录)
它适合谁 / 用来做什么?
我们把它归在这些标签下: Agent 技能。标签来自条目自身,不是我们强加的分类。
谁维护它、还在更新吗?
仓库 RusticStack/skills;我们收录于 2026-11-19;最近见到 2026-11-20。这里没有"官方认证"或"质量评级"——我们只记录来源与时间。
README 原文(英文,未翻译) · 在 GitHub 读完整 README →

RusticStack Skills

以下摘录自该仓库的 README(片段)

Agent skills that leave a repository easier to read.
Short instruction files for Claude Code, Codex, Cursor and other coding agents. Each one removes or fixes something that costs the next reader, human or model.
## Skills
| Skill | What it does | Install in Claude Code |
| prune-docs-and-comments | Prunes comments and docs that say nothing the code doesn't, or say something no longer true. | /plugin install prune-docs-and-comments@rusticstack-skills |
| prune-tests | Prunes tests that guard no behavior anyone relies on, with proof for every deletion. | /plugin install prune-tests@rusticstack-skills |
| clarify-code | Makes code say what a reader otherwise has to work out, without changing what it does. | /plugin install clarify-code@rusticstack-skills |
| prune-dead-code | Prunes code that nothing runs and nothing reaches, with proof for every deletion. | /plugin install prune-dead-code@rusticstack-skills |
| prune-dependencies | Prunes dependencies the project does not use, with a clean install and build as proof. | /plugin install prune-dependencies@rusticstack-skills |
## Install
With the skills CLI, which works with Claude Code, Cursor, Codex and other agents:
npx skills add rusticstack/skills
Or, in Claude Code, add this repository as a plugin marketplace, then install a skill with the command in its row of the table above:
/plugin marketplace add rusticstack/skills
To install by hand, copy a skill's folder into ~/.claude/skills/.
## prune-docs-and-comments
Run it on a repository or a path:
/prune-docs-and-comments
/prune-docs-and-comments src/billing
The agent asks one question of every comment and document: what does it depend on? Prose stays only if it depends on intent the code cannot express, such as why the code is this way, what must stay true, or what a caller may rely on. It goes if it depends on:
The past. It narrates an edit, describes code that is gone, or is a finished plan or status report. Git history already holds it.
Nothing. It restates the code beside it, so no reader learns anything from it.
The implementation. It copies values, file trees or call lists out of the code, and is wrong the first time someone forgets to update it.
What makes it safe to run:
It changes no code. The diff contains comments and documents only, so the build, lint, type check and tests give the same results before and after.
It leaves what tools read. Linter and type-checker suppressions, build tags, license headers, doctests, expected-output comments in tests, and docstrings that become command-line help or API schemas stay exactly as they are. references/load-bearing.md lists the known forms for more than twenty languages and formats, and ten rules for recognizing one that isn't listed.
It keeps when unsure. A deleted reason is gone for good, so prose the agent cannot fault stays.
It needs evidence for staleness. A comment is stale when the agent can point to the code that contradicts it. Age is not evidence.
It reports what it won't settle. When a comment and the code disagree and the code may be the one that is wrong, both stay and the disagreement goes in the report.
Agent instruction files (CLAUDE.md, AGENTS.md, rule files) get the same test, held more strictly, because every line in them is read on every task. Directory tours, style rules a linter already enforces, and generic advice go. Exact commands, traps, and the owner's stated preferences stay.
收录徽章(可选)
想把收录标记放进 README 就复制这行,可选,不影响收录。
[![Listed on Anclum](https://anclum.com/badge/listed.svg)](https://anclum.com/zh/s/rusticstack-skills/)
来源
github · https://github.com/RusticStack/skills
按领域浏览
Agent 技能
收录来源
4 个独立来源
收录于
今天

更多条目

全部 471 →

所属领域与标签

每个标签都能点进去看同类的全部条目。数字是当前库里的真实条数。

浏览全部领域 →

← 返回索引