dependency-management
关于
This Claude Skill enforces fixed version dependency installation across all package managers to ensure reproducible builds, supply chain security, and stability. It provides guidance for installing packages, updating dependencies, and working with configuration files like package.json and requirements.txt. Use it when reviewing dependency configurations or configuring CI/CD pipelines to maintain consistent dependency management practices.
技能文档
Dependency Management
Basic Principles
Always Use Exact Versions
- Use exact versions only:
[email protected] - Forbid:
^1.2.3,~1.2.3,latest,*, version ranges - Exception: Library peerDependencies only
Lock Files Are Mandatory
- Always commit to version control
- Forbid manual editing
- CI/CD must use frozen/locked mode
Security Audit First
- Check vulnerabilities before installation
- Automate regular audits
Installation Commands
# Node.js
npm install --save-exact [email protected]
pnpm add --save-exact [email protected]
yarn add --exact [email protected]
# Python
pip install package==1.2.3
poetry add [email protected]
# Go
go get [email protected]
# Rust
cargo add package@=1.2.3
# PHP
composer require vendor/package:1.2.3
# Ruby (Gemfile)
gem 'package', '1.2.3'
# Java/Kotlin
implementation("group:artifact:1.2.3") # Gradle
<version>1.2.3</version> # Maven
# .NET
dotnet add package PackageName --version 1.2.3
CI/CD Commands
npm ci # npm
pnpm install --frozen-lockfile # pnpm
yarn install --frozen-lockfile # yarn
poetry install --no-update # poetry
go mod verify # go
cargo build --locked # rust
composer install --no-update # php
bundle install --frozen # ruby
dotnet restore --locked-mode # .NET
Common Mistakes
| ❌ Wrong | ✅ Correct |
|---|---|
npm install (CI) | npm ci |
package@latest | [email protected] |
package@^1.2.3 | [email protected] |
| Lock file in .gitignore | Commit lock file |
| Manual lock file editing | Regenerate via package manager |
快速安装
/plugin add https://github.com/KubrickCode/ai-config-toolkit/tree/main/dependency-management在 Claude Code 中复制并粘贴此命令以安装该技能
GitHub 仓库
相关推荐技能
llamaguard
其他LlamaGuard是Meta推出的7-8B参数内容审核模型,专门用于过滤LLM的输入和输出内容。它能检测六大安全风险类别(暴力/仇恨、性内容、武器、违禁品、自残、犯罪计划),准确率达94-95%。开发者可通过HuggingFace、vLLM或Sagemaker快速部署,并能与NeMo Guardrails集成实现自动化安全防护。
sglang
元SGLang是一个专为LLM设计的高性能推理框架,特别适用于需要结构化输出的场景。它通过RadixAttention前缀缓存技术,在处理JSON、正则表达式、工具调用等具有重复前缀的复杂工作流时,能实现极速生成。如果你正在构建智能体或多轮对话系统,并追求远超vLLM的推理性能,SGLang是理想选择。
evaluating-llms-harness
测试该Skill通过60+个学术基准测试(如MMLU、GSM8K等)评估大语言模型质量,适用于模型对比、学术研究及训练进度追踪。它支持HuggingFace、vLLM和API接口,被EleutherAI等行业领先机构广泛采用。开发者可通过简单命令行快速对模型进行多任务批量评估。
langchain
元LangChain是一个用于构建LLM应用程序的框架,支持智能体、链和RAG应用开发。它提供多模型提供商支持、500+工具集成、记忆管理和向量检索等核心功能。开发者可用它快速构建聊天机器人、问答系统和自主代理,适用于从原型验证到生产部署的全流程。
