Skills Cli

作者 xixu-me498ed6f26cc3无许可证收录于 2026年10月8日更新于 2026年10月8日

Use when users ask to discover, install, list, check, update, remove, back up, restore, sync, or initialize Agent Skills, mention `bunx skills`, `npx skills`, `skills.sh`, or `skills-lock.json`, ask "find a skill for X", or want help extending agent capabilities with installable skills.

仅含说明AI & Agents
AI 生成的概览

指导使用 skills CLI 发现、安装和管理可安装的 Agent Skills。

功能
该技能提供通过 skills CLI(用 bunx 或 npx 运行)使用开放 Agent Skills 生态的说明。内容涵盖技能发现与质量核查、安装来源与参数,以及列出、检查、更新、移除、初始化、恢复和同步等日常管理操作。它还介绍一个用于声明式备份与恢复的配套工具,并给出推荐格式与故障排查指引。
适用场景
当用户要求查找、安装、列出、检查、更新、移除、备份、恢复、同步或初始化 Agent Skills,或提到 bunx skills、npx skills、skills.sh、skills-lock.json 时使用。它不适用于撰写或改进已有本地技能的内容。
运行要求
需要 skills CLI,通过 bunx(Bun)或 npx(Node.js)调用,并需要网络访问以获取技能包。该技能不附带脚本,仅为说明文档。

Use this skill to help users work with the open Agent Skills ecosystem through the skills CLI.

Overview

The skills CLI is the package manager for installable Agent Skills. Use it to discover skills, install them with the right flags, and manage them after installation.

Examples below use bunx skills, but npx skills is the same workflow if Bun is not available in the user's environment.

Always prefer the current CLI syntax:

bash
bunx skills add <source> --skill <name>

Do not use older owner/repo@skill-name examples.

When to Use

Use this skill when the user:

  • asks "find a skill for X", "is there a skill for X", or "how do I do X" and X sounds like a reusable workflow
  • asks "can you do X" and X sounds like a specialized capability that may already exist as a skill
  • wants help with bunx skills, npx skills, skills.sh, skill package installation, or skills-lock.json
  • wants to install a skill for a specific agent such as Codex or OpenCode
  • wants to list, check, update, remove, restore, sync, back up, or initialize installed skills
  • wants help searching for workflows, tools, templates, or domain-specific capabilities such as design, testing, deployment, documentation, or code review

Do not use this skill when the user already has a local skill and wants help writing or improving its contents. In that case, use a skill-authoring workflow instead.

Discovery Workflow

When a user needs a skill, follow this sequence:

  1. Identify the domain and task. Examples: React performance, PR review, changelog generation, PDF extraction. Also judge whether the task is common enough that a reusable skill is likely to exist.
  2. Check skills.sh first. Prefer well-known, well-installed skills when the domain is already covered there.
  3. If the leaderboard does not clearly answer the need, search with:
bash
bunx skills find <query>
  1. Verify quality before recommending anything:
    • install count: prefer skills with 1K+ installs and be cautious with anything under 100
    • source reputation: prefer official or well-established maintainers such as openai, anthropics, microsoft, or similarly trusted publishers
    • repository quality: check the source repository and treat skills from repos with fewer than 100 stars skeptically
  2. Present the options clearly. Include the skill name, what it helps with, the install count and source, why it looks trustworthy, the install command, and a link to learn more on skills.sh.
  3. Offer installation help if the user wants to proceed.
  4. If nothing fits, say so directly, help with the task using your general capabilities, and mention that the user can create their own package with bunx skills init.

Installation Quick Reference

Common sources

bash
# GitHub shorthandbunx skills add xixu-me/skills
# Full GitHub URLbunx skills add https://github.com/xixu-me/skills
# Direct path to one skill inside a repobunx skills add https://github.com/xixu-me/skills/tree/main/skills/skills-cli
# GitLab URLbunx skills add https://gitlab.com/org/repo
# Any git URLbunx skills add [email protected]:owner/repo.git
# Local package pathbunx skills add ./my-local-skills

Common install patterns

bash
# List skills in a package without installingbunx skills add <source> --list
# Install one skillbunx skills add <source> --skill skills-cli
# Install multiple skillsbunx skills add <source> --skill pr-review --skill commit
# Install globallybunx skills add <source> --skill skills-cli -g -y
# Install to a specific agentbunx skills add <source> --skill skills-cli -a codex -y
# Install all skills to all agentsbunx skills add <source> --all
# Install all skills to one agentbunx skills add <source> --skill '*' -a codex -y
# Copy files instead of symlinkingbunx skills add <source> --skill skills-cli -a codex --copy -y

Installation methods

When the user is choosing how to install:

  • symlink is the default and usually the best choice because updates stay centralized
  • --copy creates independent copies and is the fallback when symlinks are unsupported or inconvenient

If the user only asks to install a skill, prefer the default symlink workflow unless they mention CI packaging, portability, filesystem restrictions, or explicitly ask for copies.

Important flags

FlagUse
--skill <name>install one or more named skills
-a, --agent <agent>target specific agents such as codex
-g, --globalinstall at user scope instead of project scope
-y, --yesskip prompts
--listlist available skills in a package
--copycopy instead of symlink
--allshorthand for all skills to all agents

Managing Installed Skills

Use these commands for ongoing maintenance:

bash
# List installed skillsbunx skills lsbunx skills ls -gbunx skills ls -a codexbunx skills ls --json
# Check for updatesbunx skills check
# Update installed skillsbunx skills update
# Remove installed skillsbunx skills remove my-skillbunx skills remove my-skill -a codexbunx skills remove -g my-skillbunx skills remove --all
# Initialize a new skill packagebunx skills initbunx skills init my-skill
# Restore from skills-lock.jsonbunx skills experimental_install
# Sync node_modules skills into agent directoriesbunx skills experimental_syncbunx skills experimental_sync -a codex -y

When the user asks to initialize a skill, explain whether they want:

  • bunx skills init to create SKILL.md in the current directory
  • bunx skills init <name> to create a new subdirectory containing SKILL.md

Related Tool: Skills Vault

If the user wants declarative backup and restore of installed skills across machines or teams, use Skills Vault.

Skills Vault is a separate CLI companion for the skills ecosystem. It is not a skills add installable skill source. Use it when the user wants to snapshot installed skills into a manifest, preview restore commands, or reproduce the same setup elsewhere.

Common companion commands:

bash
# Back up installed skills into skvlt.yamlbunx skvlt backup
# Preview a restorebunx skvlt restore --dry-run
# Restore everything from the manifestbunx skvlt restore --all
# Diagnose the local environmentbunx skvlt doctor

Prefer this tool over skills experimental_* when the user explicitly wants a portable manifest workflow, cross-machine backup and restore, or team-sharing of installed skill setups.

Recommendation Format

When recommending a skill, keep the answer concrete and installable.

Use a structure like this:

text
I found a skill that should fit.
Skill: <skill-name>Why it matches: <one sentence>Source: <owner/repo or URL>Quality check: <install count / source reputation / repository confidence note>Install:bunx skills add <source> --skill <skill-name> [optional flags]Learn more: https://skills.sh/<publisher>/<package>/<skill-name>
If you want, I can install it for <agent-or-scope>.

If the user mentions a target agent or scope, include it in the command. Examples:

bash
bunx skills add <source> --skill <skill-name> -a codex -ybunx skills add <source> --skill <skill-name> -g -y

Example:

text
I found a skill that might help.
Skill: screenshotWhy it matches: it focuses on OS-level desktop and window screenshot capture.Source: openai/skillsQuality check: high install volume, trusted publisher, and a widely used source repository.Install:bunx skills add openai/skills --skill screenshotLearn more: https://skills.sh/openai/skills/screenshot

Common Skill Categories

When the user's wording is vague, map it to likely categories:

CategoryExample queries
Web Developmentreact, nextjs, typescript, css, tailwind
Testingtesting, jest, playwright, e2e
DevOpsdeploy, docker, kubernetes, ci-cd
Documentationdocs, readme, changelog, api-docs
Code Qualityreview, lint, refactor, best-practices
Designui, ux, design-system, accessibility
Productivityworkflow, automation, git

Search Tips

  • Use specific keywords. react testing is better than just testing.
  • Try alternative terms. If deploy fails, try deployment or ci-cd.
  • Check popular sources first. Many strong skills come from established publishers.
  • If the first search is too broad, narrow by domain plus task.

Common Mistakes

  • Recommending a skill from search results without checking whether it looks established.
  • Forgetting to specify -a <agent> when the user asked for one particular agent.
  • Treating bunx skills find --help like a real help command. Use bunx skills --help for command help instead.
  • Assuming no skill exists after one weak search term. Try a more specific or adjacent query first.

Troubleshooting

If the user hits an error or confusing result:

  • "No skills found" - suggest a better query, check skills.sh, or help directly and mention bunx skills init
  • interactive prompts in automation or CI - add -y
  • wrong installation scope - switch between project install and -g
  • symlink issues - retry with --copy
  • uncertainty about available package contents - run bunx skills add <source> --list
  • uncertainty about installed state - run bunx skills ls or bunx skills ls --json
  • portable backup or restore across machines - mention Skills Vault and its backup / restore --dry-run workflow

When you are unsure about exact flags, use:

bash
bunx skills --help

来源与署名

来源:xixu-me/skills位于skills/skills-cli提交498ed6f

许可证: 无许可证

内容归原作者所有。SourceWeft 从公开仓库中收录这些内容。

举报或申请下架