Sharing a Claude Code skill is simpler than most developers expect. A skill is a folder. Copy the folder and you've shared the skill. But a folder that works perfectly in your project will fail in someone else's project for three predictable reasons. Knowing those three reasons before you share saves the recipient from an hour of confused debugging.

TL;DR: Share a Claude Code skill by copying its folder to the recipient's .claude/skills/ directory. Before sharing, check that the skill uses no absolute paths, has no project-specific reference files, and includes all the reference files it needs within the same folder. For wider distribution, GitHub is the standard hosting platform. Community platforms like SkillsMP and SkillHub let others discover and install your skill.

How Do You Share a Skill with a Teammate?

Direct file sharing is the simplest method. Copy the skill folder from your project's .claude/skills/ directory and send it to your teammate. They drop it in their .claude/skills/ directory — at the project level for project-specific use, or at ~/.claude/skills/ for user-level access across all projects.

That's the entire transfer. The skill folder contains everything needed to run:

  • SKILL.md — the instructions and description
  • Any reference files the skill loads on demand
  • Any asset files (templates, approved examples) the skill uses

What it doesn't contain: your project-specific context. If your skill includes a step like "Read the project's CLAUDE.md and follow its conventions," that step will fail in the recipient's project if their CLAUDE.md has different conventions. Any instruction that references your specific project state needs to be made generic before sharing.

For a full explanation of what goes in a SKILL.md file, see What Goes in a SKILL.md File?.

What Should You Check Before Sharing?

Three checks before sending a skill to anyone:

1. No absolute paths. If your skill's instructions or reference files contain absolute paths (like /Users/your-name/projects/...), they'll break on any other machine. Replace absolute paths with relative paths: ./references/guide.md instead of /Users/your-name/.claude/skills/skill-name/references/guide.md.

2. All referenced files are included. Open SKILL.md and find every Read instruction or file reference. Verify that each referenced file is inside the skill folder itself. If your skill references a shared file that lives outside the skill folder (in your project root, or in another skill), copy that file into the skill folder before sharing. A missing-file error on the recipient's machine is the most common post-share complaint.

3. No project-specific assumptions. Instructions like "use the coding style from this project" or "follow the brand guidelines we use" only work in your context. Make the skill self-contained: include the brand guidelines as a reference file, or make the instruction generic with a path that the recipient can fill in.

How Do You Share on GitHub?

GitHub is the standard distribution channel for skills. The typical structure is a repository with one skill per folder:

your-skills-repo/
  skill-name/
    SKILL.md
    references/
      guide.md
    assets/
      template.md
  another-skill/
    SKILL.md

Recipients clone the repository and copy the skill folder they want into their .claude/skills/ directory.

A good GitHub README for a skill repository includes:

  • What each skill does (one sentence each)
  • The trigger phrase that activates each skill
  • Which Claude Code version the skills were tested with
  • Installation instructions (copy the folder to .claude/skills/)

For a detailed guide on GitHub-based skill publishing, see Can I Publish My Skill on GitHub?.

What Are the Community Platforms for Skill Distribution?

Three platforms exist specifically for Claude Code skill sharing:

SkillsMP (skillsmp.com): The largest community directory. Lists skills by category with installation instructions. Quality ranges widely, from production-grade engineered skills to basic prompts saved as SKILL.md files. Check the description quality and whether the skill includes reference files before installing. Skills with reference files are more likely to be production-ready than those that consist only of a SKILL.md.

SkillHub: Focused on curated, reviewed skills. Smaller selection than SkillsMP but higher average quality. Skills on SkillHub are reviewed before listing, which filters out the most broken submissions.

GitHub Topics: Search #claude-code-skills or #skillmd on GitHub to find repositories. Many individual developers publish skill collections this way. The advantage: you can read the full source before installing. GitHub's version history lets you see how the skill has evolved over time.

The quality difference between community skills and engineered skills is real. 700,000 skills exist across these platforms. Most of them are prompts wearing a trenchcoat: a description, some instructions, no reference files, no evals, no tested failure modes. If you need a skill for production use, evaluate it before installing.

"Developers don't adopt AI tools because they're impressive — they adopt them because they reduce friction on tasks they repeat every day." — Marc Bara, AI product consultant (2024)

How Do You Make a Skill Generic Enough to Share?

Skills built for your own use often contain assumptions that don't transfer. The generalization process:

  1. Remove project-specific references. Replace your project name, file paths, and domain-specific terms with generic equivalents or make them configurable.

  2. Include all domain knowledge in the folder. If your skill works because you have a reference file elsewhere in your project, copy that reference file into the skill folder.

  3. Add a configuration section to SKILL.md. Some skills need light customization for each project. Document what the user needs to change after installing: "Edit the brand-name value in references/config.md before using this skill."

  4. Test in a blank project. Create a new project with nothing in .claude/ except the skill. Run the skill there. If it fails, something is missing from the folder or there's a project-specific assumption that needs fixing.

Frequently Asked Questions

Can I share skills without giving people the source files? No. Skills are plain text files. There's no compiled or obfuscated format. Anyone who installs your skill can read the SKILL.md and all reference files. If your skill contains proprietary logic you don't want shared, don't share the skill.

Do I need a license if I share my skills on GitHub? Not legally required, but recommended. Without a license, the legal default is "all rights reserved," which means others technically can't use your skill. Adding an MIT license makes it explicitly free to use, modify, and distribute. GitHub can generate this automatically during repository creation.

Can I share skills with people who don't have Claude Code? No. Claude Code skills only work with Claude Code. A SKILL.md file sent to someone using a standard Claude conversation or a different AI assistant won't activate. The trigger mechanism is Claude Code-specific.

How do I update a shared skill after I improve it? For teammates: resend the updated folder or set up a shared git repository they pull from. For GitHub distributions: push updates to the repository with a version note. Users who installed from GitHub need to pull the updated folder manually. There's no auto-update mechanism.

What's the difference between sharing a skill and publishing a plugin? Plugins package skills (and MCP tools) for installation through Claude Code's plugin system. Plugins have better discoverability and installation UX than raw folder sharing. For most sharing use cases, folder-based sharing or GitHub is enough. Plugins are worth the extra packaging effort if you want formal distribution through Claude Code's official channels.

What if someone finds a bug in my shared skill? For GitHub-hosted skills: accept bug reports as GitHub Issues. For SkillsMP/SkillHub: each platform has its own reporting mechanism. Fix the bug in your source, update the hosted version, and note the fix in the README or commit message.

Last updated: 2026-04-23