Solving AI Agent Skills Fragmentation with npx skills

If you run multiple AI agents or CLIs on your computer—whether it's Hermes, Claude Code, Codex, Antigravity, Kimi, or other supported tools—you've probably hit a frustrating issue: skills fragmentation. Every agent or CLI maintains its own separate skills folder, which leads to duplicate installations, version drift, and management overhead.

In our work with AI agents, we frequently ran into the following issues:

• Multiple installs: Installing a new skill for one agent, then install it again and again for others.
• Version drift: Customising or fixing a skill for one agent, forgetting to update the rest, and ending up with agents behaving differently.
• Skills management: Having to go through every skills folder for each agent across our system to see what skills were actually installed.

The npx skills CLI, an open-source CLI from Vercel, solves the issue by letting you manage one centralised skills folder and share it across all your agents. You can check out the npx skillsGitHub repo.

Centralised Folder

The concept is simple: if we point every agent's skills folder to the same physical folder, the fragmentation problem completely disappears.

When you install a skill using npx skills CLI, the skill is downloaded to the centralised folder on your machine (~/.agents/skills). The CLI then creates a symlink (which is just a pointer) from each agent's skills folder back to that centralised folder.

For example, when Claude Code looks for skills in ~/.claude/skills/, it gets transparently redirected to ~/.agents/skills. If you make a tweak to a skill in that centralised folder, every single agent sees the update instantly. No manual copying, no sync scripts, and no wasted disk storage.

Installing npx skills and Adding a Skill

Run the following command to get it started, and add our first skill:

npx skills add https://github.com/coleam00/excalidraw-diagram-skill --skill excalidraw-diagram

The CLI will ask whether to set this up for a specific project or globally. We usually recommend choosing global so all supported agents can access it:

Next, make sure to select Symlink to ensure a single source of truth:

Once done, we can verify that the skill is installed in the centralised folder at ~/.agents/skills:

Open up the Kimi CLI (or any other agent), and the excalidraw-diagram skill is available and ready to use!

Creating a Skill from Scratch

To build your own skills, you can use Anthropic's skill-creator tool. Install it like this:

npx skills add https://github.com/anthropics/skills --skill skill-creator

Now, in Kimi or any other agent, enter the following prompt:

Use the skill-creator to help me build a skill for creating a product landing page

Kimi or your agent will automatically pick up the skill-creator and walk you through the process step-by-step:

Other Handy npx skills Commands

Here are some other handy commands:

• npx skills ls -g: List all globally installed skills
• npx skills update: Keep all skills up to date with one command
• npx skills remove: Interactively uninstall/remove a skill

Please refer to the npx skills GitHub repo for more commands and reference.

Taking It a Step Further

Since all skills are now in one place (~/.agents/skills), we can initialise it as a Git repository and push it to GitHub or GitLab.

cd ~/.agents/skills
git init
git add .
git commit -m "Initial skills folder"
git remote add origin https://github.com/Tech-Voyage-Dev/skills
git push -u origin main

Now, we have a complete history of all skill changes, meaning we can easily roll back if an update breaks something. Plus, we can seamlessly sync the skills across multiple machines (like a laptop and desktop) by running a quick git pull.

A Few Things to Keep in Mind (Security & Risks)

• Compatibility varies: Not all agent platforms support every feature (like custom hooks or specific allowed tools) yet. A skill might work great in one agent but fall short in another.
• Trust the source: Only install skills from trusted sources.
• Review before running: If you are using a custom or third-party skill, review the files (especially in the scripts/ folder) before letting your agent run it.
• Least privilege: Try not to give a skill more tool permissions or access than it absolutely needs to do its job.
• Never hardcode secrets: Keep your API keys and secrets out of your skill files.

Setting up a centralised "single source of truth" with npx skills has saved our team from frustration and duplicate work. It is highly recommended to give it a try—the AI capabilities scale as skills grow!