@dotey: Sharing how I manage Skills — a geeky approach that may not suit everyone, but might give you some ideas. 1. Install Skills only inside projects, not globally. Agent Skills can be installed globally (shared across all projects) or within a project (only usable in the current project). I choose to only install...

Here’s my approach to managing Skills — it’s a bit geeky, not for everyone, but might give you some ideas.

1. Skills live only inside projects, not globally

Agent Skills can be global (shared across all projects) or project-specific (only available in the current project). I choose project-only, mainly to save context window space.

When an agent works, it has a context window — think of it as Claude’s workbench with limited desktop space. Although a Skill only loads its name, description, and other summary info by default (full content isn’t expanded until needed), it adds up — dozens of global Skills, each with its own summary, eat up a noticeable chunk of the workbench. Plus, once Claude decides a Skill is relevant to the current task, it loads the complete content. The more global Skills you have, the higher the chance of false triggers, wasting precious space.

By installing only the Skills you truly need inside each project, you keep only the relevant materials on the workbench, leaving room for more important content. Claude works more efficiently too.

2. Use symbolic links to install Skills

This is the core of my management method. Let me first explain what a symbolic link is.

Think of a symlink as a Windows shortcut — the actual file exists in one place, but you can create shortcuts pointing to it in many places. If you modify the original, all shortcuts automatically reflect the change.

Here’s my three-step workflow:

Step 1: Download open-source Skills into a unified directory

I have a dedicated folder ~/GitHub on my machine for all GitHub projects. All downloaded open-source projects live here, for example:

~/GitHub/baoyu-skills      ← open-source project containing various Skills
~/GitHub/baoyu-design      ← another open-source project

This folder acts as a warehouse — the original copies of all Skills are stored here.

Step 2: Create symbolic links inside your own project

Suppose I have a writing project ~/GitHub/baoyu-writing that needs several Skills. Instead of copying them in, I create symlinks pointing to the originals in the warehouse:

Path inside project                 →  Actual location (original)
.agents/skills/baoyu-comic          →  ~/GitHub/baoyu-skills/skills/baoyu-comic
.agents/skills/baoyu-design         →  ~/GitHub/baoyu-design/skills/baoyu-design

Step 3: Create an entry point for Claude Code

Finally, create one more symlink so Claude Code can find all these Skills:

.claude/skills  →  .agents/skills

Now Claude Code can follow the chain and locate all the Skills it needs.

3. No need to memorize commands — let the Agent do it

You might be thinking, “I can’t remember the symlink commands!” No need at all. Just tell Codex/Claude Code what you want in natural language, for example:

Help me symlink ~/GitHub/baoyu-skills/skills/baoyu-comic to .agents/skills/baoyu-comic

Or even simpler:

Link the skill baoyu-comic from the baoyu-skills project into the current project

The Agent will create the symlink for you automatically. Subsequent maintenance, adding, and removing can also be delegated to it. Just say which Skill you want linked where, and let the Agent handle the rest.

4. Why go through all this trouble?

The initial setup takes a few more minutes than copy-pasting, but later maintenance is much easier. Two main benefits:

Benefit 1: Update once. Because all projects point to the same original files via symlinks, when the open-source project updates, I only need to pull the latest code in ~/GitHub/baoyu-skills. Every project using that Skill automatically gets the new version.

Benefit 2: Bug fixes flow back upstream. For example, while using a comic Skill in my writing project, I might discover a bug. I can ask the Agent to fix it right there. Since it’s a symlink, the Agent actually modifies the original file (~/GitHub/baoyu-skills/skills/baoyu-comic). I can then directly submit the fix back to the open-source project — a contribution to the community done almost as a side effect.

LinearUncle (@LinearUncle): I’m going crazy. I use 4 coding agents daily, and each writes its own skills in different places: /.agents/skills, /.codex/skills, ~/.claude/skills

It’s already a mess — Claude Code can’t find Codex’s skills, and vice versa. My current plan:

Write a script to symlink each agent’s unique skills into the other two locations, run it periodically via cron. Is there a more elegant solution or tool?