SI Agent News
ReportedClaudeOpenAIGeminiAgents

One folder, many agents: how a shared Skill library ends copy-paste upkeep

A dev.to write-up says the fix for reusing agent Skills is one master folder that every tool links to, not a copy for each tool. Here is how it works.

32tools Skills Manager can link one Skill library to

The short version

Why it matters

If you write reusable instructions for more than one AI agent, keeping copies in sync is a quiet source of errors. One source of truth means every agent follows the same version. It also keeps your work portable if you change tools.

The copy-paste problem

As dev.to reported, the author has one set of Skills but uses three agents. A Skill here is simply a folder holding a file called SKILL.md, which tells an agent what it can do and how to do it. Every edit meant copying the file into two other places, and before long the author could not say which version was the original and which was a stale copy from a month earlier.

The cause is that each tool looks in its own place. According to the post, Claude Code reads personal Skills from ~/.claude/skills, while Codex reads from $HOME/.agents/skills and also scans a project's .agents/skills folder. Cursor, Gemini CLI, OpenCode and others have their own folders too.

With two agents, one edit means two changes. With five, it means five. If the versions drift apart, one agent may follow the old method and another the new one. That is hard to debug because the outputs look similar.

The idea: one master, many shortcuts

The proposed fix is to stop letting each agent own its Skills. You keep a single central folder as the original, and every agent is pointed at it.

The pointing is done with a symlink, short for symbolic link. It is a shortcut to the real file, not a copy. Edit the original once and every agent sees the change straight away.

The author says this reframes the question. Instead of asking how to keep copies in sync, you ask why the copies exist at all.

Three projects, three levels

The post says several open-source projects already build on this idea, and it names three that solve different parts of the problem.

The author sees them as complementary: the first for a single machine, the second for browsing and curating, the third for teams and companies.

Trying it without an app

Skills Manager also ships a command-line tool called skm that works on machines without a screen. Its adopt command moves Skills scattered across each tool's folder into the central library and re-links them. The post stresses running it first with a --dry-run flag, because adopt moves real files and a mistake would be hard to undo. Two other commands, doctor and fix, check for broken links and repair them.

You can do the same by hand. Create a central folder, write the SKILL.md there, then run ln -s to link it into each tool's Skills folder. Check the result with ls -l. The post says this works immediately on macOS and Linux. On Windows you need Developer Mode switched on, or a similar mechanism called a junction.

For a history of changes, the author suggests making the central folder a Git repository. One caution: do not point links into a folder that some cloud-storage tools sync, since links can break when files move.

Traps to watch for

The author lists several common failures, and most of them fail quietly.

Agents also sometimes write new Skills mid-task, and these land in that tool's own folder. Adopting them back into the central library stops them being tied to one agent.

Who needs this, and a caveat

The author's advice is measured. With one agent and fewer than ten Skills, an extra app is probably overkill. With two or more agents, or Skills shared by a team, a central library cuts repeated work and makes it clearer which version each agent is using. The author started with plain symlinks and would add a tool only once the number of Skills justified it.

Two caveats apply. The article says it was written by an AI model under human review, and we have not run any of these tools. Also, the headline question, whether a Skill really behaves identically in every agent, is only partly answered. Putting a file in the right folder is not the same as an agent interpreting it the same way, so test a Skill in each tool before relying on it.

What to watch

Try it

If you use two agents, make one master folder, link it into both Skills folders with ln -s, then run ls -l to confirm. Test the Skill in each tool. Use --dry-run first if you try skm adopt.

Sources and links

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

Listen to the AI Wire headlines

Listen to the AI Wire

Listen to our daily AI shows every morning: on your commute, in the shower, wherever you are

Listen to the daily shows

First reported by dev.to · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

More AI news