How to sync your Claude skills across Cowork and Code in 2 min
(Free Skill inside)
I had 48 Claude skills spread across three environments: Claude Code, Claude Cowork and my skill vault… if you’re using Claude Code and Cowork, you know the pain!
I thought they were all the same but they weren’t even close… (As a data scientist who is used to tracking changes in my Github repo, I find it quite funny that I had not seen that coming.)
I found out the way you always find out, aka something broke.
I was in a Cowork session,
ran a skill I’d updated two days earlier in Claude Code, and
the output was wrong, like stale-wrong.
The Cowork version was running an older copy that didn’t have my latest fixes.
The updated version was sitting in my Claude Code repo, perfectly fine, completely unreachable from where I was working.
I checked a few more skills. Same story. Some were current in Cowork but missing from my repo. Others existed in the repo but never made it to my backup vault. A handful had drifted, small edits in one place that never got mirrored to the other.
I had been building Claude skills for weeks without noticing the gap widening underneath me.
I started calling this “skill drift”, meaning ****the silent divergence that happens when you create and edit skills across multiple environments without a system to keep them aligned.
And once I saw it, I couldn’t unsee it.
The Claude environment problem
If you’re building skills in Cowork, you already know the setup. But you might not realize how many places your skills actually live.
Cowork Desktop
This is where your installed skills run during sessions. Here’s the thing most people miss: it’s read-only. You can’t write to it. You can’t update a skill here from inside a session. It’s a snapshot of whatever was installed when you started.
Your Claude Code repo
A folder on your machine with a skills subdirectory. This is where you actually edit skills, test changes, iterate. It’s writable. It’s where the real work happens.
Your skill-vault
A folder of .zip backups. You can move it between machines, share it with collaborators, and restore any skill if something goes sideways.
Three copies. Three locations. And the filesystem has rules that punish you if you don’t know them.
Without a sync system, you become the person manually tracking which version lives where. You become a human version-control layer for your own tools.
That’s the friction I wanted to eliminate.
Skill Sync is one skill that does one job: audit all three environments and close the gaps between them.
Here’s the actual workflow.
Step 1: Run the inventory.
One script scans Cowork skills, Cowork plugins, remote plugins, your Claude Code repo, and your skill-vault. It produces a structured report showing skills aligned, missing, drifted, and duplicated.
48 skills. One scan. I could see everything that had drifted in about 10 seconds.
Step 2: Copy Cowork → Repo.
For every skill that exists in Cowork but not in the repo, the sync copies it over using the cat redirect method. Not cp. That lesson cost me an hour I’m not getting back.
Step 3: Sync overlapping skills.
For skills that exist in both places, it runs a diff. If they’ve diverged, it asks which direction to sync.
Step 4: Package to the vault.
Every skill in the repo gets zipped and sent to the vault. The script writes to /tmp first, then uses cat to move the .zip file over.
Step 5: Flag cleanup.
The script identifies old duplicates .skill files that now have .zip versions, renamed skills where the old name is still hanging around.
Step 6: Verify.
Run the inventory one more time. Confirm zero gaps.
That’s the whole system. Six steps, one command to kick it off.
What surprised me
The inventory was the real eye-opener because I expected only one or two skills to be out of sync.
I found:
four missing from my repo entirely,
three that had never been packaged to the vault, and
two duplicates under different names that I’d forgotten about.
But the part I didn’t expect was that once everything was aligned, I started building WAY faster because I stopped second-guessing which version was current.
And in the end that’s what really matters when building with AI: the confidence to build without worrying about losing work.
Where to start if you want to FINALLY have an organized skill-vault.
First, download the Skill Sync Claude Skill here:
Run the inventory and see where you stand.
Skills are becoming the building blocks of how we work with AI with full workflows built from prompts, tools, and real process logic stacked together, compounding over time. Keeping them aligned across environments isn’t a nice-to-have anymore ahah.
If you want to learn how to build an AI-powered data portfolio in 7 days, I am launching a FREE 7-day guided challenge in 17 days to help you build your data portfolio project using Claude Code.
📌 You can sign up here to be the first to know when it launches (and get all the bonuses): https://ai-powered-data-career.vercel.app/
Keep Building,
Thanks for this. I have a few repos locally, Claude Cowork, etc and really feeling the skill creep.