TechLatest

Claude Code now brings your claude.ai skills to the terminal

Skills you create or enable at claude.ai now download into Claude Code when you sign in with the same account. How it works, its limits and how to stop it.

Published 4 min read

Baca dalam BM

GitHub card for the anthropics/claude-code repository, with the Anthropic logo and a count of 150k stars
Image: GitHub

If you have made a skill on claude.ai, say one that formats your company's quotations or writes in your team's house style, Claude Code can now use it in the terminal without you copying any files. Anthropic added the sync in September and was still adjusting it in October.

For developers and small teams who use both the Claude app and Claude Code, this means one place to keep skills. You edit them on claude.ai, and every machine where you sign in to Claude Code picks up the change.

What changed

The Claude Code changelog for version 2.1.275, released on 18 September (Malaysia time), says Claude Code now syncs the skills and plugins enabled on your claude.ai account to terminal sessions signed in with that account. According to Anthropic's documentation:

  • When a session starts, Claude Code downloads your account's skills into ~/.claude/skills/synced/ in the background.
  • It then keeps checking claude.ai. When a skill is added, edited or turned off there, the running session updates without a restart.
  • The synced skills include ones you create or turn on at claude.ai, skills your organisation provides, and Anthropic's built-in skills such as pdf and xlsx.
  • Run /skills to see them. They are listed under claude.ai sync.

Two later updates changed the details. Since version 2.1.293 (8 October), an idle session checks for changes about every 40 minutes instead of every 10. While you are working, it still checks about every 10 minutes. Version 2.1.296 (10 October) fixed synced skills that never finished installing when several sessions share one config folder.

Where to make the skill

On claude.ai, skills live under Customize > Skills, where you can create one or upload a skill folder as a ZIP file. Anthropic's help centre says skills need code execution and file creation turned on.

It only goes one way

Claude Code downloads synced skills and never uploads them. If you or Claude edit a file in ~/.claude/skills/synced/, the change is not saved to your account, and the next sync can overwrite or remove it. To change a synced skill, edit it on claude.ai.

The sync only runs when you sign in with your claude.ai account through /login. The docs say it does not run in sessions that use an API key, Amazon Bedrock, bare mode or --safe-mode. If you sign in partway through a session, restart Claude Code to start syncing.

When names clash

You can call a synced skill by its short name, /<name>, or its full name, /anthropic-skills:<name>. If a local skill or command already uses the short name, the local one wins. With a local deploy skill and a synced deploy, /deploy runs the local one and /anthropic-skills:deploy runs the synced one.

Safety limits

A skill can contain lines starting with !, which run a shell command when the skill loads. Anthropic's docs say Claude Code does not run these commands on your machine for synced skills, and does not attach the files that @ references point to. It also cleans up a synced skill's description so the text cannot imitate Claude Code's own formatting. In cloud sessions, which run in an isolated container, synced skills behave like local ones.

If your organisation turns Skills off at claude.ai, Claude Code removes the downloaded copies and moves them to ~/.claude/skills/.trash/.

How to turn it off

To stop syncing on one machine, add this to your user settings file, ~/.claude/settings.json:

{
  "syncClaudeAiSkills": false
}

Claude Code then stops downloading, and the next time it starts it moves the skills it already synced to the trash folder. A separate setting, syncClaudeAiPlugins, does the same for plugins.

Check what your skills cost

Every installed skill adds its name and description to Claude's context on every turn, whether or not Claude uses it. The /skill-doctor command, added to Claude Code in early September, shows what each skill costs in context and how often it gets used, and flags skills that have never been invoked. Anthropic's advice is to start by turning off unused skills with the highest context cost. Worth a look once your claude.ai skills start appearing in every terminal session.

Sources

ShareWhatsAppXFacebook
Meja Berita

The saksama editorial team