Your claude.ai skills follow you into Claude Code 2.1.275

Hit ctrl+enter while Claude is mid-answer and it stops the turn and fires off everything you queued while you waited. That send-now key is one of the smaller lines in the v2.1.275 notes on Github; the bigger one is that the skills and plugins enabled on your claude.ai account now follow you into terminal sessions signed in with that account.
At a glance
- Skills and plugins enabled on your claude.ai account now sync into terminal sessions signed in with it, and turning that off takes two separate settings: syncClaudeAiSkills: false and syncClaudeAiPlugins: false.
- The send-now key, ctrl+enter or ctrl+x ctrl+s, interrupts the current turn and sends every queued message at once; sent and queued messages stay gray until the model receives them.
- Plugins installed from an npm source are now fetched with npm pack --ignore-scripts and integrity-verified, so a package's install scripts no longer run when you add the plugin.
If you have not been following the skills story: according to Verdent AI, Anthropic launched Agent Skills publicly on October 16, 2025, with pre-built skills for PPTX, XLSX, DOCX and PDF work, and on December 18, 2025 published the Agent Skills specification as an open standard at agentskills.io. Until this release, a skill you set up in the browser was not there when you opened the terminal.
Ctrl+enter cuts the turn short and flushes the queue
The key is ctrl+enter, or ctrl+x ctrl+s if your terminal swallows the first one. It interrupts the turn Claude is in the middle of and sends everything you have queued at once, instead of waiting for the model to finish. Messages already sent and messages still waiting show in gray until the model receives them.
Queued messages got attention elsewhere too. In VS Code, a message sent while Claude is working waits at the bottom of the conversation until Claude starts on it, and a slash command typed during a reply runs when the reply finishes instead of going to the model as plain text. A resumed cloud session also stops leaving an unanswered question open in the transcript after a queued message superseded it.
Synced skills are files you can edit, but not from the terminal
Skills and plugins enabled on your claude.ai account sync to terminal sessions signed in with that account. Two settings switch it off separately: syncClaudeAiSkills: false and syncClaudeAiPlugins: false. Write and Edit results for files in the synced account-skills folder now say the change is not saved to your account, and how to save it.
The ListPlugins tool description was rewritten so Claude knows it lists the plugins enabled on your claude.ai account, not the ones installed locally with /plugin. Separately, /plugin install <plugin> --marketplace <source> offers to add the marketplace before installing the plugin. Per Claude Code's documentation, using a marketplace has been a two-step job: add the catalog, then install from it.
Gateway sign-in names the account before the credential is saved
Signing in through a Claude apps gateway shows you the account the gateway names and waits for you to confirm it before the credential is stored. Afterwards /status shows it. And /logout on a gateway sign-in now also ends the session on gateways that advertise token revocation.
Two quiet failures got noise attached to them. Users behind a network gateway that rewrites API error responses were seeing API Error: 400 on every turn when a beta request header was rejected; that is fixed. A configured otelHeadersHelper that fails now produces a startup warning, so a session exporting no telemetry at all gets noticed.
Tokens typed into a marketplace URL stop showing up in logs
Plugin and marketplace messages, logs and claude plugin marketplace list were printing a password or token stored in a git, ssh or marketplace URL. They stop doing that, and so does the Manage plugins dialog in VS Code.
Plugins installed from an npm source are fetched with npm pack --ignore-scripts and integrity-verified, so a package's install scripts do not run. And claude plugin marketplace update stops deleting a GitHub marketplace's local copy when the fetch failed and the marketplace was named after its repository.
What does a synced skill actually move?
Not a program. As Han, Not Solo describes it, skills do not run Python or JavaScript and there is no HTTP server or function call behind them: they are specialized prompt templates injected into the conversation through a meta-tool called Skill, and Claude picks one by reading the descriptions in its system prompt, with no classifier, embedding or pattern match doing the choosing.
Think of a recipe card slipped into the conversation rather than a machine switched on. The practical result in this release is a folder of files you can open locally, with Write and Edit now telling you that a change there is not saved back to your account and what to do if you want it to be.
The notes do not say when the sync runs, what happens when a synced skill and a local one share a name, or what an organization admin can do about either. And in our view the default reads oddly: the sync is on after you upgrade, and switching it off takes two keys, one for skills and one for plugins.
The first sync after you upgrade When you next sign in on a machine running 2.1.275, the thing to watch is which skills and plugins actually land in the session, and whether /status names the account you expected the gateway to name. The notes do not say whether syncClaudeAiSkills and syncClaudeAiPlugins have to be set on every machine you work from, and no date is given for the next release.
Related stories
- Claude Code 2.1.274 warns before memory turns critical
- Without CLAUDE.md, Claude Code 2.1.277 reads AGENTS.md
- Leaked Claude Code Hub lists local and cloud sessions
- Usage details button spotted in Claude for iOS
- Slash commands are coming to Claude Code on iOS
- Computer use lands in Claude Cowork and Claude Code beta
Comments
No comments yet. Be the first.
Join the conversation
Sign in with Google to leave a comment. Your name and avatar come from your Google profile, and the comment appears after moderation.
