Anthropic · 2026-09-17 · major
Claude Code 2.1.275 — claude.ai skills and plugins sync to the terminal
Claude Code 2.1.275 syncs the skills and plugins enabled on your claude.ai account into terminal sessions signed in with it, and adds a key that sends every queued message at once. A same-day 2.1.276 fixes a gateway regression.
The skills and plugins you switch on at claude.ai now follow you into the terminal.
Quick facts
| Version | v2.1.275 |
|---|---|
| Latest patch | v2.1.276 |
| Maker | Anthropic |
| Released | 2026-09-17 |
| Opt-out settings | syncClaudeAiSkills, syncClaudeAiPlugins |
| New command | /plugin install <plugin> --marketplace <source> |
What is it?
Account-level syncing arrives in Claude Code 2.1.275: terminal sessions signed in with a claude.ai account now pull down the skills and plugins enabled on that account, instead of needing each one installed locally. Two settings, syncClaudeAiSkills and syncClaudeAiPlugins, turn it off. The release also adds a send-now key and a marketplace-aware plugin install.
How does it work?
The sync keys off the signed-in identity, which this release also makes more visible — when the Claude apps gateway names the account, you confirm it before the credential is saved, and /status shows which account is in use. Queued input gets its own path too: a send-now key interrupts the current turn and hands the model every queued message together, with sent and queued text greyed out until the model actually receives it.
Why does it matter?
Anyone running Claude Code across several machines had to re-install the same plugins on each one; the sync removes that step and keeps a team on one source of truth. Two fixes matter for long sessions: a restored memory file's age note used to change between requests after a compaction or resume, silently costing prompt-cache hits, and subagent messages spawned by a forked skill were dropped from stream-json and SDK output. Upgrade past this version, though — 2.1.276 landed hours later to undo a 2.1.275 regression that made every request fail with a 400 when ANTHROPIC_BASE_URL points at a proxy or gateway.
Who is it for?
developers running Claude Code on more than one machine
Frequently asked questions
- Should I upgrade past Claude Code 2.1.275?
- Yes — install Claude Code 2.1.276 instead. In 2.1.275, every request failed with a 400 error naming `Input tag 'advisor_20260301'` whenever ANTHROPIC_BASE_URL pointed at a proxy or gateway. Anthropic shipped 2.1.276 a few hours later with that single fix, so anyone routing Claude Code through a gateway should skip straight to it.
- How do I stop Claude Code from syncing my claude.ai skills and plugins?
- Claude Code reads two settings for this. Set syncClaudeAiSkills to false to stop skills syncing and syncClaudeAiPlugins to false for plugins; either can be disabled on its own. The sync only applies to terminal sessions signed in with a claude.ai account, so sessions using an API key are unaffected.
- What does the send-now key in Claude Code do?
- The send-now key — ctrl+enter, or ctrl+x followed by ctrl+s — interrupts the turn Claude Code is currently running and delivers every message you have queued at once, rather than feeding them one at a time. Messages you have sent and messages still waiting are shown in grey until the model receives them, so you can see what has actually gone through.
- Does Claude Code 2.1.275 change anything for telemetry?
- Claude Code 2.1.275 adds a startup warning when a configured otelHeadersHelper fails. Before this, a broken helper left the session exporting no telemetry at all with nothing to indicate it, so a team could believe it was collecting traces that never arrived. The warning surfaces the failure as the session begins.
Try it
npm i -g @anthropic-ai/claude-code@2.1.276