Dev tools

Claude Code 2.1.275 syncs skills from claude.ai to the terminal and stops plugin install scripts from running

September 17, 2026 at 4:30 PM PT

The anthropics/claude-code repository on GitHub

Image: GitHub

Why it mattersThe plugin change removes a way that installing a marketplace plugin could run arbitrary code on a developer's machine, and the claude.ai sync means the set of skills a team uses in the browser now travels with the terminal without a second setup step.

Anthropic released Claude Code v2.1.275 on 17 September at 22:33 UTC. The changelog is one of the longer ones in the 2.1 series and it carries two changes that reach anyone using the CLI or the VS Code extension, plus a set of security fixes.

What is new for someone using it today

Terminal sessions signed in with a claude.ai account now receive the skills and plugins that account has enabled. Anthropic says the sync happens automatically, and a team that wants to opt out can set syncClaudeAiSkills: false or syncClaudeAiPlugins: false. That closes the split where a skill added in the browser was not available in the terminal until a person set it up a second time.

A new keystroke, Ctrl+Enter (or Ctrl+X Ctrl+S), sends every queued message at once and interrupts the current turn. The changelog notes that sent and queued messages show in gray until the model receives them, so it is clear which messages made it through.

The plugin installer gained a --marketplace <source> flag: running /plugin install <plugin> --marketplace <source> now offers to add the marketplace before installing the plugin, so a marketplace and its first plugin can be added in one step instead of two.

The security-relevant changes

The one worth pausing on is a change to how plugins are fetched from npm. Anthropic writes that plugin installs from an npm source now run through npm pack --ignore-scripts and are integrity-verified, so a package's install scripts no longer run when Claude Code adds it as a plugin. Before this release, installing a marketplace plugin from npm meant running whatever lifecycle scripts that package shipped, on the developer's machine. That is now off by default.

Two other fixes belong in the same class: plugin and marketplace messages, logs, and claude plugin marketplace list used to show a password or token if one had been typed into a git, ssh, or marketplace URL, and now redact it. Sandboxed Bash commands on Linux running under zsh had been reporting exit code 0 for failed commands, which meant scripts that keyed off exit codes silently ran the wrong branch.

The rest

The release also fixes a memory-file age note that was changing between requests and blowing the prompt cache after a compaction, several session-resume and transcript failures on malformed history entries, and one that made VS Code renaming of a running session revert to the generated name. A separate stream of items covers Claude Tag, Anthropic's Slack integration, including new Amazon CloudWatch, SNS, Google Cloud Monitoring, and Datadog presets for its access-bundle credentials tab.

Source

Source: Anthropic

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

More from AI News

headcount packages 146 Claude Code skills into 16 department plugins, hits 846 stars in three days

A Claude Code project called headcount packs 146 skills into 16 department plugins, each independently installable, and picked up 846 stars in the three days since its public creation.

Source: GitHubOpen source

Anthropic redesigns Claude Code Projects with a coordinator that splits a goal across parallel threads, and warns each thread counts as its own session

The New Stack reports that Anthropic is rolling out a redesigned Claude Code Projects in beta on Pro and Max plans, with a coordinator that breaks an engineering goal into parallel threads, shared memory across them, and the warning that each thread counts as a full Claude Code session against usage limits.

Source: PressDev tools

Anthropic banned a Claude Code account 15 minutes after it was pointed at OpenAI's GPT-5.6 Sol through a proxy

The Information reports that Anthropic shut down developer Alex Getman's Claude Code account 15 minutes after he used a proxy to route it to OpenAI's GPT-5.6 Sol, and his posts about the ban drew a million views within hours.

Source: PressDev tools