Release archive

Claude Code v2.1.218: Background /code-review, Background Forked Skills, and Manual-Only /deep-research

Claude Code v2.1.218 runs /code-review as a background subagent, runs context: fork skills in the background by default, makes /deep-research manual-only, and fixes Windows paths, paste, and resume.

• 10 min read • By ClaudeWorld

Claude Code v2.1.218 is a broad release with 36 changelog entries. It moves more work into the background — /code-review now runs as a background subagent and skills with context: fork run in the background by default — while auto mode hands more checks to its classifier and --ax-screen-reader mode gains new announcements. The rest of the release fixes Windows path handling, multi-line paste, session resume, Bedrock setup, and more.

What’s New

/code-review Runs as a Background Subagent

/code-review now runs as a background subagent, so review work no longer fills your conversation, and it keeps stacked slash commands as its review target. In non-interactive sessions, /code-review ultra now launches the cloud review instead of silently running a local review.

Skills with context: fork Run in the Background

Skills with context: fork now run in the background by default. You can opt out per skill with background: false.

/deep-research Is Manual-Only

/deep-research now starts only when you invoke it manually; Claude no longer launches it on its own.

Auto Mode Defers More Checks to Its Classifier

In auto mode, the dangerous-rm, background-&, and suspicious-Windows-path checks no longer open permission dialogs; the auto-mode classifier adjudicates them instead. Plan mode with auto also no longer prompts for Bash commands the static analyzer can’t prove read-only — the classifier judges those too.

Screen-Reader Announcements for Deletions

In --ax-screen-reader mode, word and line deletions (Option+Delete, Ctrl+W, Cmd+Backspace, Ctrl+U, Ctrl+K) now announce the deleted text.

Frontmatter Rules for Skills, Plugins, and Agents

Skill and plugin frontmatter booleans now accept yes/no/on/off/1/0 (case-insensitive) alongside true/false. Agent markdown files now reject agent names containing :, which is reserved for plugin namespacing, and hooks in agent frontmatter now require the agent file’s own folder to have accepted workspace trust.

Fixes and Improvements

Review commands

  • /ultrareview no longer fails on descriptive arguments like “review my auth changes”; it now reviews your current branch and applies the text as a note to the findings.
  • /ultrareview gives better error feedback, so Claude can correct an invalid argument instead of retrying it unchanged.

Windows

  • Windows paths with \u-prefixed segments (like C:\Users\unicorn) are no longer corrupted into CJK characters in tool inputs, which had made those files inaccessible.

Input and history

  • The left arrow key no longer discards the conversation with no undo: presses right after editing now ask to confirm, and Esc in the agent view returns to the conversation it backgrounded.
  • Multi-line paste no longer collapses into one line with j in place of newlines in terminals that encode pasted newlines as Ctrl+J.
  • Prompt history entries are no longer dropped or duplicated when history writes race or fail.

Accessibility

  • In --ax-screen-reader mode, VoiceOver now echoes a typed space at the end of the input instead of reading “new line”.
  • Plugin and settings panels now move the terminal cursor to the focused row, so screen readers and magnifiers can follow arrow-key navigation.

Sessions, context, and SDK

  • /context no longer reports stale pre-compact token usage after compacting from the message picker.
  • A resumed session no longer fails every turn, or crashes on resume, when its history holds a malformed delta attachment.
  • Fork-session lineage is no longer lost after compaction in headless and SDK sessions.
  • A retry loop that re-sent identical doomed requests after a context-overflow error with a large thinking budget is fixed, and Ctrl+B backgrounding now applies the same background-shell caps as other paths.
  • Remote sessions no longer keep sending heartbeats after their worker is replaced, which had left long-lived desktop and IDE processes retrying a rejected request every few seconds.

Bedrock and gateway

  • The Bedrock setup wizard no longer fails profile verification for assume-role profiles in partitioned AWS regions and on proxy-only networks.
  • Gateway spend metering now prices Bedrock application-inference-profile ARNs and other config-mapped upstream model IDs at the configured model’s rates.

MCP

  • The “N MCP servers need authentication” startup notice no longer over-counts claude.ai connectors that aren’t connected in claude.ai.

Trust and managed settings

  • Trust dialogs now name the repository root the grant covers.
  • Benign feature and cost toggles in server-managed settings no longer trigger the settings-approval prompt.

Models

  • Claude Code now announces when fast mode changes as a result of switching models via /config model=<x> or Remote Control.

This release contains 36 changes in total; see the Claude Code changelog for the complete list.

How to Update

  1. Check your installed version: claude --version
  2. Update in place: claude update
  3. Review your skills and agents: skills with context: fork now run in the background unless they set background: false, and agent names containing : are now rejected.