Claude Code 2.1.288 adds draft recovery after Ctrl+C, a --max-findings flag for /code-review and a re-authenticate prompt for MCP servers, plus more than 60 fixes. It was published on npm on October 2, 2026 at 18:30 UTC (npm), and its changelog entry lists 89 items in all. The ones that change day-to-day use are below.

On this page

Which channel gets it

On the npm registry, 2.1.288 carries the latest tag, while stable still points to 2.1.285 (npm). Anthropic’s setup docs describe "latest" as the default auto-update channel and "stable" as a version “typically about one week old, skipping releases with major regressions”. The channel is set with autoUpdatesChannel:

{
  "autoUpdatesChannel": "stable"
}

New in 2.1.288

From the changelog:

  • Draft recovery after Ctrl+C. Pressing Up on an empty prompt brings back a prompt you cleared with Ctrl+C, including pasted text and images.
  • /code-review --max-findings. --max-findings <n>|all reports more or fewer findings than the usual limit. The choice is reused until you pass --max-findings default.
  • MCP scope re-authentication. When an MCP server asks for more OAuth scope during a tool call, Claude Code now shows a re-authenticate prompt.
  • Agents view navigation. Ctrl+F finds a session by name, Alt+↑/↓ jumps between groups, and both (plus rename) can be rebound in keybindings.json.
  • CLAUDE_CODE_DISABLE_STRUCTURED_OUTPUTS. A new environment variable that turns structured outputs off, added with a fix for session titles, memory recall and prompt hooks failing on Mantle or behind gateways that reject structured outputs.
  • Built-in gh api in cloud sessions whose image has no GitHub CLI.
  • $.ui.selection() for mods, which returns the text you last selected in fullscreen mode.

Behavior changes

  • The background command time limit now applies only in unattended sessions (-p, Agent SDK, CI, cloud). Terminal, desktop app and VS Code sessions have no limit.
  • claude project purge is now claude purge. The old name still works and prints a notice.
  • /autocompact saves the auto-compact window per model, so each model keeps its own setting when you switch.
  • In auto mode, a conversation that grows too long for the client-side safety classifier to review is now compacted instead of prompting for, or failing, every tool call.

Fixes worth knowing about

  • Mid-response API timeouts no longer fail the turn: non-interactive sessions and subagents continue from the partial response.
  • Long conversations no longer fail with “Prompt is too long” instead of auto-compacting when the last reply reported zero token usage.
  • Several --resume fixes: dropped files after a compaction, a missing last response, transcripts loaded cut short, and lost thinking for conversations started on 2.1.286 or earlier.
  • Path-scoped .claude/rules and nested CLAUDE.md files now load when Write or Edit touches a file in their scope; before, only Read loaded them.
  • LSP tool calls no longer hang when a language server stops responding; requests time out after 60 seconds, set per server with requestTimeout.
  • /login no longer reports “Login successful” when credentials could not be saved to secure storage.

A permissions fix

One fix concerns bypassPermissions mode: a dangerous rm (such as one on / or the home directory) inside a bash -c or sh -c script could run without a prompt in that mode or under a shell allow rule (changelog). Anthropic’s permission modes docs say rm and rmdir removals targeting a critical path are among the actions no mode auto-approves.

The docs carry two warnings about this mode, which --dangerously-skip-permissions also turns on: “Only use this mode in isolated environments like containers, VMs, or dev containers without internet access, where Claude Code cannot damage your host system”, and “bypassPermissions offers no protection against prompt injection or unintended actions.” For fewer prompts with background safety checks, the docs point to auto mode instead:

claude --permission-mode auto