Claude Code 2.1.126: Gateway Models, Project Purge, OAuth Fixes

TL;DR

• Model picker now pulls from gateway endpoints when using Anthropic-compatible gateways
• New claude project purge command to nuke all project state in one go
• OAuth login now works in WSL2, SSH, containers, and slow connections
• Major fixes for permissions, image handling, stream timeouts, and Windows shell detection

New

• Gateway model listing — `/model` picker now reads from your gateway's `/v1/models` endpoint when `ANTHROPIC_BASE_URL` points at an Anthropic-compatible gateway
• claude project purge — Delete all Claude Code state for a project (transcripts, tasks, file history, config) with `--dry-run`, `-y/--yes`, `-i/--interactive`, and `--all` flags
• OAuth code paste fallback — `claude auth login` now accepts OAuth codes pasted into the terminal when browser callback can't reach localhost (fixes WSL2, SSH, container auth)
• Slash command telemetry — `claude_code.skill_activated` OpenTelemetry event now fires for user-typed slash commands with new `invocation_trigger` attribute
• Permission bypass flag — `--dangerously-skip-permissions` now bypasses prompts for writes to `.claude/`, `.git/`, `.vscode/`, and shell config files (catastrophic commands still prompt)
• Windows PowerShell detection — Now detects PowerShell 7 from Microsoft Store, MSI without PATH, and .NET global tool; treats PowerShell as primary shell when enabled
• Permission stall indicator — Auto mode spinner turns red when a permission check stalls instead of appearing to run normally

Fixed

• OAuth login timeout on slow/proxied connections, IPv6-only devcontainers, and localhost-unreachable scenarios
• Image paste breaking sessions when larger than 2000px — now auto-downscales on paste and removes oversized images from history
• Login screen showing for "OAuth not allowed for organization" errors — now displays admin contact guidance
• Rare race condition where concurrent credential writes could clear valid OAuth refresh tokens
• API retry countdown stuck at "0s" instead of counting between attempts
• "Stream idle timeout" error after waking Mac from sleep mid-request
• Background and remote sessions falsely aborting with "Stream idle timeout" during long model thinking pauses
• Assistant finishing thinking but showing no output after empty turns
• Overly fast trackpad scrolling in Cursor and VS Code 1.92–1.104 integrated terminals
• Claude.ai MCP connectors suppressed by manual servers stuck in needs-auth state
• Japanese/Korean/Chinese text rendering as garbled on Windows in no-flicker mode
• Ctrl+L clearing prompt input — now only forces screen redraw per readline behavior
• Deferred tools (WebSearch, WebFetch) unavailable to skills with `context: fork` on first turn
• Plan-mode tools unavailable in interactive sessions launched with `--channels`
• Blank remote-session transcripts when certain messaging tools unavailable
• /plugin Uninstall reporting "Enabled" instead of "Uninstalled"
• /remote-control retries appearing stuck on "connecting…" — each retry now shows result
• Remote Control failure notifications not showing error reason for initial connection failures
• Windows clipboard writes exposing copied content in process command-line arguments visible to EDR/SIEM; also fixes >22KB selections not reaching clipboard
• PowerShell tool mis-flagging bare `--` as `--%` stop-parsing token
• Agent SDK hang when model emits malformed tool name in parallel tool call batch
• Read tool malware-assessment reminder causing spurious refusals on legacy models
• Security: `allowManagedDomainsOnly` / `allowManagedReadPathsOnly` being ignored when higher-priority managed-settings source lacked `sandbox` block
• Host-managed deployments auto-disabling analytics on Bedrock/Vertex/Foundry

Update: npm install -g @anthropic-ai/claude-code or check your IDE's extension marketplace for the latest version.

Source: Claude Code