· Updated

Claude Code Just Patched the NTLM Credential Leak — And 50 Other Fixes You'll Actually Feel

#claude-code#release#security#coding-agents

Claude Code v2.1.234 dropped 14 hours ago. The headline is security — Windows NT-namespace paths (\\??\\) are now rejected across remote file reads, session restore, CLAUDE.md includes, workflow scripts, and file uploads. This closes the NTLM credential-leak vector that’s been haunting Windows users.

But scroll past the security bullet and you’ll find 50+ fixes that change daily workflow. Here’s what matters.

The Security Fix That Matters

Windows users: any pre-approval file access that touched \\??\\ paths could leak NTLM hashes. That’s now blocked at the gate. If you’re on Windows and use --dangerously-skip-permissions, this alone justifies the update.

GitLab Merge Requests in Your Statusline

Repos with a GitLab remote and authenticated glab CLI now show MR !N with draft/pending/green states right in the footer. No more context-switching to the browser to check review status.

Auto-Resume When Your Usage Limit Resets

Hit your claude.ai limit? Claude Code now continues your session automatically when it resets. Turn it off in /config → “Continue automatically at usage limit” if you prefer manual control.

Your Prompts Now Render Markdown

This one’s subtle but huge: your own prompts now render markdown (highlighted code blocks, inline code, lists) the same way replies do. Your transcript is finally readable as documentation, not a raw log.

Session Titles That Make Sense

Auto-generated titles shifted from “Fix the login button on mobile” → “Login button bug”. Short. Specific. Scannable in the session list.

Permissions Mid-Turn

/permissions can now open while Claude is working. Rule changes apply to the rest of the current turn. Same for /add-dir, /autocompact, /theme, /help, /config, /advisor — all open mid-turn in the fullscreen TUI.

The Fixes You’ll Notice Daily

Annoyance Fixed
Auto mode re-checking sandboxed commands after compaction
Session-scoped permission answers dropped during subagent prompts
Crash on malformed API responses (third-party gateways)
Markdown rendering crawl on weird Unicode
SendMessage rejecting recipients at 200-char session name cap
MCP diagnostics printing resolved secrets
Modal text losing characters when copied in fullscreen
Horizontal rules running into next line
Shell commands splitting into multiple rows with todo interleaving
Dialogs dismissed when ! shell command finished
Queued ! command sent as plain text after up-arrow edit
Queued messages reappearing in prompt history
Accepting fullscreen renderer prompt restarting without permissions
/tui dropping --allowed-tools/--disallowed-tools on restart
Trust prompts missing repo-wide scope warning
IDE diff tab closing during re-prompt answering wrong prompt
Files in Remote Control sessions not uploading to phone/web
Stale-token reminder leaking into resumed turn after /login
Permission previews relaying to unadmitted channel servers
Credential masking hiding commands/paths from approver
Provider tokens not masking when followed by shell delimiters
Cross-session messages silently dropped in Desktop/VS Code
Remote Control: account switch now stops session in seconds
Remote Control: phone/claude.ai/code now sync permission mode & model
Effort picks on phone now apply to terminal/Desktop sessions
SendMessage/ListAgents now admit when session list too long
Expired credentials now point to /login
Built-in claude-api skill context cost: 200k+ → ~25k tokens
Redundant “Allowed by auto mode classifier” line removed
“Default teammate model” setting removed (uses leader’s model)
Elapsed-time counter dimmed on running tool header
Background task notifications now in <system-reminder> tags
Mantle skips admin-pin probe when main-loop model picked
Windows startup no longer stalls on read-only ~/.claude.json

The CLAUDE_CODE_PROJECT_DIR_NAME Env Var

Hosts giving each session its own config directory can now set a short name for the per-project transcript directory. Small thing, matters at scale.

selection:clear Keybinding

Bind a key to clear in-app text selection. Works in the agents view too.

Updating

# Auto-updates on next launch, or force:
claude update

Full changelog: v2.1.233...v2.1.234

Why This Release Matters

v2.1.233 (two days ago) fixed the Windows git crash and interactive session crash. v2.1.234 plugs the NTLM hole and cleans up a year’s worth of paper cuts. The transcript markdown rendering alone changes how you review sessions. The GitLab MR badge changes how you stay in flow.

If you’re on Windows, this is mandatory. If you’re on Mac/Linux, the quality-of-life fixes stack up. Update.


Update 2026-08-18: Added v2.1.234 release details. Security fixes backported to v2.1.233 for Windows users.

FREE RESOURCE

Get the AI Agent Cheat Sheet

All 19 coding agents in one comparison table — pricing, features, benchmarks. Updated weekly. Delivered to your inbox.

k
kira_bug_hunter
Security & Bug Hunter
Former pen tester. Finds the bugs nobody wants to exist. Skeptical of everything, especially status indicators.

Related articles