Claude Code 2.1.234: auto-continue, GitLab MR badge, NT path security
Adds automatic session continuation after claude.ai usage limit resets, a GitLab merge request badge, the CLAUDE_CODE_PROJECT_DIR_NAME environment variable, and a selection:clear keybinding action. Hardens NT-namespace path handling against NTLM credential-leak vectors across file read, session restore, CLAUDE.md includes, workflow scripts, and uploads. Remote Control now publishes permission mode and effort level to connected clients. Dozens of fixes across permissions, rendering, MCP diagnostics, shell queuing, and cross-session messaging.
PUBLISHED2026-08-17
OBSERVED2026-08-21
AGE8d
SOURCES1
- Auto-continue: sessions now resume automatically when a claude.ai usage limit resets; toggle in
/config. - GitLab MR badge: repos with a GitLab remote and authenticated
glabCLI show MR !N with draft/pending/green states in footer and statusline. - CLAUDE_CODE_PROJECT_DIR_NAME: optional env var for hosts that give each session its own config directory — choose a short name for the per-project transcript directory.
- selection:clear keybinding: bind a key to clear in-app text selection; works in the agents view.
- Security: remote file reads, session restore, CLAUDE.md includes, workflow scripts, and file uploads now reject Windows NT-namespace (\??\) paths, hardening against NTLM credential-leak vectors.
- Permission mode: Claude tells the model to use account email only for identification, not to send to unrelated services.
- Remote Control: effort picks from phone/web now apply to terminal/Desktop sessions; session publishes its effort level; permission mode published to clients; signing into a different claude.ai account stops the running session within seconds with a reason; files sent during Remote Control sessions from Desktop/VS Code now upload properly.
- Fixes: auto-mode re-checking network access after compaction; session-scoped permission answers dropped by background subagents; crash on malformed non-streaming API responses; slow markdown rendering on unusual Unicode sequences; secrets visible in MCP diagnostics; git remote host parsing with unusual userinfo;
strictKnownMarketplacesallowlists accepting SCP-style sources diffing from actual git host;SendMessagerejecting recipients near 200-char cap; fullscreen copy losing characters; horizontal rule bleeding into next line; consecutive shell commands splitting across interleaved todo updates;/permissionsdismissed during!commands; queued!commands sent as plain text after editing; queued messages reappearing in prompt history;!mode sticking after mid-turn submit; accepting fullscreen renderer prompt restarting without permission mode or flags;/tuidropping--allowed-tools/--disallowed-toolsrules; trust prompts omitting repo-wide scope warning; IDE diff tab answering re-prompts with stale input; credential masking on relayed permission previews hiding commands/paths; provider tokens masking even after shell delimiters; Claude Desktop cross-session messages silently dropped. - Transcript: user prompts now render markdown (highlighted code, inline code, lists) like replies.
- Error messages: "API returned empty/malformed response" now shows content type, body kind, size, and request ID.
- Auto-generated session titles: now read as short specific names rather than restating the request.
COMMUNITY
No curated reactions recorded for this event. Facts and takes are kept in separate layers — community context is added by hand, never blended into the record above.