Claude Code 2.1.234 picks up when the limit resets

Claude Code 2.1.234 resumes a stopped session on its own once a claude.ai usage limit resets, and cuts the context cost of the built-in claude-api skill from more than 200,000 tokens to roughly 25,000 by loading reference docs on demand. The release notes on Github list the automatic continuation as optional, disabled in /config under "Continue automatically at usage limit".

At a glance

• Remote file reads, session restore, CLAUDE.md includes, workflow scripts and file uploads now reject Windows NT-namespace paths written as \??\, hardening the pre-approval file accesses left over from the NTLM credential-leak fix.
• Repositories with a GitLab remote and an authenticated glab CLI show a merge request badge reading MR !N in the footer and statusline, with separate draft, pending and green states.
• Remote Control sessions now stop within seconds when the computer signs in to a different claude.ai account or organization, with a stated reason instead of an HTTP 404 hours later.

The through line of this build reads as session continuity: a run that stops at a usage limit, a restart triggered by a renderer prompt or a /tui switch, and a compacted conversation that starts re-checking sandboxed network access all cost the same thing, a working context that has to be rebuilt by hand. Trimming the built-in claude-api skill points the same way, at how much room a long run has left.

Automatic continuation at a usage limit reset is switched off in /config

Claude Code continues the session automatically when the limit resets unless the /config option is switched off. Dialogs also change: /permissions opens while Claude is working, with rule changes applying to the rest of the current turn, and /add-dir takes a path mid-turn. In the fullscreen TUI, /autocompact, /theme, /help, /config and /advisor also open mid-turn.

/goal now clears itself with a notice when a turn dies on an unrecoverable error such as revoked auth, an exhausted credit balance or a context overflow, and Claude checks in on background tasks that keep a goal waiting for 30 minutes or more; CLAUDE_CODE_GOAL_CHECKIN_MINUTES=0 opts out.

The transcript now renders markdown in the user's own prompts, with highlighted code blocks, inline code and lists, and auto-generated session titles read as short names such as "Login button bug" rather than sentences restating the request. The redundant "Allowed by auto mode classifier" line under every Agent tool call is gone.

MCP diagnostics now show the configured ${VAR} form instead of resolved secrets

Claude is now told to use the account email only to identify the user and not to send it to unrelated services unless asked. MCP diagnostics no longer print resolved secrets: scope-conflict warnings show the configured ${VAR} form, and connection-failure details show only the server origin.

strictKnownMarketplaces allowlists no longer accept SCP-style git marketplace sources whose host differs from the one git would actually connect to. Credential masking on relayed permission previews can no longer hide commands, paths or destinations from the approver, and oversized private-key blocks redact under full-strength redaction.

Permission previews relay only to channel servers admitted by the inbound trust gate, and a server's explicit permission-capability opt-out is honored. Provider API tokens that mask on permission previews now mask even when directly followed by shell delimiters. After /login while CLAUDE_CODE_OAUTH_TOKEN is set, the stale-token reminder appears only to the user rather than leaking into the automatically resumed turn.

Remote Control sessions publish permission mode and effort level to connected clients

Sessions started from Claude Code Desktop or VS Code keep phones and claude.ai/code updated on the session's permission mode as it changes, and claude.ai/code on the model. Effort picks made on a phone or on claude.ai/code now apply to terminal- and Desktop/VS Code-hosted sessions, and the session publishes its effort level to connected clients.

Files sent to the user during Remote Control sessions hosted by Claude Code Desktop or VS Code now upload and open on phone and web instead of showing an empty card. Claude Desktop inter-session messages are no longer silently dropped by the recipient session when cross-session messaging read as disabled, a case that left the sender's query thinking for many minutes.

SendMessage and ListAgents now report when the account's session list was too long to check completely instead of treating unseen sessions as absent, and SendMessage accepts a recipient copied from ListAgents when the session name hits the 200-character cap or is emoji-heavy.

What /tui will not restart

/tui previously dropped launch --allowed-tools and --disallowed-tools rules when it restarted; it now declines to switch and states the reason when the session carries restrictions a restart cannot carry over. Accepting the fullscreen renderer prompt no longer loses the session's permission mode, tool allow and deny rules, model or effort flags. The notes name no plan for carrying those restrictions through a restart.