Claude Code adds phone push notifications and 67% faster /resume

• ClaudeDevs' push notification announcement says Claude Code can now ping your phone when a long task finishes or when the agent needs input.
• According to ClaudeDevs' setup instructions, the flow is simple: pair the Claude mobile app with /remote-control, then enable Push when Claude decides in /config.
• In a separate update, ClaudeDevs' stability thread says the last four CLI releases shipped 50-plus fixes, including /resume running up to 67% faster on sessions larger than 40MB.
• ClaudeDevs' auth-fix post, ClaudeDevs' memory post, and ClaudeDevs' hang-fix post point to a quieter story underneath the feature news: fewer forced re-logins, less Linux RAM creep, and fewer stalls around WebFetch, proxies, and parallel requests.

You can open the remote control docs, skim the full Claude Code changelog, and watch ClaudeDevs' demo video show the handoff from terminal to phone. The most useful bit is how narrow the change is: Anthropic did not ship a full mobile coding client, it shipped a way to leave the terminal without losing the thread.

Claude Code can now send mobile alerts for two moments: when a long-running task finishes, and when Claude needs user input. The docs link in ClaudeDevs' post points to the new mobile push notifications section under remote control.

The setup sequence from ClaudeDevs' setup instructions is:

1. Install the Claude mobile app.
2. Run /remote-control to pair the app.
3. Open /config.
4. Enable Push when Claude decides.

That is a small feature with a very specific workflow payoff: long agentic runs no longer require babysitting the terminal just to catch the next prompt.

The bigger ship on the same day was reliability. According to ClaudeDevs' stability thread, the last four Claude Code CLI releases bundled more than 50 stability and performance fixes.

The headline number came from transcript-heavy projects. ClaudeDevs' big-sessions post says /resume is now up to 67% faster on sessions above 40MB, /branch no longer carries a 50MB cap, and session loading is more resilient after unclean shutdowns.

Those fixes target a real pain point in daily use: once a codebase conversation gets large, the friction is usually not model quality, it is state management.

The bug list breaks into three practical buckets:

• Auth: ClaudeDevs' auth-fix post says Anthropic shipped about a dozen login fixes, including hardened macOS keychain handling and reactive token refresh on 401 errors.
• Memory and startup: ClaudeDevs' memory post says Linux sessions with background tasks were leaking RAM through a re-render loop, and MCP servers now connect concurrently so startup does not wait on the slowest server.
• Hangs and crashes: ClaudeDevs' hang-fix post says WebFetch no longer stalls on huge pages, proxy 204 responses no longer crash sessions, and an intermittent API 400 caused by a cache-control race under parallel requests was closed.

The official changelog, linked in ClaudeDevs' changelog pointer, is where Anthropic is parking the running ledger for these CLI-level fixes.

One of the least flashy fixes is also one of the most terminal-native. In the fifth post of the thread quoted inside ClaudeDevs' thread on paste and scroll, ClaudeDevs says pasting from Windows or Xcode now preserves line breaks, scrollback survives resize events without duplicating itself, and fullscreen scrolling no longer snaps back down.

That thread lines up with what trq212 on renderer work said later the same day: the team is pushing on a no-flicker renderer hard enough that it wants to make it the default. trq212 on big file writes added one more concrete bug in flight, a case where Claude Code can appear to hang during big file writes.

That last detail is new information, and it says a lot about where Claude Code is maturing next: not headline model changes, just fewer papercuts between prompt, terminal, and long-running work.