🧭 Claude Code Projects Redesigned: A Coordinator Now Directs Parallel Agent Threads with Shared Memory

Anthropic has shipped a complete redesign of Claude Code Projects, now available in beta for Pro and Max subscribers who use cloud sessions. The core change: a Project is no longer just a shared context folder — it is now a coordinator that can spawn and direct multiple parallel Claude Code cloud sessions (threads), each working on its own Git branch, while sharing a common memory store and file library. You describe the goal at a high level; Claude scopes the work, delegates to threads, monitors progress, resolves merge overlaps, and assembles the result.

How the new Projects model works

• Coordinator layer: One Claude instance holds the project goal, breaks it into sub-tasks, and assigns each to a parallel thread. The coordinator reviews thread outputs and can redirect or restart threads that diverge.
• Parallel threads = parallel cloud sessions: Each thread is a full Claude Code cloud session with its own working branch. Threads can open pull requests independently; the coordinator tracks which need to merge first to avoid conflicts.
• Shared memory + library: Threads draw on the same project memory and a shared library of uploaded files and generated artifacts — so context built in one thread (e.g. a profiling report) is immediately available to another (e.g. an optimisation thread).
• Async, phone-friendly oversight: Because the work happens in cloud sessions, you can step away and steer progress later — including from a mobile browser — without keeping your laptop open.

Example: retiring a deprecated API endpoint across three repos

Connect your API, web, and mobile repositories to a project. Set the goal: "Retire the v1 checkout endpoint." Claude creates one thread per repository to migrate callers, run test suites, and open pull requests. The coordinator identifies that the API repo PR must merge before the mobile PR can unblock, and surfaces that ordering in its progress summary.

🧭 Claude Code v2.1.274–276: Critical-Memory Warning, Send-Now Key, and Claude.ai Skills Sync to Terminal

Three rapid-fire releases landed this week, each fixing a real developer pain point. The headline change from v2.1.274 (108 CLI changes) is a visible critical-memory warning that surfaces when the session is approaching its RAM limit — showing actionable steps to free memory or restart safely before Claude crashes mid-task. v2.1.275 added the most-requested terminal shortcut: a send-now key (ctrl+enter) that interrupts the current Claude turn and immediately delivers all queued messages. And both v2.1.275 and the same-day v2.1.276 hotfix addressed a regression in proxy and gateway base URL handling.

Highlights across all three releases

• Critical-memory warning (v2.1.274): When a session's memory usage is critical, Claude now shows a banner with specific steps (clearing caches, reducing active tool results, restarting the session) before an out-of-memory crash. The warning also includes a CLAUDE_CODE_MCP_STARTUP_WAIT_MS env var to bound how long the first non-interactive turn waits for MCP server connections — useful for slow-starting remote MCP servers.
• AGENTS.md support (v2.1.274): Claude Code now reads AGENTS.md files at project root and sub-directory levels, giving teams a dedicated file to declare agent conventions, tool permissions, and skill boundaries alongside CLAUDE.md.
• Send-now key (v2.1.275): ctrl+enter (or ctrl+x ctrl+s) interrupts the current turn and sends all queued input at once. This is the same send-immediately pattern Claude.ai chat has had for months — now in terminal. Also configurable to ctrl+s for single-key muscle memory.
• Claude.ai skills and plugins sync to terminal (v2.1.275): The skills and plugins you have enabled on your claude.ai account now automatically sync to terminal sessions. No more re-enabling your preferred tool set after switching from the web UI.
• Proxy/gateway 400 fix (v2.1.276): Hotfix for every request failing with a 400 Input tag 'advisor_20260301' error when ANTHROPIC_BASE_URL points at a proxy or gateway — a regression introduced in v2.1.275 that affected any team routing Claude Code through an internal LLM gateway.