Claude Code shipped four releases in the first week of July: 2.1.198, 2.1.199, 2.1.200, and 2.1.201. Most changes are reliability fixes for background agents — but one detail in 2.1.200 deserves special attention.
Manual Mode Becomes the Default
The former ‘default’ permission mode is now officially called ‘Manual’. Sounds like a rename, but it’s a statement: Anthropic is making clear that Claude Code should ask for permission before every write action by default. Auto mode remains available, but it’s deliberately not the default.
For teams rolling out Claude Code, this matters — the default determines how new users experience the tool.
Claude in Chrome Is GA
The biggest feature news of the week came with 2.1.198: Claude in Chrome is now generally available. The browser extension that lets Claude read, click, and navigate websites had been in beta for months. It’s now available to all users on paid plans.
Background Agents: Dozens of Fixes
The focus of this release week was clearly on background agent stability. Some highlights:
Sessions interrupted by sleep/wake or stalls no longer stop silently. A stale daemon lock whose PID the OS reused no longer blocks new agents. Subagents cut off by a rate limit now return their partial work to the parent agent instead of failing silently.
Particularly useful: background agents launched from claude agents now automatically commit, push, and open a draft PR when they finish code work in a worktree.
SSL and Network Resilience
Version 2.1.199 brings better error handling for SSL certificate issues (TLS-inspecting proxies, missing NODE_EXTRA_CA_CERTS) — these errors are now reported immediately with a fix hint instead of burning through retries. Brief network drops mid-response are also caught and retried with backoff.
Takeaway
Four releases in one week shows how fast Claude Code is evolving. The focus on background agent reliability makes sense — if you’re running agents in the background, you need to trust that they won’t silently stop working.
Sources: