On June 16, Anthropic shipped two Claude Code versions back to back: 2.1.178 and 2.1.179. No fireworks of new features — more like what you want to see after a stretch of outages: stability and polish.
2.1.178: security and cleanup
The bigger of the two releases turns several knobs. It brings stricter permission rules, support for nested .claude directories, and a safer auto mode. On top of that, improvements to /doctor, Remote Control, and the /bug command. And the usual but important part: crash fixes, fixes for authentication and WebSocket issues, and more reliable subagent execution.
That fits the broader arc of the last few weeks — nested sub-agents (up to five levels deep), finer model and region handling, and a searchable plugin manager. The direction is clear: Claude Code should stay dependable even in complex, multi-step setups.
2.1.179: the fix you want during outages
The smaller follow-up has exactly one highlight — and it lands: when the connection drops mid-stream, the partially received response now stays put instead of being replaced by a raw error. Anyone who battled Claude’s outages over the past few days knows how valuable that is. It also fixes mouse-wheel scrolling under WSL2.
My take
The timing is almost poetic: while the infrastructure groans under load, along comes the fix for interrupted responses. That’s exactly where good tooling shows — not in the big feature, but in making sure a dropped connection no longer throws away the whole job. The stricter permission rules and safer auto mode pay into the same account: trust. Which is precisely what Anthropic urgently needs elsewhere right now.
Sources: Releasebot: Claude Code, Claude Code Changelog