A developer used Opus 5.5 to build a fully Claude-native home automation platform, then shut down their Home Assistant and Apple Home setups entirely. The platform integrates Hue, Lutron, Aqara, and Ring devices.
Opus 5.5 fast mode is now generally available on the Claude Platform, with no waitlist. It runs up to 2.5x faster at $8/$40 per million tokens, is available on Microsoft Foundry (Azure), and Google Cloud support is coming soon. Fast mode in Claude Code remains separate and still uses extra usage credits.
The desktop app now launches sessions faster, makes it easier to split work across multiple sessions, and keeps each session's status visible at a glance. An Anthropic team member is amplifying the update with a straightforward endorsement.
Anthropic is partnering with Gray Swan AI on a $300,000 red teaming competition focused on prompt injection, alongside Meta, Google DeepMind, and other frontier labs. The move reflects Anthropic's view that solving prompt injection is increasingly urgent as AI agents take on more consequential tasks.
Claude Code now automatically compacts long idle sessions just before the prompt cache expires (around 1 hour). When you return, it reprocesses only a summary rather than the full conversation, reducing token usage.
Idle compaction causes Claude to continue from a summarized context rather than the full conversation. To disable it, set 'idleCompaction': false in settings.json. To adjust when it triggers, use CLAUDE_CODE_IDLE_COMPACT_MIN_TOKENS (default is 200k tokens).
Idle compaction, now on by default, means Claude resumes from a summary rather than the full conversation history. It can be disabled with 'idleCompaction': false in settings.json, and the token threshold (default 200k) is configurable via CLAUDE_CODE_IDLE_COMPACT_MIN_TOKENS.
Idle compaction is enabled by default in Claude Code. It can be turned off in settings, and the settings link was shared for reference.
A team member quoted the Claude Code UI simplification announcement with a playful 'Today's Weather: Partly Agentic' headline, riffing on the agentic nature of the new multi-session interface.
In response to a user struggling to track prior Claude Code sessions from the CLI, the suggestion is to use the Claude desktop app, which handles session history better. The --continue flag (equivalent to --resume for the last session in a folder) is the CLI workaround.
Official documentation for the new observability features is now published, covering OpenTelemetry gen_ai conventions, span metadata (duration, status, token counts, model, tool names), and full content export for prompts and tool results.
