Claude Code 2.1.282 ends the web-search 400 that broke whole chats
Also caps prose on wide terminals, makes --resume far less lossy, and ignores telemetry vars in project settings — plus Gemini Live Avatar hits GA.
Copy markdown
The web-search 400 that broke whole conversations is fixed
If a chat's history held web search results the API couldn't decrypt — say, a turn answered through a third-party gateway — every following request died with a 400. 2.1.282 fixes it, so those conversations work again instead of being permanently stuck.
Cap Claude's prose width on wide terminals
The new maxProseWidth setting limits how wide prose runs on big monitors while tables and code blocks still use the full width. If you work on an ultrawide, you finally get readable line lengths without shrinking the window.
--continue and --resume stop silently degrading sessions
More cases where resumed sessions re-sent earlier messages in altered form — making the API drop Claude's prior reasoning — are fixed, plus a retry-once path for the 'redacted_thinking' 400 that killed every turn and faster resume for very large, never-compacted sessions. Picking up a long session is far less likely to quietly wreck it.
A cloned repo can't quietly flip on your telemetry
Project and local settings that switch on OpenTelemetry export, set its endpoint, or capture content (CLAUDE_CODE_ENABLE_TELEMETRY, OTEL_LOG_*) are now ignored, with a startup notice plus /status and claude doctor listing exactly what was dropped. One less thing a repo you pull can do behind your back.
anthropic-skills and claude-ai names locked to synced skills
Skill(anthropic-skills:*) and Skill(claude-ai:*) allow-rules now cover only skills synced from claude.ai, and plugins, commands, workflows, or MCP servers squatting those names no longer load or list. If you named a local plugin or MCP server that, rename it to get its skills and prompts back.
Gemini 3.8 Live Avatar hits GA — talking video agents via the Live API
Google shipped Gemini 3.8 Live with Live Avatar to general availability: lip-synced conversational video, native speech-to-speech with interruption recovery, background tool calls, and 97 languages with auto-detection, all via the Live API and Agent Development Kit. You can build a real-time on-screen agent that watches a shared screen or camera and talks back — no separate avatar vendor.