Recent releases from organizations and products in the AI category.
Projects without a CLAUDE.md now fall back to AGENTS.md, configurable under Project instructions in /config (not yet on Bedrock, Vertex or Foundry). Fixed claude -p and Agent SDK sessions hanging with no result after an internal error, which now report the error and exit with code 1, plus a fix for conversations failing with "text content blocks must be non-empty" after --resume.
Fixed a 2.1.275 regression where every request failed with "400 ... Input tag 'advisor_20260301'" when ANTHROPIC_BASE_URL points at a proxy or gateway.
Fixed a restored memory file's age note changing between requests after compaction or resume, which caused prompt cache misses, and stopped plugin and marketplace messages, logs, and listings from showing passwords or tokens stored in git, ssh, or marketplace URLs. Also added a send-now key (ctrl+enter) that interrupts the current turn and sends all queued messages, and syncing of claude.ai skills and plugins to terminal sessions.
Projects are no longer a folder of chats but one conversation where each request spawns a thread that Claude works on in the cloud and reports back as it goes, including what is ready for review and what is waiting on you. Available in beta for select Pro and Max users, rolling out in stages across Claude Code, chat, and Cowork, with existing projects unchanged for now.
Squashed some bugs and improved the overall experience. Yours, Claude
Voice calls with Devin now run on a live speech model, with call controls and Slack thread call attribution. Devin Review always checks for security bugs and the Security scan toggle was removed, GitLab connections gained an Automation scope setting for public repos, MCP credentials moved to each server's page, and the v3 blueprint API can manage blueprints from git or Devin's database.
Sessions stuck endlessly retrying "unexpected tool_use_id" 400 errors now self-heal corrupted transcripts where possible, and otherwise end the loop with a clear error and a /rewind hint. Also adds CLAUDE_CODE_MCP_STARTUP_WAIT_MS to bound how long the first non-interactive turn waits for connecting MCP servers, a critical memory-usage warning, and dozens of fixes across MCP, the Claude apps gateway, plugins, and permissions.
Harvey Agent can now construct and maintain structured review tables directly through agent actions.
