Claude Code 2.1.271 adds a fast mode for Remote sessions, per-command allowed_domains restrictions for Bash, PowerShell, and Monitor in auto mode, and a claude self-hosted-runner --drain-marker-file option for host draining, alongside fixes for stale cached org policy, enterprise managed-mcp.json being ignored, and cloud sessions rejecting subagent tool calls.
why it mattersThe auto-mode domain restrictions and the enterprise MCP fix close real gaps for teams running Claude Code at scale.
Anthropic (Claude Code changelog)
Cursor's new Projects feature runs a coordinator agent that delegates to multiple subagents rather than writing code itself, executing on its own machine in the cloud so work continues after the laptop closes. It can also be configured to watch Slack channels, follow pull requests, or run on a schedule without a manual prompt.
why it mattersPersistent, event-triggered cloud agents push Cursor further from an IDE plugin toward an always-on background worker.
Cursor
Manifold Security disclosed eight flaws across seven command-line coding agents in which a repository's own .git/config can specify a command that runs on the developer's machine outside the agent's sandbox and without an approval prompt. Claude Code, Cursor, Codex, and Goose have shipped fixes, but Hermes Agent, Qwen Code, Grok Build, and a second path in Claude Code remained exploitable as of the researchers' September 1 retest.
why it mattersTeams that handle repos as file copies rather than git clone (shared drives, zip files, USB) should check their agent's patch status now.
The Hacker News