AI pair programmers and autonomous coding agents — IDE-integrated assistants and cloud agents that read, edit, and ship code.
This week: Security hardening dominates as Claude Code patches permission bypasses →
Ultra sessions now notify you when the lead action model degrades or recovers. PR diff files are virtualized for faster session switching, and lightboxes, attachment previews, and CI failure shortcuts see improvements.
Claude Code fixed a sandbox bypass and added a default model variable, while prompt caching now works through gateways.
Fixed prompt caching for sessions using an LLM gateway or custom base URL. Added a built-in "Concise" output style that leads with results and skips preamble, selectable under Output style in /config.
New ANTHROPIC_DEFAULT_MODEL env var sets a session's starting model while preserving /model overrides, and cross-session SendMessage gains one-shot idle notifications. Also fixed sandbox read-deny wildcards being bypassed via renaming, plus clipboard/renderer/crash regressions across fullscreen, /model picker, and remote control.
Claude Code fixed a premature-approval bug in its permission prompts, invalidation on language-server reconnects, and wrapped in a spellcheck option and UI refinements.
Cloud agents can now subscribe to PRs, Slack threads, or scheduled tasks, wake on events, and drive work to completion. Skills can be pinned as custom modes, subagents can run on isolated VMs, and /goal gives agents long-lived objectives; follow-up messages now wait for the next tool call instead of interrupting.
Fixed Shift+Tab in permission comment fields no longer approving edits prematurely, and fixed whole-prompt-cache invalidation on language server reconnect. Also improved memory/CPU use for background cloud sessions, hardened permission dialogs, and added an optional spellcheck setting.
Claude Code closed the remaining pre-approval file-access NTLM credential-leak vector by rejecting Windows NT-namespace paths, and added automatic session resume after claude.ai usage limits reset.
/export, either to the clipboard or a new file. ([#37358](https://github.com/o...Added a setting to open ChatGPT directly in Codex Remote on launch, plus support for standard MCP forms and editable message approvals. Fixed large task responses failing to load, tasks disappearing after idle or reconnect, and improved diff review stability and voice reliability.
Remote file reads, session restore, CLAUDE.md includes, and workflow scripts now reject Windows NT-namespace paths, fixing the remaining pre-approval file-access NTLM credential-leak vector. Sessions now resume automatically when a claude.ai usage limit resets, with a /config toggle to disable; 30+ other fixes cover shell-command queueing, permission previews, and Remote Control.
Cursor began rolling out Origin, its own code hosting platform in early beta on paid plans, with repos, pull requests, code browsing, and two-way GitHub sync, plus agent-powered code Q&A and Vercel, Depot, and Buildkite integrations.
Cursor now hosts code in early beta on paid plans, starting with repos, pull requests, code browsing, and GitHub sync. GitHub repos can sit alongside Cursor-hosted ones with two-way PR comment sync, and agents can answer questions about code, make changes, and push branches. Vercel, Depot, and Buildkite integrations are available now.
Devin introduced Coach prompt suggestions and default enterprise Local enablement, while Claude Code patched an NTLM credential leak and removed todo tools on newer models.
Devin Coach surfaces prompt-improvement suggestions in the session input box as you type. Devin Review skips re-reviewing unchanged PR diffs, sessions subscribe to Slack thread replies, and Devin Local is now enabled by default for enterprise customers, plus Teams, sidebar, and command palette polish.
xAI1Launchers can now override config settings via GROK_CONFIG and GROK_CONFIG_PATH environment variables, and tool calls no longer fail for the rest of a session if /dev/null is removed. Adds safeguards around automatic worktree reclamation, clearer hook-policy blocking messages, and better Arabic/Persian text reordering in the terminal UI.
Todo and task-tracking tools are no longer available on Opus 4.8, Sonnet 5, Fable 5, Mythos 5, and newer models — set CLAUDE_CODE_ENABLE_TODO_TOOLS=1 to bring them back. Also patched an NTLM credential-leak vector involving Windows NT device-prefix paths, fixed a 2.1.232 auto-mode approval regression, and reverted the 2.1.232 Bash permission changes for Cygwin symlinks and input redirections. Plus improved worktree GitLab MR support, memory limits for Bash, and faster self-hosted runner startups.
Claude Code turned on subagent forking by default, added cross-session messaging, and patched permission and OAuth security issues.
Subagent forking is now on by default: subagent_type: "fork" inherits the full conversation and prompt cache, and non-teammate agent spawns in interactive sessions run in the background by default. Also fixed PowerShell and Git Bash permission bypasses that could overwrite settings and redirect file access, plus hardening around nested git trust and cross-session messaging sockets.
Fixed MCP OAuth sign-in failing with a redirect URI mismatch for servers that use a pre-registered OAuth client, such as Slack.
Devin's Slack integration now auto-handles channel access for spawned sessions while side chats let you interrupt mid-session, and Claude Code patches streaming and crash bugs plus adds server-supplied hooks for self-hosted runners.
Side chats let you start a conversation anchored to any point in a session without interrupting Devin's main work. Slack improvements include requesting channel access directly from DMs, automatic read access for Slack-spawned sessions, and a machine-capacity notice. Automation conditions now support "not contains", "not starts with", and "not ends with" operators.
xAI1Web search can now be restricted to allowed or excluded domains via [toolset.web_search] in config.toml. A new Follow-up Behavior setting lets queued messages steer immediately, and queued messages no longer auto-submit while still being edited. Includes numerous stability fixes for sessions, subagents, and terminal input handling.
Computer History is a new opt-in feature in the ChatGPT desktop app on macOS that turns activity across apps and websites into memories and a timeline that ChatGPT and Codex can use. Choose which apps and websites contribute, pause collection, and review or delete your history at any time; available to ChatGPT Pro, Business, and Enterprise users, with Business and Enterprise admins needing to enable access first.
Cloud Agents now boot from Builds — ready-to-use environment copies Cursor prepares in the background — cutting time to first token by 3x. When a bad commit or dependency update breaks the environment, agents keep using the last successful build; a Builds tab in the dashboard shows status, logs, and commit SHAs.
Fixed long responses being printed twice and vanishing mid-stream, plus crashes from non-string tool arguments, narrow terminal rendering, and extended-length Windows paths. Added server-supplied hooks for self-hosted runners, SSE keepalive pings for gateway streams, and plugin marketplace command sources.
Claude Code patched a handful of correctness bugs, including /tui sessions reverting to an earlier model after /model changes, self-hosted runners mishandling failed checkout hooks, and interactive sessions that could stop redrawing after a rare layout error.
xAI1Subagent spawning is much faster when you have many sessions in ~/.grok, and TUI rendering now automatically matches high-refresh displays (120 Hz+). /session-info rows are now clickable to copy values, with a copy-all shortcut.