Cursor Composer 2.5 and Claude Code Agent Fixes Lead Weekly Updates

What shipped

From 15 to 22 May, three major coding agents pushed incremental releases that focus on reliability and task persistence rather than broad new features. Cursor emphasized longer-running tasks inside its IDE. Claude Code refined background sessions and command options in the terminal. OpenAI Codex added practical controls for remote use and goals in its CLI.

Cursor

Cursor shipped four updates centered on its Composer tool and new automation options. The releases target sustained performance on extended tasks and direct links to project management systems like Jira. These changes give users concrete ways to trigger code changes from tickets or manage automations across multiple codebases.

• •Composer 2.5 Cursor released Composer 2.5 with stronger instruction following and better results on long tasks. The update adds standard and fast pricing tiers plus double usage limits for the first week, helping users complete multi-hour coding jobs without repeated restarts.
• •Jira Integration Cursor added a direct link to Jira that lets users assign tasks or start agents from ticket comments. The agent pulls context from the ticket to fix bugs or write tests then posts the pull request link back, reducing manual handoffs for teams that track work in Jira.
• •Automations in Agents Window Cursor moved automation management into the Agents Window and added support for multi-repo and no-repo templates. Users can now set up Slack or analytics workflows without leaving the main interface, cutting setup time for routine cross-tool tasks.
• •Composer 2.5 Availability Cursor made Composer 2.5 generally available with gains in intelligence and task endurance. The release includes revised token pricing and extra usage in week one, giving frequent users a low-risk window to test longer agent runs.

Claude Code

Claude Code delivered ten updates across two weeks that mostly refine session handling and command behavior. The changes improve background task management, add JSON export for scripting, and fix several tool regressions. These tweaks matter most for users who keep multiple agents running or integrate the tool into existing terminal workflows.

• •Model Command Scope Claude Code limited the /model command to the current session only. Users press d in the picker to set a permanent default, preventing accidental model switches across separate tasks.
• •Background Session Resume Claude Code added /resume support for background sessions started with claude --bg. These sessions now appear with a bg label and can be continued like normal ones, helping users pick up long-running work after a restart.
• •OpenTelemetry Attributes Claude Code added agent_id and parent_agent_id fields to its tool spans. The change improves trace nesting for background agents and makes debugging multi-agent runs easier in observability tools.
• •JSON Agent Listing Claude Code introduced claude agents --json to export live sessions. The flag supports scripting with tools like tmux or status bars without parsing plain text output.
• •Code Review Command Claude Code renamed /simplify to /code-review and added effort levels plus GitHub comment output. Users can now request targeted bug reports that post directly as pull request comments.
• •Pinned Session Stability Claude Code made pinned background sessions survive idle periods and restart after updates. They are only removed under memory pressure after non-pinned sessions, giving priority to important long tasks.
• •Bash Exit Code Fix Claude Code corrected a regression that caused every Bash command to return exit code 127. The fix restores normal tool behavior for users affected since version 2.1.147.
• •Version 2.1.144 Release Claude Code v2.1.144 added background resume support and session duration notices. It also reduced rendering issues inside VS Code terminals for smoother daily use.
• •Version 2.1.145 Release Claude Code v2.1.145 added JSON session export and plugin previews before install. Permission prompts for Bash commands were also hardened against bypass attempts.
• •Version 2.1.146 Release Claude Code v2.1.146 introduced the /code-review command and fixed PowerShell tool failures on Windows. Auto-update reliability and diff rendering speed both improved in this build.

OpenAI Codex

OpenAI Codex released four CLI updates that activate Goal Mode by default and add remote control features. The changes focus on macOS screenshot capture, permission profiles, and TUI responsiveness. These updates give users clearer ways to run persistent tasks and share plugins within business accounts.

• •Codex CLI 0.133.0 OpenAI Codex shipped version 0.133.0 with Goal Mode enabled by default and dedicated storage. Remote control now runs as a foreground command while plugin discovery and several TUI startup bugs were addressed.
• •Appshots and Goal Mode OpenAI Codex introduced Appshots on macOS that send active window screenshots to the agent. Goal Mode graduated from experimental, remote computer use for locked Macs was added, and ChatGPT Business plugin sharing became available.
• •Codex CLI 0.131.0 OpenAI Codex released version 0.131.0 with richer TUI session controls and responsive tables. File and plugin search now use unified @ mentions, and a new codex doctor diagnostic command helps troubleshoot setups.
• •Codex CLI 0.132.0 OpenAI Codex released version 0.132.0 with native Python SDK authentication and faster TUI startup. Output schemas for resumed automations and several Windows installation fixes were included.

What this means for you

What to watch next

Watch for follow-up fixes to the new Composer pricing tiers and any Claude Code plugin store rollout. Track whether Codex expands Appshots beyond macOS or adds similar capture to its CLI.