Claude Code adds Opus 4.8 and background agents, Codex refines TUI, Cursor holds steady

What shipped

From 22 to 29 May the main activity came from terminal coding agents. Claude Code delivered the largest set of changes centered on Opus 4.8 and agent workflows. Codex shipped smaller TUI and diagnostics updates while Cursor stayed quiet.

Cursor

Cursor: No updates shipped this week so builders continue with the existing agentic coding features already in the IDE.

Claude Code

Claude Code delivered eleven updates that center on Opus 4.8 integration and day-to-day agent controls. The changes let users run background shell tasks, reload skills on the fly, and apply code-review fixes directly to files. Several releases also tighten tool permissions and add fallback model handling for steadier sessions.

• •Code review fixes The /code-review --fix command now writes suggested changes straight into the working tree so reviewers can test improvements without extra copy steps.
• •Git LFS skip A new skipLfs flag lets users bypass large file downloads during clones which speeds up work on repositories that do not need every binary asset.
• •Background shell sessions Commands started with ! or the --bg flag run detached so users can keep the main agent free while long tasks finish in the background.
• •Opus 4.8 support The agent now defaults to high-effort mode with Opus 4.8 and offers a cheaper fast mode for simpler tasks that still need strong reasoning.
• •Thinking block fix An earlier API error that broke Opus 4.8 thinking blocks has been removed so sessions no longer fail mid-run.
• •Skill reload command The new /reload-skills instruction rescans skill folders without restarting the agent which keeps custom behaviors current during long projects.
• •Update notices One-time notices now appear when global npm updates fail so users know when their local tools are out of date.
• •Fallback model switch If the main model fails the agent automatically continues with the configured fallback model for the rest of the session.
• •Tool restrictions Skills and slash commands can now list disallowed tools in frontmatter so the agent stays within the exact capabilities the user wants.
• •Automatic review fixes The updated /code-review flow suggests reuse and simplification then applies the edits so teams spend less time on manual cleanup.
• •Terminal variables Status line commands now receive current COLUMNS and LINES values which improves layout in narrow or resized terminal windows.

OpenAI Codex

Codex released five targeted fixes that improve the terminal interface and diagnostics. The updates address rendering bugs, add named permission profiles, and expand the doctor command with deeper environment details.

• •TUI rendering fixes Markdown tables and multiline lists now display cleanly and stderr no longer corrupts output on macOS or Zellij.
• •Named permission profiles The /permissions command lists saved profiles so users can switch between preset access rules without typing long flags each time.
• •Vim mode upgrades Text-object editing and better word navigation arrived in Vim mode along with a configurable interrupt binding for faster terminal edits.
• •Codex doctor expansion The doctor command now reports Git state, thread counts, and app-server details which shortens support turnaround for complex setups.
• •CLI 0.134.0 release Local conversation search and a primary --profile flag landed while Windows TUI corruption and remote reliability issues were also addressed.

What this means for you

For Vibe Builders: You can now run background shell tasks and reload skills without restarting Claude Code so agent workflows stay live during longer builds. Codex profile and permission tools let you switch access rules quickly when testing different agent behaviors. Cursor staying quiet means your current setup remains the baseline while the terminal agents add the new controls.

For Non-techies: Claude Code now applies code review fixes automatically and lets you skip large file downloads which keeps daily terminal work moving faster. Codex added clearer permission profiles and better error reports so you spend less time fixing setup problems. No Cursor changes means your existing editor tools stay the same for now.

For Developers: Claude Code introduced automatic fallback switching and disallowed-tools frontmatter so production agent sessions become more reliable without extra wrapper code. Codex 0.134.0 added local search and richer diagnostics that help debug MCP and Git issues in live environments. Cursor silence gives you a stable reference point while you benchmark the new terminal features against your current stack.

What to watch next

Track whether Claude Code extends background agent orchestration beyond this week and whether Codex adds local model options. Watch for any Cursor patch that re-enters the weekly cycle.