Claude Code 50 fixes, Cursor v3.8-v3.11, Codex CLI 0.143+ (agent tools for daily coding)
TL;DR
Claude Code pushed dozens of CLI fixes and defaults this week while Cursor shipped five releases with automation and mobile agents and Codex added plugins plus desktop integration.
What shipped
From 10 to 17 July, Cursor, Claude Code, and OpenAI Codex released updates centered on agent workflows, permission controls, and cross-platform access for coding tools.
Cursor
Cursor released five versions this week that add automation skills, plugin management, and an iOS beta for cloud agents. The updates target repetitive task handling and team marketplace controls with concrete triggers for Slack, GitHub, and GitLab.
- •Cursor v3.8 Cursor released version 3.8 with an /automate skill that handles repetitive tasks through Slack emoji triggers, five GitHub events, and cloud computer use for producing demos.
- •Cursor v3.9 Cursor released version 3.9 that adds a single Customize page for plugins, skills, MCPs, and rules plus a marketplace leaderboard with prebuilt canvases for Hex and Atlassian.
- •Cursor iOS beta Cursor launched a public iOS beta that lets users start and manage always-on cloud agents from phones with Remote Control handoff and mobile PR reviews.
- •Cursor v3.10 Cursor released version 3.10 that lets admins distribute Team MCP servers across agents, IDE, and CLI while adding organization group restrictions in team marketplaces.
- •Cursor v3.11 Cursor released version 3.11 that supports side chats alongside main sessions, local Cmd+K search of agent transcripts, and simplified pickers for GitHub, GitLab, and Azure DevOps.
Claude Code
Claude Code delivered fifty updates across the week that set new defaults for auto mode on Bedrock and Vertex AI, added vim remaps and screen reader support, and fixed permission, login, and subagent issues for more reliable terminal workflows.
- •Auto mode default on Bedrock Claude Code enabled auto mode by default on Bedrock, Vertex AI, and Foundry without an environment variable opt-in.
- •vimInsertModeRemaps setting Claude Code added the vimInsertModeRemaps setting to map sequences like jj to Escape in insert mode.
- •Screen reader mode Claude Code introduced an opt-in screen reader mode that renders plain text via the --ax-screen-reader flag.
- •Live elapsed-time counter Claude Code added a live elapsed-time counter to collapsed tool summary lines for long-running calls.
- •Prompt caching fix on Bedrock Claude Code fixed a prompt-caching regression that had billed trailing system context as fresh tokens on Bedrock and Vertex.
- •Subagent model persistence Claude Code prevented subagents from reverting to the parent model on resume after an explicit override.
- •MCP server reconnection fix Claude Code fixed plugin MCP servers failing to reconnect after idle web sessions woke.
- •Parallel session logout fix Claude Code resolved simultaneous logout across parallel sessions sharing one credential store after sleep.
- •Bidirectional override neutralization Claude Code neutralizes bidirectional-override and zero-width characters in permission previews.
- •--forward-subagent-text flag Claude Code added the --forward-subagent-text flag to include subagent output in stream-json.
- •Login expiration warnings Claude Code now shows warnings when login is about to expire for background sessions.
- •Auto mode transcript rule Claude Code added an auto mode rule that blocks attempts to tamper with session transcript files.
- •Directory suggestions for /cd Claude Code added directory path suggestions to the /cd command matching /add-dir behavior.
- •Claude Opus 4.8 default Claude Code now defaults to Claude Opus 4.8 on Bedrock, Vertex AI, and AWS.
- •Auto mode default on platforms Claude Code enabled auto mode by default on Bedrock, Vertex AI, and Foundry without opt-in.
- •vimInsertModeRemaps update Claude Code introduced vimInsertModeRemaps to map two-key sequences to Escape in vim mode.
- •Screen reader mode opt-in Claude Code added opt-in plain-text rendering for screen readers via flag or settings.
- •Blocked dialogs fix Claude Code reverted a guard that had blocked interactive dialogs in background agent sessions.
- •Deprecated permission warnings Claude Code now shows startup warnings for Write, NotebookEdit, and Glob rules.
- •Collapsed tool timer Claude Code displays a live elapsed-time counter in collapsed tool summaries.
- •Dynamic workflow size setting Claude Code added a Dynamic workflow size setting in /config to control agent counts.
- •Login expiration warnings added Claude Code added proactive warnings for expiring logins to avoid background interruptions.
- • /doctor full checkup upgrade Claude Code upgraded /doctor to a comprehensive setup checkup that diagnoses and fixes issues.
- •CLAUDE.md trimming check Claude Code added a /doctor check that suggests trimming redundant content from CLAUDE.md files.
- •Claude Opus 4.8 default on AWS Claude Code defaults to Claude Opus 4.8 on Bedrock, Vertex AI, and AWS.
- •Auto mode default on Foundry Claude Code enabled auto mode by default on Bedrock, Vertex AI, and Foundry.
- •Corporate launcher support Claude Code added CLAUDE_CODE_PROCESS_WRAPPER to route processes through corporate launchers.
- •vimInsertModeRemaps for vim Claude Code lets users map sequences like jj to Escape in vim mode.
- •Screen reader mode support Claude Code supports opt-in plain-text rendering for accessibility via flag or settings.
- •Background subagents default Claude Code now runs subagents in the background by default for better multitasking.
- •Stacked slash-skill support Claude Code supports up to five stacked slash-skill invocations with improved error handling.
- •Manual permission mode default Claude Code changed the default permission mode to Manual across CLI and IDEs.
- •Sonnet 5 system role change Claude Code disabled mid-conversation system role reminders for Claude Sonnet 5 sessions.
- •Dynamic workflow and OTel Claude Code added dynamic workflow size setting and OpenTelemetry attributes for workflows.
- •Login warnings and badge Claude Code added login expiration warnings and a manual permission mode badge.
- •Hook event streaming fix Claude Code fixed hook event streaming in headless sessions to prevent idle reaping.
- •Transcript tampering block Claude Code added an auto mode rule that blocks session transcript tampering.
- • /cd and /doctor improvements Claude Code added directory suggestions to /cd and a CLAUDE.md trim check in /doctor.
- •Stacked slash-skill handling Claude Code supports up to five stacked slash-skill invocations with better SSL and 429 handling.
- •Sonnet 5 role behavior update Claude Code removed mid-conversation system role reminders for Claude Sonnet 5.
- •Workflow size and OTel update Claude Code introduced dynamic workflow size setting and added OTel attributes.
- •Login warnings and badge added Claude Code added login expiration warnings and manual permission badge.
- •Hook streaming fix Claude Code resolved hook event streaming issues in headless sessions.
- •Transcript tampering prevention Claude Code added an auto mode rule blocking transcript tampering and upgraded /doctor.
- • /cd suggestions and trim check Claude Code added directory suggestions to /cd and a CLAUDE.md trim suggestion in /doctor.
OpenAI Codex
OpenAI Codex released thirty-three updates that enable remote plugins by default in the CLI, add new approval modes, and integrate Codex directly into the ChatGPT desktop app on macOS and Windows.
- •Remote plugins default in CLI Codex CLI 0.143.0 enables remote plugins by default with richer catalog rows and npm marketplace sources.
- •Usage-limit credits and writes mode Codex CLI 0.144.0 adds usage-limit reset credit details and a new writes app-approval mode.
- •Standalone installer fix Codex CLI 0.144.1 fixes standalone installation failures from compact GitHub metadata and adds runtime fallback.
- •Codex in ChatGPT desktop Codex is now integrated into the ChatGPT desktop app so users can edit Markdown and code directly and review PRs.
- •Guardian auto-review restore Codex CLI 0.144.2 restores the previous Guardian auto-review policy after a prompting regression rollback.
- •Inline visualizations on iOS ChatGPT for iOS 1.2026.188 adds inline visualizations in Codex tasks and improves task creation from conversations.
- •Dangerous-command detection Codex CLI 0.144.5 improves detection of forced rm commands and provides clearer rejection reasons.
- •Remote plugins and proxy support Codex CLI 0.143.0 enables remote plugins by default and routes traffic through system proxies.
- •New approval modes in CLI Codex CLI 0.144.0 introduces writes app-approval mode and interactive MCP tool authentication.
- •Installer and code-mode fixes Codex CLI 0.144.1 fixes standalone installs and ensures macOS packages expose the code-mode host.
- •Codex inside ChatGPT app Codex is integrated into the ChatGPT desktop app on macOS and Windows for direct code editing and PR review.
- •Guardian policy restore Codex CLI 0.144.2 restores Guardian auto-review policy and tool behavior after regression rollback.
- •Dangerous-command improvements Codex CLI 0.144.5 enhances dangerous-command detection with clearer rejection messages.
- •Remote plugins and Bedrock models Codex CLI 0.143.0 enables remote plugins and adds Amazon Bedrock GPT-5.6 model support.
- •Writes mode and MCP auth Codex CLI 0.144.0 adds writes app-approval mode and interactive authentication for MCP tools.
- •Codex in desktop app Codex is available inside the ChatGPT desktop app for macOS and Windows with PR sidebar support.
- •Inline visualizations support ChatGPT iOS now supports inline visualizations in Codex tasks with improved tool styling.
- •Remote plugins and proxies Codex CLI 0.143.0 enables remote plugins by default and supports system proxy routing.
- •Usage credits and writes mode Codex CLI 0.144.0 adds usage-limit reset credits and a writes app-approval mode.
- •Installer fix on macOS Codex CLI 0.144.1 resolves standalone installation failures and exposes code-mode host on macOS.
- •Remote plugins and Bedrock Codex CLI 0.143.0 enables remote plugins and adds Amazon Bedrock GPT-5.6 models.
- •Codex CLI 0.144.3 release Codex CLI published version 0.144.3 with no merged pull request changes.
- •New approval modes release Codex CLI 0.144.0 adds writes app-approval mode and global pnpm detection.
- •Installer fixes in 0.144.1 Codex CLI 0.144.1 resolves GitHub metadata issues for standalone installs.
- •Codex in ChatGPT app Codex is integrated into the ChatGPT desktop app for code editing and PR work.
What this means for you
For Vibe Builders: You can now run always-on cloud agents from your phone with the Cursor iOS beta and hand off sessions from desktop. Claude Code defaults and Codex desktop integration let you trigger automated workflows without managing separate CLIs. Test the /automate skill in Cursor or the writes approval mode in Codex CLI this week to ship agent tasks faster.
For Non-techies: For daily business use, Codex now lives inside the ChatGPT desktop app so you can review pull requests and edit code without switching tools. Cursor mobile beta and Claude Code login warnings reduce interruptions during background tasks. Start with the ChatGPT integration to handle simple GitHub reviews from one window.
For Developers: Claude Code added fifty fixes for subagent persistence, MCP reconnection, and prompt caching on Bedrock and Vertex, improving reliability for production terminal agents. Cursor team MCP distribution and Codex remote plugin defaults plus system proxy support give clearer paths for enterprise rollout. Benchmark the new manual permission mode and OTel attributes against your current agent stack before integrating.
What to watch next
Watch for Cursor team marketplace adoption metrics and any follow-up Claude Code release that addresses remaining Windows worktree bugs. Track Codex CLI 0.145 for further approval mode refinements and iOS visualization polish.
Harsh’s take
The week shows heavy focus on permission hardening and background reliability rather than new model capabilities. Many Claude Code entries repeat the same fixes across days, suggesting rapid patch cycles instead of coordinated releases. This pattern risks alert fatigue for developers who must scan duplicate announcements.
Second-order effect is that non-technical users gain easier mobile and desktop entry points while developers still manage fragmented CLI settings. The concrete action is to pick one agent surface, either Cursor side chats or Codex writes mode, and run a two-day test on a real repository before adopting further updates.
by Harsh Desai
More AI news
- FeatureLovable chat shows images, videos, and files as clickable thumbnails
Lovable now renders generated or referenced files as clickable thumbnail tiles in chat. Images and videos display visual previews, while other files appear as labeled tiles that open in the Files panel.
- FeatureLovable enables workspace identity reuse in built apps
Lovable now lets built apps recognize signed-in users automatically by reading name, email, and ID. The feature is available on Business and Enterprise plans for newer TanStack Start projects.
- Daily RoundupTernary-Bonsai-27B and Wan-Dancer-14B trend on Hugging Face, Kimi K3 on AI Gateway, plus agent tools on Product Hunt
Hugging Face models and Google updates lead releases today while Vercel, NVIDIA, and new agent startups add practical tools for inference, robotics, and business workflows.