Cursor iPad and Router releases, Claude Opus 5 in Code, Codex CLI session tools (agent workflows you can test today)
TL;DR
Cursor rolled out iPad support, Slack multi-repo planning, and model routing across twenty updates while Claude Code added Opus 5 with deeper subagents and Codex CLI introduced session naming plus plugin catalogs.
What shipped
From 31 July to 7 August 2026 Cursor, Claude Code, and Codex each shipped repeated version updates that expand agent reach into mobile, terminal, and desktop environments.
Cursor
Cursor delivered twenty updates focused on mobile access, workspace integrations, and intelligent routing. The releases center on paid-plan features such as iPad PR reviews and Slack task planning that let non-coders trigger agents from familiar apps.
- •Cursor Slack Upgrade Cursor's Slack integration now shares execution plans before tasks and works across multiple repositories with cross-channel context reading.
- •Cursor Router Launch Cursor Router selects models by task complexity and offers Intelligence, Balance, or Cost modes with admin governance controls.
- •Cursor Start Plan India Cursor Start offers a 649-rupee monthly plan in India with UPI payments, Grok 4.5 access, and always-on cloud agents.
- •Cursor iPad Release Cursor for iPad adds pinned sidebar chats, split-screen PR reviews, Apple Pencil markup, and an Inbox for agent tasks on paid plans.
- •Google Workspace Plugins Cursor agents can now search Drive files, draft Gmail messages, and manage Calendar events inside the editor.
- •Cursor Slack Multi-Repo Cursor in Slack shares task plans upfront and reads context across channels and threads in multi-repository setups.
- •Cursor Router Auto Mode Cursor Router analyzes prompts and routes them to optimal models under Cost, Balance, or Intelligence settings with team controls.
- •Cursor Start Local Pricing Cursor Start provides a 649-rupee plan in India that includes Grok 4.5, Composer, cloud agents, and the iOS app.
- •Cursor iPad PR Tools Cursor for iPad supplies larger layouts, pinned chats, Apple Pencil support, and full Inbox tracking for reviews.
- •Google Workspace Agent Access New plugins let Cursor agents pull files from Drive and update Calendar events without leaving the IDE.
- •Cursor Slack Execution Plans Cursor for Slack displays task plans before running and supports named multi-repo environments with cleaner UI.
- •Cursor Auto Mode Router Cursor powers Auto mode by routing to frontier or cheaper models and adds admin allow or block lists.
- •Cursor Start India Plan The new plan grants Indian developers UPI billing plus daily Grok 4.5 and Composer access with cloud agents.
- •Cursor iPad Mobile Reviews Cursor for iPad includes split views, Apple Pencil markup, Bitbucket support, and multi-PR session handling.
- •Cursor Google Workspace Plugins Agents gain direct search and edit rights in Gmail, Drive, Calendar, Docs, and Sheets.
- •Cursor 3.11 Side Chats Version 3.11 adds side chats, Cmd+K transcript search, and new hooks to monitor cloud agent reasoning.
- •Cursor Slack Cross-Channel Updates The integration posts progress across threads and reads context from multiple Slack channels.
- •Cursor Router Model Selection Router chooses models by Cost, Balance, or Intelligence modes and supplies admin default controls.
- •Cursor Start UPI Support The India plan bundles Grok 4.5, Composer, iOS remote control, and MCP server access for 649 rupees.
- •Cursor iPad Inbox and Reviews Paid iPad users receive side-by-side chats, file diffs, and an Inbox to organize active PRs and agents.
Claude Code
Claude Code issued fifty updates centered on version 2.1.216 through 2.1.224. The changes emphasize subagent nesting, sandbox controls, and stability fixes that affect terminal workflows for developers running long sessions.
- •Claude Code 2.1.218 Reviews Version 2.1.218 runs code reviews as background subagents and improves screen-reader deletion announcements.
- •Claude Code 2.1.219 Opus 5 Version 2.1.219 sets Claude Opus 5 as default with 1M context and allows subagents to nest up to depth 3.
- •Claude Code 2.1.220 Fixes Version 2.1.220 delivers general reliability improvements and stability fixes across interactive sessions.
- •Claude Code 2.1.221 Focus View Version 2.1.221 adds a VS Code Focus view toggle and masks sandbox credentials on Linux and WSL.
- •Claude Code 2.1.222 Worktree Version 2.1.222 enforces git worktree isolation for destructive commands and improves raw diff displays.
- •Claude Code 2.1.223 Marketplace Version 2.1.223 adds owner wildcards for managed marketplaces and fixes Bash permission bypasses.
- •Claude Code 2.1.224 Runners Version 2.1.224 introduces self-hosted runners for Team plans and cross-session SendMessage support.
- •Claude Code Filesystem Toggle Version 2.1.216 adds a setting to disable filesystem isolation while keeping network egress rules.
- •Claude Code Emoji Autocomplete Version 2.1.217 adds emoji shortcode suggestions and caps concurrent background subagents at 20.
- •Claude Code Background Reviews Version 2.1.218 shifts code reviews to subagents to preserve main chat context.
- •Claude Code Opus 5 Support Version 2.1.219 enables Claude Opus 5 with strict network allowlists and DirectoryAdded hooks.
- •Claude Code Stability Update Version 2.1.220 applies general bug fixes and performance improvements across CLI workflows.
- •Claude Code VS Code Focus Version 2.1.221 introduces a Focus view and credential masking for sandboxed Linux setups.
- •Claude Code Worktree Isolation Version 2.1.222 applies isolation to file edits and Bash across all session types.
- •Claude Code Marketplace Controls Version 2.1.223 adds GitHub org wildcard rules and fixes workflow sandbox isolation issues.
- •Claude Code Sandbox Option Version 2.1.216 lets users bypass filesystem isolation while retaining network controls.
- •Claude Code Prompt Emojis Version 2.1.217 adds inline emoji shortcode autocomplete and transcript failure warnings.
- •Claude Code Subagent Reviews Version 2.1.218 runs reviews as background subagents and caps concurrent execution.
- •Claude Code Opus 5 Default Version 2.1.219 sets Opus 5 as the default Opus model with 1M context windows.
- •Claude Code Bug Fixes Version 2.1.220 includes maintenance updates and stability improvements for tool usage.
- •Claude Code Credential Masking Version 2.1.221 adds Focus view and prompt-audit tools for sandboxed sessions.
- •Claude Code Git Isolation Version 2.1.222 removes the ultraplan feature and resolves usage credit display issues.
- •Claude Code Permission Fixes Version 2.1.214 addresses bypasses in Windows PowerShell and Bash sessions.
- •Claude Code Manual Verify Version 2.1.215 requires explicit user invocation for verify and code-review skills.
- •Claude Code Filesystem Setting Version 2.1.216 resolves long-session latency spikes and improves permission prompts.
- •Claude Code Subagent Limits Version 2.1.217 caps concurrent subagents and fixes memory leaks from MCP outputs.
- •Claude Code Accessibility Updates Version 2.1.218 adds screen-reader announcements and refines permission checks.
- •Claude Code Nested Subagents Version 2.1.219 expands default subagent spawn depth to 3 with strict allowlists.
- •Claude Code Reliability Patch Version 2.1.220 delivers essential bug fixes for stable session execution.
- •Claude Code VS Code Toggle Version 2.1.221 adds prompt auditing commands and security fixes for Bash.
- •Claude Code Fork Command The update modifies /fork to clone conversations into separate background sessions.
- •Claude Code EndConversation Tool The new tool terminates abusive sessions and tightens PowerShell permission checks.
- •Claude Code Skill Invocation Version 2.1.215 disables automatic runs for verify and code-review skills.
- •Claude Code Sandbox Toggle Version 2.1.216 fixes quadratic slowdowns in long sessions.
- •Claude Code Emoji Support Version 2.1.217 sets a default limit of 20 concurrent subagents.
- •Claude Code Review Subagents Version 2.1.218 preserves main chat context during background reviews.
- •Claude Code Network Allowlists Version 2.1.219 introduces strict sandbox network allowlist configuration.
- •Claude Code Maintenance Release Version 2.1.220 applies general bug fixes to CLI and background services.
- •Claude Code Fork Workflow The update rebrands in-session subtasks to /subtask for continuous interaction.
- •Claude Code Security Checks The release adds progress heartbeats and expands OpenTelemetry attributes.
- •Claude Code Manual Triggers Users must now invoke verify and code-review via slash commands.
- •Claude Code Performance Fixes Version 2.1.216 resolves auto mode permission classifier errors.
- •Claude Code Autocomplete Caps Version 2.1.217 adds transcript write failure warnings.
- •Claude Code Background Agents Version 2.1.218 refines dynamic plan and research workflows.
- •Claude Code Opus 5 Nesting Version 2.1.219 adds mid-session directory tracking hooks.
- •Claude Code Abuse Tool The EndConversation tool safely terminates jailbreak sessions.
- •Claude Code Explicit Commands Users invoke skills manually after the update.
- •Claude Code Isolation Toggle Version 2.1.216 improves terminal UI rendering.
OpenAI Codex
Codex released thirty-four updates that add session management, plugin catalogs, and mobile task controls. The changes focus on CLI 0.146.0 features and desktop app improvements that let users organize conversations and review multi-repo diffs.
- •Codex CLI Flag Removal The deprecated --full-auto flag is removed and users must switch to --sandbox workspace-write.
- •Codex Web Search Caching Codex CLI now supports cached web search and remote conversation compaction on Amazon Bedrock.
- •Codex MCP Protocol Support Codex adds opt-in support for the MCP 2026-07-28 protocol with paginated discovery.
- •Codex Cursor Skill Import Codex CLI imports Cursor-managed skills and syncs updates across Claude and Cursor histories.
- •Codex Approve For Me Flag The new --approve-for-me flag enables automatic approvals to streamline execution.
- •Codex Conversation Sections Users can organize conversations into persistent manually ordered sections.
- •Codex Agent Plugin Install Codex supports portable Agent Plugins across local, personal, workspace, and remote catalogs.
- •Codex CLI 0.146.0 Release Version 0.146.0 adds session naming with /new and /clear plus thread forking.
- •Codex Sign in with ChatGPT The beta now supports partner integrations such as GitLab, Notion, and Vercel.
- •Codex Desktop Multi-Repo Desktop app version 26.727 adds multi-repository code reviews and a new Activity view.
- •Codex GPT-5.4 Deprecation GPT-5.4 and GPT-5.4 mini retire on August 31 and users should migrate to GPT-5.6 Terra.
- •Codex CLI 0.146.1 Update Version 0.146.1 introduces safer automatic-review defaults for cyber-capable models.
- •Codex Session Naming Version 0.146.0 introduces thread pinning and remote Code Mode over WebSockets.
- •Codex Partner Beta Sign in with ChatGPT rolls out to Airtable, HubSpot, Supabase, and additional partners.
- •Codex Browser Upgrades The desktop update adds address bar search and canvas image editing controls.
- •Codex Model Retirement Users must update workspace defaults to gpt-5.6-terra and gpt-5.6-luna before August 31.
- •Codex Plugin Manifests Version 0.146.0 supports Agent Plugins manifests and extensive proxy handling fixes.
- •Codex Account Linking The beta simplifies account creation for plugins with added privacy controls.
- •Codex Multi-Folder Reviews The desktop app enables cross-repository diff reviews for multi-folder projects.
- •Codex Migration Notice Workspace settings should switch to recommended replacements before the cutoff date.
- •Codex Remote GA Codex Remote reaches general availability with one-to-one QR pairing and DigitalOcean integration.
- •Codex iOS Task Management ChatGPT for iOS 1.2026.181 adds full Codex task management from chat threads.
- •Codex Desktop Integration Codex joins the ChatGPT desktop app with inline code editing and PR Chat.
- •Codex iOS Visualizations Version 1.2026.188 introduces inline visualizations in Codex tasks.
- •Codex iOS Mermaid Support Version 1.2026.195 adds support for inline Mermaid diagrams in task transcripts.
- •Codex Voice Steering Desktop update 26.715 introduces voice-driven task steering powered by GPT-Live.
- •Codex iOS Composer Version 1.2026.202 improves composer autocomplete for long prompts.
- •Codex Beta Rollout Sign in with ChatGPT simplifies authentication across select partner sites.
- •Codex Browser Navigation Desktop update 26.727 adds history search and focused image editing views.
- •Codex Model Switch Users should migrate to GPT-5.6 Terra and Luna before the August 31 retirement.
- •Codex Thread Forking Version 0.146.0 adds paginated history support for forked threads.
- •Codex Plugin Beta The beta streamlines plugin setup for Airtable, GitLab, and Notion users.
- •Codex Activity View The latest desktop update expands the sidebar Activity view.
- •Codex ChatGPT Sign-Ins GPT-5.4 models retire from Codex for ChatGPT-authenticated users on August 31.
What this means for you
For Vibe Builders: Cursor's iPad app and Slack task planning let you trigger agents from mobile or chat without writing code. Claude Code subagent nesting and Codex session naming give you reusable workflows you can test in the terminal or desktop app this week.
For Non-techies: Cursor Start plan in India and iPad reviews make it easier to manage code tasks from a tablet or phone. Codex desktop multi-repo views and Claude Code stability fixes reduce the chance of stalled sessions when you run daily business automations.
For Developers: Claude Code 2.1.219 Opus 5 and strict sandbox allowlists plus Codex CLI 0.146.0 thread forking give you production-grade controls for long-running agents. Evaluate the new subagent depth and plugin manifests against your current stack before integrating.
What to watch next
Track whether Cursor Router reaches general availability and whether Claude Code adds deeper nesting defaults. Watch Codex CLI for the first Agent Plugin catalog updates and any follow-up on GPT-5.6 model performance.
Harsh’s take
The week shows heavy overlap in announcements across Cursor, Claude Code, and Codex, with many items repeating the same features in slightly different wording. This suggests vendors are racing to match mobile and plugin capabilities rather than delivering distinct reliability gains. Builders should pick one environment, measure actual subagent success rates on their own tasks, and ignore marketing volume until concrete benchmarks appear.
by Harsh Desai
More AI news
- Daily RoundupGemini trip planning, WeatherNext 2 forecasts, and Vercel agent plugins roll out
Google and Vercel pushed agent features forward while OpenAI expanded free ChatGPT access and new models appeared on gateways and hubs for builders testing responsive agents.
- Daily RoundupMuse Spark 1.2 on Vercel, v0 API launch, and fresh agent tools for builders
Vercel rolled out Muse Spark 1.2, expanded Sandbox limits, and the v0 API while Google shifted DeepMind leadership and new models appeared on Hugging Face and Fal.
- Daily RoundupDeepSeek V4 Flash 90% off, NVIDIA Alpamayo 2 for AVs, and agent tools shipping today
Vendors pushed model discounts, autonomous vehicle models, faster deploys, and browser-equipped agents while open models and cloud deals advanced across the board.