Cursor 3.11 side chats and router, Claude Code Opus 5, and Codex CLI 0.146 thread tools (three agent platforms advance)
TL;DR
Cursor, Claude Code, and OpenAI Codex each shipped agent workflow upgrades this week, from parallel chats and model routing to background sessions and plugin manifests.
What shipped
From 24 to 31 July, three major coding agent platforms released clustered updates focused on parallel workflows, model routing, and team controls. Cursor accounted for the largest volume of changes while Claude Code and Codex refined CLI reliability and integrations. The releases center on practical agent features that reduce context switching during development tasks.
Cursor
Cursor delivered the majority of this week's updates across 25 items centered on version 3.11 and supporting features. The changes emphasize parallel exploration tools, intelligent model routing, and expanded access options including regional pricing and tablet support. These build on prior Slack and marketplace enhancements to support more distributed team workflows.
- •Side chats and transcript search Cursor 3.11 added side chats via /side for parallel questions without breaking the main thread, plus Cmd+K search across thousands of transcripts.
- •Slack multi-repo support Cursor in Slack now posts execution plans before tasks and allows mid-task repository switching in multi-repo setups with cross-channel messaging.
- •Cursor Router for Auto mode The new router analyzes requests and routes them to optimal models under Cost, Balance, or Intelligence modes with admin controls for teams.
- •Cursor Start plan in India Developers in India can access Grok 4.5 and cloud agents for ₹649 per month via UPI on the new local pricing tier.
- •Cursor for iPad release Paid users gained split-screen layout, Apple Pencil markup, inbox task tracking, and full PR review and merge tools on iPad.
- •Version 3.11 side chats The update introduced durable side chats for research tangents alongside redesigned project pickers and cloud agent lifecycle hooks.
- •Slack execution plans Slack integration now shares plans upfront and supports cross-thread context while improving artifact rendering in messages.
- •Auto mode optimization modes Cursor Router powers Auto mode with three selectable optimization profiles and per-team model block lists across platforms.
- •India local pricing tier The Start plan bundles Composer models, always-on agents, and iOS remote control at ₹649 monthly with UPI support.
- •iPad PR review tools The tablet app includes inbox management and full pull request capabilities on all paid plans with optimized larger-screen layout.
- •Team MCP marketplace controls Admins can centrally configure Team MCP servers for distribution across agents, IDE, and CLI with organization group restrictions.
- •Transcript search addition Full-text search now covers agent transcripts while simplified pickers and new cloud hooks improve session management.
- •Multi-repo Slack workflows The integration enables repository switching prompts and cross-channel updates during active tasks.
- •Admin routing controls Team admins can set default modes, enablement rules, and block lists for the router across desktop, web, and CLI.
- •Grok 4.5 access via Start Indian developers receive plugin and MCP server support plus mobile remote control under the new monthly plan.
- •Organization group access Marketplace access now respects both SCIM directory groups and custom organization groups for finer control.
- •Durable side chats in 3.11 The release added command palette transcript search and project picker redesigns alongside cloud agent hooks.
- •Slack plan sharing Tasks begin only after an execution plan is posted, with improved message footers for multi-repo environments.
- •Dynamic model routing Requests route automatically to frontier or efficient models based on task complexity under the three optimization modes.
- •UPI payment support The India plan includes iOS remote control and Composer access at the reduced local rate.
- •iOS public beta launch Mobile users can manage cloud agents, review diffs, and control desktop sessions via Live Activities.
- •Central MCP distribution Team admins distribute configured servers to agents and CLIs while controlling marketplace visibility by groups.
- •Parallel question handling Side chats let users explore tangents locally while the main agent context stays intact.
- •Cross-channel Slack updates Refined formatting supports context reading across channels during agent tasks.
- •Cost and intelligence modes Router options now include explicit Cost, Balance, and Intelligence profiles with granular admin settings.
Claude Code
Claude Code released 24 updates focused on CLI stability, subagent behavior, and sandbox controls. The changes introduce Claude Opus 5 as the default model and refine permission handling plus background session tools. These adjustments target long-running agent sessions and team reliability needs.
- •Fork command background sessions The /fork command now duplicates conversations into separate background sessions while the original stays active.
- •EndConversation tool addition A new tool handles policy-violating sessions alongside fixes for Bash and PowerShell permission bypasses.
- •Manual verify and review invocation The /verify and /code-review skills now require explicit slash command calls instead of running automatically.
- •Sandbox filesystem toggle A new setting disables filesystem isolation while keeping network egress control for flexible environments.
- •Emoji autocomplete and MCP fixes Prompt input gained shortcode autocomplete with memory leak repairs for truncated MCP tool outputs.
- •Background code review subagents The /code-review skill now runs as a background subagent to keep main conversation context concise.
- •Claude Opus 5 default model Opus 5 with a 1M context window became the default Opus model with updated sandbox allowlists.
- •CLI reliability improvements General bug fixes addressed session stability and terminal path handling across workflows.
- •Background session forking /fork creates dedicated background sessions and renames prior subagent behavior to the /subtask command.
- •Permission check hardening Periodic progress heartbeats and OTel logging enhancements accompany fixes for shell permission bypasses.
- •Restricted auto skills execution Users must manually invoke verification skills via slash commands after the automatic trigger was removed.
- •Long session performance fixes The filesystem.disabled setting and other tweaks resolve multi-second stalls in extended agent runs.
- •Transcript failure warnings Emoji autocomplete pairs with clear warnings for transcript write issues and subagent execution caps.
- •Screen reader accessibility Code reviews as background subagents now include announcements for text deletion and improved auto mode checks.
- •Subagent nesting depth increase Opus 5 default supports nesting up to depth 3 with DirectoryAdded hooks and stream-json forwarding.
- •General stability updates Version 2.1.220 focused on bug fixes that improve overall CLI session reliability.
- •Distinct background sessions The fork update separates conversations into new sessions while preserving the active one.
- •Policy session handling tool EndConversation plus heartbeat logging addresses permission issues in Bash and PowerShell.
- •Explicit skill invocation required Verification commands now demand manual slash command use after automatic execution was disabled.
- •Optional filesystem isolation The sandbox setting lets users skip isolation while retaining network controls.
- •Prompt emoji shortcodes Autocompletion joins memory leak fixes for MCP outputs and subagent concurrency limits.
- •Concise context via subagents Background execution of reviews preserves main thread focus with added accessibility features.
- •1M context Opus 5 default The model ships with strict allowlists and nested subagent support for complex workflows.
- •System reliability patches Final fixes target broad CLI stability across terminal environments.
OpenAI Codex
OpenAI Codex contributed 15 updates centered on CLI 0.146 and desktop integrations. The releases add thread forking, plugin manifests, and partner sign-in flows alongside voice and multi-folder project support. These target smoother multi-agent and external tool connections.
- •Codex CLI 0.146.0 session tools The release added /new and /clear for session naming, thread pinning, and agent plugin manifest support.
- •Sign in with ChatGPT beta Partner integrations now include Airtable, GitLab, HubSpot, Notion, Supabase, and Vercel for simplified authentication.
- •Desktop browser and review upgrades ChatGPT desktop gained browsing history search, multi-repo code review diffs, and canvas image refinement.
- •Thread forking with pagination Codex CLI supports forking conversations with paginated history that can stay hidden from listings.
- •Agent plugin marketplace expansion Manifest support now covers workspace publishing plus Amazon Bedrock and Claude Code environments.
- •Standalone web search for providers Custom model providers gained independent web search during active CLI sessions.
- •WebSocket remote Code Mode App-server instances can connect directly to remote execution hosts via WebSocket.
- •Executor skill discovery The CLI now detects and securely accesses skills provided by executors with explicit selection.
- •Session naming and pinning Users name sessions with /new or /clear and pin important threads for quick switching.
- •Settings import from other tools Version 0.145.0 added migration support for Cursor and Claude Code configurations.
- •ChatGPT Voice in desktop GPT-Live voice lets users talk through tasks with multi-folder local project support added.
- •Paginated thread history Experimental history pagination pairs with Amazon Bedrock login and audio input options.
- •Multi-agent V2 stability Updates improved multi-agent handling alongside streaming V3 conversation support.
- •Multi-folder desktop projects Local projects can now span multiple related folders for search and editing.
- •Partner account linking flow Beta sign-in simplifies plugin enablement on participating external services.
What this means for you
For Vibe Builders: Cursor side chats and iPad PR tools let you explore ideas in parallel and review work on the go without switching devices. Claude Code background subagents and Codex thread forking keep main sessions clean while you test agent workflows. Start with Cursor Router or Claude Code sandbox toggles to match cost and control needs for your next non-code project.
For Non-techies: These releases add Slack plans and partner sign-ins that simplify account setup across tools like Notion or Vercel. Cursor India pricing and iPad access lower barriers for occasional users managing tasks on mobile. Watch for voice features in Codex desktop and manual review controls in Claude Code to reduce unexpected agent actions in daily operations.
For Developers: Cursor 3.11 router and Team MCP controls plus Claude Code Opus 5 default and permission fixes give clearer options for production agent stacks. Codex CLI 0.146 plugin manifests and WebSocket remote hosts support custom executor integrations. Benchmark the new background session patterns against your current CLI setup before rolling out to teams.
What to watch next
Track Cursor 3.12 picker refinements and Claude Code subagent depth changes. Watch Codex for expanded Bedrock marketplace reach and desktop multi-folder search updates. Test the new router modes and background fork commands in your own agent sessions this week.
Harsh’s take
The week showed heavy overlap in parallel session tools across all three platforms, suggesting vendors are racing to solve the same context-management pain rather than differentiating on core reliability. Cursor's volume of releases masks that many items repeat prior Slack and routing announcements, while Claude Code focused on narrower CLI hygiene fixes. Builders should pick one platform's background session approach and measure actual token usage and error rates before adopting the full feature set.
by Harsh Desai
More AI news
- Daily RoundupGemini Robotics ER 2 and Inkling-Small top releases, plus Vercel agent tools
Model releases on Hugging Face and Replicate join Vercel platform updates and OpenAI price cuts to shift AI from chat interfaces toward runnable agents and cheaper inference.
- Daily RoundupLyria 3.5 music model, AI Gateway fast mode, and agent tools on Product Hunt
Vercel and Google rolled out new AI features while Hugging Face models and Product Hunt agents showed fresh options for builders and teams.
- Daily RoundupPixelship agent on Replicate, Gemini 3.6 Flash batch on OpenRouter, and agent tools for builders
New agentic image tools and trending multimodal models arrived alongside expanded Gemini agent features and deployment options for production use.