Skip to content
Cursor iPad and Origin beta, Claude Code subagent forking, and Codex CLI 0.149 updates you can run today | My AI Guide (programmatic OG fallback)

Cursor iPad and Origin beta, Claude Code subagent forking, and Codex CLI 0.149 updates you can run today

By Harsh Desai
Share

TL;DR

Cursor added iPad support and Google Workspace plugins, Claude Code enabled default subagent forking with GitLab tools, and OpenAI Codex shipped new CLI versions plus cross-app sync features.

What shipped

From 14 to 21 August 2026, three major coding platforms released frequent updates focused on mobile access, agent collaboration, and CLI tooling. Cursor emphasized iPad workflows and background builds, Claude Code led with the highest volume of agent and integration fixes, and OpenAI Codex improved desktop-to-mobile continuity plus security tiers.

Cursor

Cursor released 25 updates centered on iPad availability for paid plans and Google Workspace marketplace plugins. Background builds cut cloud agent startup times by up to 10x while Origin beta added native hosting and GitHub sync. A new India-specific Start plan at ₹649 per month brought localized pricing.

  • Cursor iPad app Cursor released an iPad version with split-screen PR reviews, Apple Pencil markup, and inbox task tracking for paid users.
  • Google Workspace plugins Cursor added Drive, Gmail, and Calendar plugins so agents can search files, draft emails, and schedule events inside the editor.
  • Background builds Cursor introduced pre-warmed environments that let cloud agents start up to 10x faster with automatic fallback on failures.
  • Origin code hosting beta Cursor launched Origin beta on paid plans with two-way GitHub sync, native repos, and integrations for Vercel and Buildkite.
  • Cloud agent subscriptions Cursor enabled agents to subscribe to PR and Slack events while running isolated subagents on separate VMs.
  • iPad PR reviews Cursor expanded the iPad app with multi-agent sidebars and full Bitbucket plus Azure DevOps pull request support.
  • Workspace plugin updates Cursor refined Google Drive, Gmail, and Calendar plugins for direct file and email management by coding agents.
  • Pre-warmed agent environments Cursor added background build preparation that delivers 3x faster time-to-first-token for cloud agents.
  • Origin GitHub sync Cursor introduced Origin early beta featuring bi-directional sync and agent interaction across hosted codebases.
  • Agent event subscriptions Cursor let cloud agents monitor PRs and Slack threads with persistent goals and custom pinned modes.
  • India Start plan Cursor launched a ₹649 monthly plan for Indian developers with UPI payments and daily Grok plus Composer access.
  • iPad multi-PR sessions Cursor added split-screen reviews and inbox tracking for multiple pull requests on iPad.
  • Drive and Calendar plugins Cursor released official plugins allowing agents to read, search, and manage Google Workspace content.
  • Ready environment builds Cursor prepared development setups in the background at no extra cost for faster agent starts.
  • Origin repository hosting Cursor opened Origin beta on paid plans with PR review workflows and extension support.
  • Cursor Router for Auto mode Cursor activated dynamic model routing across cost, balance, and intelligence tiers with admin controls.
  • Localized India pricing Cursor offered the Start tier at ₹649 monthly with UPI and card support for local developers.
  • iPad full PR workflow Cursor delivered multi-column layouts and complete review-and-merge flows on iPad for paid tiers.
  • Official Workspace plugins Cursor connected agents directly to Drive, Gmail, and Calendar for file and event handling.
  • Faster cloud agent boot Cursor used background builds to reach 10x quicker startup and resilience against broken updates.
  • Router-powered Auto mode Cursor analyzed prompt complexity to route requests while giving teams granular governance.
  • Cursor Start tier Cursor introduced India-specific pricing with daily model access and iOS remote control.
  • iPad agent workspace Cursor expanded the iPad app with unified inbox and end-to-end pull request reviews.
  • Gmail and Drive integration Cursor enabled agents to draft messages and edit files through new marketplace plugins.
  • Background environment prep Cursor prepared ready-to-use setups that automatically revert on update failures.

Claude Code

Claude Code delivered 48 updates, the largest set this week, with default subagent forking, GitLab merge request support, and cross-session messaging. Memory limits for Bash tools, spellcheck options, and self-hosted runners for Team plans appeared repeatedly. Fixes targeted OAuth redirects, terminal redraws, and gateway prompt caching.

  • GitLab worktree support Claude Code added MR URL handling to worktrees and agent views plus opt-in Linux memory cgroups.
  • GitLab MR badges Claude Code placed merge request status badges in the footer and statusline with auto-resume after limit resets.
  • Prompt spellcheck setting Claude Code introduced optional aspell or hunspell checks to underline typos in prompts.
  • Default model variable Claude Code added ANTHROPIC_DEFAULT_MODEL for persistent startup model selection across sessions.
  • Concise output style Claude Code built a /config option that skips preamble and delivers direct results.
  • Readline keybindings Claude Code added Bash-style navigation and plugin marketplace command sources.
  • Subagent forking default Claude Code enabled subagent forking by default to share context across agents.
  • GitLab MR URL support Claude Code extended worktree flags and agent views to handle GitLab merge requests.
  • Auto-resume on limits Claude Code automatically continues sessions after usage limit resets with GitLab badges.
  • Spellcheck underline Claude Code added aspell-powered typo detection and refined permission dialogs.
  • Idle notification support Claude Code introduced notify_when_idle for cross-session SendMessage alerts.
  • Gateway caching fix Claude Code resolved prompt cache issues when using LLM gateways or custom base URLs.
  • Remote Control resume Claude Code added claude remote-control --continue and SSE keepalives for long pauses.
  • MCP OAuth redirect fix Claude Code corrected sign-in failures caused by URI mismatches on pre-registered clients.
  • Cross-session @mentions Claude Code enabled direct messaging between active sessions via SendMessage.
  • Memory cgroup limits Claude Code introduced opt-in Linux memory controls for Bash tool execution.
  • Usage limit auto-continue Claude Code resumes sessions automatically when claude.ai limits reset.
  • Terminal spellcheck option Claude Code added underline for misspelled words using system spellcheckers.
  • Windows Git path fix Claude Code resolved Git discovery failures when launched from parent directories.
  • Plugin command sources Claude Code allowed dynamic resolution of marketplace plugin directories.
  • Pre-registered OAuth fix Claude Code fixed redirect mismatches for servers like Slack using pre-registered clients.
  • Default subagent forking Claude Code turned on background subagent spawning with inherited context.
  • GitLab token redaction Claude Code added secret masking and marketplace support for GitLab URLs.
  • Gateway spend warnings Claude Code introduced operator-defined spend limit alerts in usage messages.
  • Self-hosted runners Claude Code launched self-hosted runners for Team and Enterprise plans.
  • Image viewing in Remote Control Claude Code enabled direct photo display from the Claude app.
  • General reliability fixes Claude Code shipped miscellaneous stability improvements across the CLI.
  • Subscription flag evaluation Claude Code fixed feature flag checks on expired tokens at startup.
  • Synced skill hardening Claude Code prevented command injection from claude.ai synced skills.
  • SSE keepalive pings Claude Code added pings during long thinking pauses to avoid gateway timeouts.
  • OAuth redirect correction Claude Code resolved sign-in failures for pre-registered Slack-like clients.
  • Subagent context sharing Claude Code enabled default forking with background spawns.
  • User identity forwarding Claude Code passed gateway headers for spend attribution on upstreams.
  • MR status badges Claude Code displayed GitLab merge request states in the statusline.
  • Cross-session SendMessage Claude Code introduced messaging between sessions on different machines.
  • Directory trust prompts Claude Code added workspace verification before opening agents in untrusted folders.
  • Routine bug fixes Claude Code delivered ongoing stability and crash prevention updates.
  • Slash-command menu polish Claude Code improved suggestion performance and visual styling.
  • Terminal redraw fixes Claude Code resolved rare layout error stalls during interactive sessions.
  • Plugin marketplace sources Claude Code allowed dynamic command loading from external directories.
  • MCP form support Claude Code added editable approvals and standard MCP forms in iOS.
  • Bedrock integration Claude Code enabled Amazon Bedrock Runtime with GPT-5.6 routing in CLI.
  • Markdown export command Claude Code added /export for conversation downloads in Markdown.
  • Agent plugin catalogs Claude Code introduced portable plugin catalogs and secret redaction.
  • Linux desktop preview Claude Code supported importing settings from Cursor and Claude tools.
  • Computer History opt-in Claude Code turned app activity into timeline memories for Pro users.
  • Codex CLI 0.149 Claude Code added interactive agents dashboard and Vim motion expansions.
  • Codex CLI 0.148 Claude Code introduced session forking and asynchronous MCP hooks.

OpenAI Codex

OpenAI Codex released 24 updates including Codex CLI versions 0.147 through 0.149 with plugin catalogs and Bedrock support. Pinned threads now sync across macOS and iOS while Computer History became available in Europe for Pro plans. Daybreak Blue and Red tiers gave vetted defenders access to specialized GPT-5.6 models.

  • Pinned thread sync OpenAI Codex synchronized pinned chats between macOS desktop and iOS apps.
  • Read-only thread shares OpenAI Codex allowed macOS users to share local Codex threads with revocable links.
  • Computer History in EEA OpenAI Codex enabled cross-app activity timelines for Pro users in Europe.
  • Site URL edits OpenAI Codex let workspace owners change hosted Site addresses without new deployments.
  • Workspace co-editing OpenAI Codex permitted member editors to update Sites while owners retain control.
  • Apple Messages plugin OpenAI Codex added direct message management from the macOS desktop app.
  • Codex CLI 0.149 OpenAI Codex introduced an interactive agents dashboard and directory commands.
  • Codex CLI 0.148 OpenAI Codex added Markdown export, session forking, and Amazon Bedrock support.
  • iOS Codex Remote launch OpenAI Codex added a direct launch setting and MCP form support on iOS.
  • Bedrock Runtime routing OpenAI Codex enabled GPT-5.6 routing through Amazon Bedrock in the CLI.
  • Agent Plugins catalog OpenAI Codex released portable plugin catalogs and MCP 2026-07-28 support.
  • iOS task reliability OpenAI Codex improved reconnection and prompt editing performance on iOS.
  • Daybreak security tiers OpenAI Codex launched Blue and Red access levels for defensive cybersecurity work.
  • Linux desktop preview OpenAI Codex released .deb and .rpm builds for Ubuntu, Debian, and Fedora.
  • macOS Computer History OpenAI Codex turned app activity into memories for Pro, Business, and Enterprise.
  • Portable plugins release OpenAI Codex added organized transcript sections and --approve-for-me flag.
  • Defender access tiers OpenAI Codex provided vetted users GPT-5.6 Sol and Cyber models for red teaming.
  • External agent imports OpenAI Codex allowed import of skills and history from Claude Code and Cursor.
  • Linux app packages OpenAI Codex delivered preview builds supporting local files and Codex integration.
  • Activity timeline feature OpenAI Codex converted website visits into ChatGPT memories on macOS.
  • CLI 0.147 updates OpenAI Codex introduced persistent sections and Bedrock compaction optimizations.
  • Daybreak Blue tier OpenAI Codex gave approved defenders access to GPT-5.6 Sol for analysis tasks.
  • Settings import tool OpenAI Codex synced instructions and recent work from competing agents automatically.
  • macOS activity memories OpenAI Codex enabled opt-in Computer History for Enterprise plans.

What this means for you

For Vibe Builders: You can now run full PR reviews on an iPad with Cursor, connect agents to Gmail and Drive without code, and test subagent forking in Claude Code on your existing projects. The India Start plan and Linux desktop previews lower the barrier to daily use. Pick one tool this week and ship a small agent workflow inside your current stack.

For Non-techies: SMB owners gain mobile access to code reviews in Cursor and cross-device chat sync in Codex without new hardware. Claude Code self-hosted runners and spend alerts help control team costs. Start with the iPad app or macOS desktop preview to handle routine tasks like email drafting or calendar checks inside AI tools.

For Developers: Evaluate Cursor background builds against Claude Code memory cgroups and Codex Bedrock routing for your production pipelines. Subagent forking and cross-session messaging in Claude Code plus MCP protocol updates in Codex CLI signal a shift toward composable local agents. Benchmark the new GitLab worktree support and gateway spend limits against your current runner setup before integrating.

What to watch next

Track Cursor Origin rollout to more regions and any Claude Code updates on Windows NT path handling. Watch for Codex CLI 0.150 and expanded Daybreak tier availability in additional countries.

Harshs take

The week showed heavy overlap in mobile and plugin features across all three tools, with Claude Code producing the most incremental fixes while Cursor and Codex focused on user-facing polish. This volume of small releases suggests vendors are racing to close feature gaps rather than delivering differentiated reliability. Builders should pick one platform, measure actual agent uptime on their workloads, and ignore marketing claims about 10x speed until independent benchmarks appear.

Second-order risk is rising dependency on vendor-specific plugins and marketplaces that may not interoperate. The practical move is to test one new capability, such as Cursor iPad PR reviews or Claude Code subagent forking, inside a non-critical repository this week and log failure modes before scaling.

by Harsh Desai

More AI news

Everything AI. One email.
Every Monday.

New tools. Model launches. Plugins. Repos. Tactics. The moves the sharpest builders are making right now, before everyone else.

No spam. Unsubscribe anytime.