releases.shpreview
Home/Collections/Coding Agents/Week of May 11, 2026

Agent infrastructure, smarter sheets, and Opus 4.7 lands everywhere

May 11–17, 2026

Cursor and Claude Code shipped major agent workflow upgrades — Cursor with configurable development environments and full-screen tabs, Claude Code with an agent view and goal-completion mode — while Anthropic's Opus 4.7 reached fast mode across the ecosystem and OpenAI's Codex gained a richer TUI and iOS notifications.

Agents get workspaces, views, and goals

The most consequential releases this week gave developers more control over how their coding agents set up and work through complex tasks. Cursor's cloud agents can now be configured with full development environments — cloned repos, installed dependencies, credentials, and build tooling — making it practical to run parallel agents across a fleet. Cursor also added a new full-screen tab mode that maximizes the right panel for focused work on files, changes, or PRs, with a floating prompt bar replacing the agent chat when expanded.

Claude Code v2.1.139 introduced an agent view (claude agents) that surfaces every running, blocked, or completed session in a single list, plus a /goal command that sets a completion condition so Claude keeps working across multiple turns until the target is met, with live token and turn counters. Subsequent patches hardened the experience: v2.1.140 fixed /goal hanging under certain hook configurations and background service startup failures, v2.1.141 fixed permission-prompt dismissal bugs and background job state management, and v2.1.143 added plugin dependency enforcement so you can't accidentally disable a plugin another plugin depends on.

Opus 4.7 fast mode and model upgrades

Anthropic's Opus 4.7 arrived in fast mode across multiple platforms. Claude Code v2.1.142 made it the default for fast mode (replacing Opus 4.6, with an environment variable to pin back if needed) and fixed MCP tool timeout propagation. Windsurf also made Opus 4.7 fast mode available, promising ~2.5x higher output speeds with full intelligence. The speed-ups matter most for long-running agent tasks where per-token latency compounds — exactly the workflows these tools are targeting.

Codex gets a richer TUI and mobile reach

Codex CLI 0.131.0 was the week's most feature-dense release: a reworked TUI with data-driven commands, blended token usage displays, and responsive Markdown tables; remote daemon workflows with registry-backed environments; a Python SDK moved to openai-codex; and a unified @-mention picker that searches files, directories, plugins, and skills. iOS notification support lets you open completed Codex tasks directly from notifications and review changed files. New documentation covers ChatGPT mobile app setup, and Amazon Bedrock guidance arrived for teams wanting AWS-managed deployment.

Smaller but notable

Cursor's Composer 2.5 ships with improved long-running task behavior and instruction following. Grok Build 0.1.209 and 0.1.210 fixed Windows rendering, keyboard layout issues, and added plan-mode polish. Devin added collapsible session folders and archive-all, while Devin's earlier release added snapshot build deletion and multiline MCP environment variables. Cursor in Microsoft Teams lets you delegate tasks to cloud agents by mentioning @Cursor in any channel.

Releases covered