Claude Code on desktop now embeds the iOS simulator directly in a panel beside the conversation window. Build and run your app, and Claude works alongside you in real time. Available in public beta today.
During an iOS simulator session, Claude can observe the running app, interact with it, and iterate autonomously. Users can also tap through the simulator themselves at any point. Requires macOS and Xcode.
In response to a question about Android emulator support, a team member confirmed it is coming soon, following the iOS simulator launch.
A user reported that Claude Code frequently spawns dozens of subagents even for simple, non-coding tasks. Dynamic workflow size can be adjusted or disabled entirely in '/config', and setting 'disallowedTools: Agent' in a subagent's frontmatter prevents it from spawning its own child agents.
Managed Agents now support per-agent effort configuration, session seeding with events, up to 500 skills per session, webhooks for environments and memory stores, and event streaming for sub-agents.
Effort levels can now be configured per Managed Agent. Lowering effort reduces both response time and cost, giving developers a straightforward cost-versus-quality dial.
Session creation and event-sending no longer require separate API calls. A new session can be seeded with up to 50 user_message and define_outcome events in a single create call.
Skills provide task-specific instructions to Managed Agents. You can now add up to 500 skills across all Managed Agents in a single session, up from the previous limit.
A cookbook for the new Managed Agents features is available, alongside the built-in "claude-api" skill for asking Claude Code about the updates directly, and full release notes.
You can set effort level per subagent by adding an `effort` field to the subagent's YAML frontmatter alongside `name` and `model`. This lets you tune effort independently from the parent session rather than relying solely on model selection.
A user asked for subagents to spawn with isolated sessions and return only short summaries to keep the main agent's context lean. Subagents already run in fresh contexts, and only their final message is returned to the main agent; the team is asking for specifics if users are still seeing context blowup.
A user requested a command to hand off prompts from the main agent into a fresh context. The response floats '/subtask' as a possible name, with a screenshot suggesting it may already be in consideration.
A user asked about accessing projects from their phone without being at their desktop. The '/remote-control' command keeps a session running on the local machine while allowing you to continue it from the Claude mobile app or web.
A user wanted to enable '/remote-control' globally without typing it each time. The setting lives in '/config' under 'Enable Remote Control for all sessions'.
A user wanted to start a session in the terminal and continue it in the Claude Code desktop app. The '/desktop' command does exactly that, transferring the active session over.
A user requested the ability to branch chat conversations. The feature already exists via the /branch slash command.
Version 2.1.217 adds emoji shortcode support to the prompt input. Type `:heart:` and it converts to the emoji; type a partial like `:hea` and suggestions appear as you type. The behavior can be toggled via the `emojiCompletionEnabled` setting.
The Claude Code Desktop app has reached the point where an Anthropic team member uses it for all daily coding. Key features cited: multiple workflow management, built-in browser and iOS simulator, file explorer, git, terminal, and easier settings.
Bun v1.4 will be the runtime's biggest Node.js compatibility jump since v1.0. One engineer pairing with Claude Code passed more of Node's test suite in a single month than the entire Bun team did across nine months between v1.2 and v1.3.
