The July 20-24 release added a live iOS Simulator pane for macOS and a separately installed plugin that scans code and proposes developer-reviewed patches.
By RuntimeWire Staff · Published
Primary source: Claude Code Docs
Why it matters #
The July additions let Claude Code inspect a running iOS app and scan repositories for vulnerabilities, while keeping simulator control, finding selection and patch application with the developer.
According to Anthropic's July release notes, Anthropic, led by co-founders Dario and Daniela Amodei, paired a public-beta iOS simulator pane with a separately installed security plugin. The additions extend Claude Code Desktop beyond source editing into visual testing and repository review. Anthropic was founded by siblings Dario and Daniela Amodei and other former OpenAI employees. The documented release covers iOS on macOS.
The July 20-24 release added two targeted features to Claude Code Desktop: one lets the coding agent inspect a running iOS application, while the other brings vulnerability scans and proposed patches into the coding session.
That direction carries the imprint of the co-founders, who started Anthropic around the belief that capability and safety should be developed together. Daniela Amodei described the founding motivation in a July 2023 interview with Stripe as wanting to "help make AI systems safer from day one." Claude Code puts that principle under commercial pressure because useful coding agents need permission to read repositories, execute commands and interact with development environments.
Claude gets a view of the iPhone screen
Anthropic's release notes describe the iOS Simulator pane as a public beta in Claude Code Desktop for macOS on Pro, Max and Team plans. It requires Xcode with the iOS platform installed and Claude Desktop version 1.24012.0 or later. When Claude builds, launches or checks an application in Apple's simulator, the pane opens beside the conversation and streams the device screen live.
Claude can tap through an onboarding flow or inspect its own changes while the developer watches and can take control. That gives the agent a visual feedback channel beyond source files and command-line test results. It can check whether screens render, controls respond and a user can complete a flow inside the running application.
The distinction matters in application development. A model can generate interface code that compiles while missing layout failures, broken navigation or controls that do nothing. Simulator access lets Claude compare its implementation with the application's behavior before presenting the work for review.
Anthropic still leaves the developer in the loop. The user initiates the task, watches the simulator and can operate the virtual device directly. The July release also stopped Claude from invoking /verify
, /code-review
and /deep-research
on its own; users must call those tools explicitly.
Device interaction is becoming another competitive surface for coding agents. OpenAI described its Codex desktop app as a command center that supports delegating long-running work to multiple agents. GitHub said its Copilot coding agent could work asynchronously in the background and return a draft pull request for review. Anthropic's July addition is narrower: it gives Claude Code Desktop a live view into one Apple development tool on supported Macs.
Security scans enter the coding session
The separately installed Claude Security plugin runs a multi-agent vulnerability scan inside a Claude Code session. Anthropic's release documentation says agents map the repository's architecture, produce a threat model, search for vulnerabilities and independently review findings before generating a report.
A developer can scan an entire repository or limit the job to a branch diff, pull request or individual commit. Selected findings can then be converted into proposed patches, which the developer reviews and applies.
Giving a coding agent repository and shell access creates its own security exposure. In a GitHub security advisory published on September 9, 2025, Anthropic described a high-severity flaw affecting Claude Code versions earlier than 1.0.105. The advisory said a malicious Git user.email
value could trigger arbitrary code execution before workspace trust was accepted. Anthropic identified version 1.0.105 as the patched release.
In a February 25, 2026 report, Check Point Research said malicious Claude Code project configurations could enable remote code execution and theft of Anthropic API credentials. The firm attributed the attack paths to abuse of hooks, MCP servers and repository-controlled environment variables, including redirection of API traffic through a malicious ANTHROPIC_BASE_URL
. Check Point Research said it collaborated with Anthropic to ensure the vulnerabilities were fully remediated.
The July release tightened several related boundaries. More Bash permission checks now fail closed, including commands with certain redirects and commands longer than 10,000 characters. Anthropic also added a filesystem-isolation setting and retained human approval for patches produced from security findings.
The plugin's value will depend on false-positive rates, severity ranking and patch quality. Anthropic's release documentation describes the workflow without publishing independent measurements for those outcomes. Requiring developers to choose findings and apply patches limits the cost of a poor result, though it preserves much of the review burden.
For small engineering teams, the iOS pane and security plugin reduce handoffs between implementation, application testing and vulnerability review. Developers still approve access, decide which findings matter and review every proposed patch. The July release shows how the Amodeis are translating Anthropic's safety argument into product controls around a coding agent with growing access to the developer's machine.