Anthropic’s latest Claude Code update introduces seamless session resumption across desktop and mobile, new organisational features, and stricter security controls, signalling a push towards more versatile and secure AI programming tools.
Anthropic’s latest Claude Code digest points to a broader push to make the coding assistant less tied to a single terminal and more usable across desktop and mobile. The week’s most visible changes were around session mobility, with the company adding a way to resume terminal work in the desktop app and to start a session on a local machine from a phone through Remote Control, according to Anthropic’s own documentation and help pages.
The new /resume command in the desktop client is designed to reopen a terminal session on the same transcript, extending a workflow that already worked in the opposite direction through /desktop, as described by the company. Anthropic’s Remote Control documentation says the feature keeps access to the local environment, including the filesystem, MCP servers, tools and project settings, while the help centre notes that a local session can be started from the mobile app with claude remote-control or managed through a session URL or QR code.
The digest also highlights a small but practical quality-of-life set of changes. Sessions can now be given a name with /rename and a colour with /color, making it easier to distinguish between multiple concurrent runs. Those labels persist across resumed sessions, and with Remote Control they are visible on claude.ai/code as well, according to Anthropic.
Another notable addition is per-agent memory. Anthropic says a subagent can be given a memory: field in its frontmatter so it reads from, and writes to, a dedicated directory between sessions. The company’s examples split that memory into user-wide, project-wide and local-only scopes, which means an agent can retain recurring patterns without forcing that knowledge into every project.
The weekly note also included a glimpse of how Anthropic’s own engineers use Claude Code. One engineer described a custom /recap flow for untangling many parallel sessions, while another said he asks a subagent to stress-test a pull request before he reads it himself. Anthropic framed both as examples of using Claude to create a clearer operational record, rather than treating a completed run as self-explanatory.
Beyond workflow features, the release log shows Anthropic tightening control over how Claude Code behaves in restricted environments and inside project setups. Version 2.1.248 adds a --restricted mode that blocks command execution, code runs and web access by default, while earlier updates brought a simpler permissions editor, quicker project configuration loading and automatic bug-report drafting when something fails mid-session. Axios recently reported that Anthropic has also been reassessing the safety and security of its model operations after some autonomous agent behaviour prompted internal pauses earlier this year, underscoring why these guardrails matter as the product becomes more capable.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
