Claude Code: Remote Control, Auto Memory, Plugins & More

Anthropic shipped a wave of updates to Claude Code and Cowork in the last few weeks. No single headline feature - just a stack of meaningful improvements that compound. Remote session control from your phone. Scheduled recurring tasks. Two new plugin repositories. Auto memory that persists context across sessions. And some adoption stats that should get your attention.

Here's what changed and why it matters.

Remote Control From Your Phone

This one is more useful than it sounds. You can now access your active Claude Code session from your phone or any web browser via a slash command. Start a session on your laptop, walk away, and pick it up on your phone to monitor progress, answer questions, or redirect the agent.

The flow is simple: run the command in your terminal, get a link, open it on your phone. You see the full session - what Claude is working on, what it's asking, what it's outputting. You can respond to prompts, approve tool use, or kill the session entirely.

This matters for long-running agents. If you've kicked off a multi-file refactor and walked to get coffee, you don't need to rush back when Claude asks "should I also update the tests?" You answer from your pocket.

Scheduled Tasks in Cowork

Cowork now supports recurring scheduled tasks. Think cron jobs, but described in natural language and executed by Claude.

The use cases Anthropic highlighted: daily summaries of repository activity, recurring research pulls, file organization, email follow-ups. You define the schedule and the task description. Cowork handles execution on the cadence you set.

This is the kind of feature that's easy to overlook and hard to stop using once you start. If you're already running Claude Code for one-off tasks, scheduled tasks let you automate the patterns you keep repeating manually. "Every Monday morning, summarize all PRs merged last week and post to Slack" - that kind of thing.

Two New Plugin Repos

Anthropic released two new plugin repositories: one for knowledge work, one for financial services. Both are installable from the Cowork marketplace and - this is the important part - editable in natural language after installation.

You install a plugin, then modify its behavior by describing what you want changed. No code editing. No YAML wrangling. Just tell it what to do differently. The example Anthropic showed was an equity research idea generation plugin: install it, customize it to your coverage universe, and run it.

The plugin architecture itself is straightforward. Each plugin is a set of skills and agent definitions that get loaded into your Cowork environment. The marketplace is the distribution layer. Natural language editing is the customization layer. The combination means you can take someone else's workflow, fork its behavior through conversation, and end up with something tailored to your work without writing a line of config.

Claude Code Auto Memory

This one fixes a real friction point. Claude Code now automatically remembers project context across sessions using an editable markdown file.

Previously, every new Claude Code session started cold. You'd re-explain your project structure, your conventions, your preferences. Auto memory changes that: Claude writes relevant context to a markdown file (visible and editable by you), and loads it at the start of each session.

The file lives in your project's .claude/ directory. You can read it, edit it, delete lines you don't want persisted, or add context manually. It's not a black box - it's a markdown file you own.

This is the right design. Transparent, user-controlled, file-based. No hidden database. No opaque embeddings. Just a file that Claude reads and writes, and you can too.

# Auto memory lives here
cat .claude/CLAUDE.md

If you've been maintaining your own CLAUDE.md with project instructions, auto memory now supplements that with learned context. Your explicit instructions stay. Claude's observations get appended separately.

Ask User Tool Upgrades

When Claude Code needs to ask you a question mid-session, it can now render markdown diagrams and code snippets in the prompt. Previously, questions were plain text. Now Claude can show you a proposed file structure as a tree diagram, a code diff it wants you to approve, or a dependency graph - all rendered inline in the terminal.

Small change. Meaningful improvement to the feedback loop. When an agent asks "should I restructure the imports like this?" and shows you the actual code instead of describing it in prose, you make faster and better decisions.

The Stats That Matter

Anthropic shared a number: 4% of all public commits on GitHub are now authored by Claude Code. Their projection is 20% by end of 2026.

That's not "AI-assisted" commits where a human used Copilot for autocomplete. That's commits where Claude Code was the author - autonomous agent commits pushed to public repositories.

Whether the 20% projection holds is anyone's guess. But 4% today is already significant. It means Claude Code isn't a demo anymore. It's production infrastructure for a meaningful slice of the open source ecosystem.

Preview: Simplify and Batch

Anthropic teased two upcoming skills: Simplify and Batch.

Simplify takes complex code and breaks it down - not just refactoring, but genuinely reducing complexity while preserving behavior. Batch takes a task and fans it out across multiple isolated agents using worktrees, running them in parallel.

If you've used the worktree isolation pattern from the previous update, Batch is the automated version. Instead of manually spawning sub-agents, you describe the batch job and Claude handles the fan-out, isolation, and result collection.

Both are previews. No ship date. But they signal where Anthropic is heading: agents that manage other agents, with structural isolation built in.

The Bigger Picture

None of these features exist in isolation. Remote control makes long-running agents practical. Scheduled tasks make recurring agent work automatic. Plugins make agent behaviors shareable and customizable. Auto memory makes every session smarter than the last. Better ask-user prompts make human-in-the-loop faster.

Stack them together and the workflow changes. You're not "using Claude Code" as a tool. You're managing a team of agents that remembers what they've learned, runs on schedules you set, and checks in with you on your phone when they need a decision.

That's the trajectory. Each update nudges it forward.

• Claude Code: Remote Control, Auto Memory, Plugins & More - full walkthrough of all new features

Official docs:

• Claude Code Documentation - Anthropic's official Claude Code docs
• Cowork Documentation - scheduled tasks and plugin marketplace

This article is based on a Developers Digest video. All feature behavior is based on direct testing with Claude Code at time of publication.

Further Reading:

• Anthropic: Introducing Claude Code - official announcement and feature overview
• Claude Code Sub-Agents Guide - how to configure and deploy sub-agents
• Claude Code Worktrees - parallel development with git worktree isolation
• Claude Skills Documentation - reusable agent behaviors

Watch the Video