Claude Code's Hidden Features That Actually Matter

Here's the thing about Claude Code that nobody wants to say out loud: it's shipping features so fast that even power users are basically stumbling around in the dark. The WorldofAI channel just dropped a breakdown of hidden capabilities that are either buried in config files, undocumented, or just... never announced in any meaningful way.

And the wild part? Some of these features solve problems people are actively complaining about. Right now. Today.

The Memory Problem You Didn't Know Was Fixed

Let's start with the biggest one: context memory degradation. If you've been using Claude Code for a while, you've probably noticed it gets dumber the longer a session runs. Context gets muddy. Previous decisions get forgotten. It's frustrating.

Turns out there's a feature called /dream (also "autodream") that's basically REM sleep for your AI agent. The video creator describes it as a system that "automatically consolidates Claude's memory between sessions. It prunes stale or irrelevant notes versus useful insights and essentially cleans up the memory so it stays accurate and useful over time."

The feature was discovered in Claude's system prompt on GitHub, not in any official documentation. To enable it, you type /memory in your Claude instance and toggle on autodream. It's not even a proper slash command yet—you can't invoke /dream directly. But if you type something like "consolidate my memory using dream," it triggers manually.

This raises an obvious question: how many people are fighting with context issues when the fix already exists in their installation?

NO_FLICKER Mode and the UI Stability You've Been Missing

Another quality-of-life improvement that's apparently just... sitting there: NO_FLICKER mode. If your terminal has been jumping around or glitching out, there's an experimental renderer you can enable by setting CLAUDE_CODE_NO_FLICKER=1 in your config files.

The trade-offs are real though. You lose Command+F functionality and copy-paste behavior changes. But in exchange, you get stable performance, smoother rendering, and even mouse support in the terminal. Whether that's worth it depends entirely on how you work, but at least now you know the option exists.

The Codex Integration Nobody Told You About

Here's where things get interesting for people hitting rate limits. OpenAI released a Codex plugin that integrates with Claude Code, essentially giving you a second AI agent for code review without leaving your workflow.

This matters because Claude Code recently cut rate limits for Pro plans. The Codex integration means you can offload some work to a different model with different limits. As the video explains: "Once you have installed the plugin, you essentially can use the Codex reviewer tool as a normal read-only check. You can also make Codex question assumptions, have it hand off tasks to Codex for a second pass."

It's lightweight, uses your local Codex setup, and stays inside Claude Code. For refactors or high-stakes code changes where you want a second opinion, this is a legitimate workflow improvement.

Computer Use: Mac Only, Unfortunately

The computer use feature through MCP (Model Context Protocol) just launched last week. It lets Claude Code open apps, click through UIs, find bugs, fix them, and verify fixes—all from a single prompt. No more leaving the CLI to test things manually.

The catch: it's Mac-only right now, in research preview for Pro and Max plans. If you're on Windows or Linux, there's an unofficial plugin that replicates the functionality, but it's not from Anthropic.

Enable it with the /mcp command, right-click on computer use MCP, and you're in. The video creator notes it works with "anything you can open on the Mac with SwiftUI apps, local Electron builds, or GUI tools without a CLI."

Session Management: Teleport and Remote Control

Two features address the "I started this on my laptop but now I'm at my desktop" problem.

Teleport (/teleport) moves cloud sessions between devices. Started something on mobile or web? Pick it up in your terminal.

Remote Control lets you control a locally running session from another device—your phone, tablet, or browser. Run /remote-control to connect. Everything stays local on your machine; other devices just act as windows into that session.

You can even enable remote control for all sessions in your config file so any local session is instantly accessible from any device.

The Learning Features Hiding in Plain Sight

The /powerup command gives you interactive lessons inside the terminal. It's Claude Code teaching you how to use Claude Code as you go. The video describes it as "a really solid step towards making the tool more accessible, especially with hidden features that are present within your Cloud Code instance."

Then there's /output-style, which modifies the system prompt to change how Claude interacts with you. Three modes:

• Default: efficient software engineering
• Explanatory: adds educational insights while coding
• Learning: Claude guides you to write key parts yourself

If you're learning a new codebase, learning mode means "you'll actually understand what Claude is building instead of just accepting diffs." That's a meaningful difference.

Hooks, Forks, and Worktrees for Advanced Workflows

For people who actually want deterministic control over Claude's behavior, hooks let you run logic at specific points in the agent's lifecycle. You can:

• Load context dynamically each time Claude starts
• Log every bash command the model uses
• Route permission requests to WhatsApp for approval
• Poke Claude to keep going whenever it stops

Session forking (/branch command) lets you experiment without affecting your main workflow. Create a separate copy of your session, try new ideas, and if they work, merge them back.

Git worktrees support means you can run multiple parallel sessions in the same repository. The /worktree command or checkbox in Claude desktop lets you manage multiple sessions and branches simultaneously without the usual juggling act.

Effort Levels: Speed vs. Depth on Demand

You can set effort levels in skills or slash commands to control how long Claude thinks before answering. Override your session default to fine-tune responses for speed or depth on the fly. Some tasks need quick answers. Others need deep reasoning. Now you can actually control that trade-off.

What This All Means

The pattern here is clear: Claude Code has serious capabilities that most users will never discover because they're not surfaced in any intuitive way. Features that solve real problems—memory degradation, UI instability, rate limits, session management—are hidden in config files or undocumented commands.

This isn't unusual for developer tools, especially ones moving as fast as AI coding assistants. But it does mean the gap between power users and everyone else is widening. The people who dig into GitHub repos and Discord channels get a fundamentally different tool than the people who just install and use the obvious features.

Whether that's a bug or a feature depends on your perspective. But either way, now you know what you're missing.

— Tyler Nakamura, Consumer Tech & Gadgets Correspondent