/powerup

Problem: Claude Code has way more features than most people will ever find. Hooks, sub-agents, plan mode, rewind, worktrees, skills, MCP servers, cloud tasks. New stuff ships every week, and the average user pokes at maybe a fifth of it. Difficulty is not the wall. Discovery is. What you can't see, you can't use.

Anthropic shipped a fix on April 1, 2026. It is an interactive lesson system that runs inside the shell, and the command is /powerup. The community said nothing about it. Not because the feature is weak. Because a terminal Tamagotchi burned through every headline the same day.

What /powerup Actually Is

The changelog for Claude Code v2.1.90 gives one line. From the v2.1.90 changelog: Added /powerup, interactive lessons teaching Claude Code features with animated demos.

Nothing else. No Anthropic blog post. No docs page of its own. No tweet. As of April 2, 2026, the commands reference page still doesn't list it. The feature sits in the binary, shows up in the changelog, and ships to everyone on v2.1.90.

Unpacked, that one line means something real. /powerup is the first in-terminal learning system for Claude Code that Anthropic themselves built. Type /powerup and you don't get a wall of text or a link to a doc site. You get lessons, with animated feature walkthroughs, rendered inside the same session you were already working in.

No third-party plugin. No community side project. A tutorial system built straight into the CLI by the people who make it.

Why It Matters More Than a Terminal Pet

Look at the timeline. Anthropic pushed two releases inside 22 hours on April 1, 2026:

• v2.1.89 (01:07 UTC): /buddy, the terminal Tamagotchi. 18 species, five rarity tiers, shiny variants, hat unlocks, stat systems with categories like CHAOS and SNARK. Within hours, someone spun up a Solana memecoin ($Nebulynx) tied to the rarest variant.

• v2.1.90 (23:41 UTC): /powerup, interactive lessons with animated demos. Zero community articles. Zero Reddit threads. Zero Twitter viral moments.

What that gap tells you about April Fools' developer culture is the whole story. A collectible pet with a 0.01% shiny legendary drop rate spawns memecoins and reverse-engineering repos. A learning tool that targets one of the best-documented pain points in Claude Code onboarding lands in silence.

The pet is entertainment. The lesson system is the feature you'll reach for six months from now.

The Problem Being Solved

Spend a week in Claude Code and this happens to you. You stumble on a feature you wish you'd known about from day one. Plan mode rewires how you attack hard tasks. The skills system turns Claude from a general assistant into a narrow expert. Voice mode lets you speak to your terminal. None of these are small. They change the shape of your day.

Discovery is the real hole. Here is what your options looked like before /powerup:

See the pattern. Every resource besides /help drags you into a browser. And /help is just a list of commands. It doesn't tell you what any of them do or show them in action. Want to know how interactive mode differs from auto mode? You context-switch out of the exact tool you're trying to learn.

/powerup cuts that loop. Stay where you are. Watch features play out as animated examples instead of reading static prose. The terminal is the classroom.

How /powerup Probably Works

The lesson list and the UI flow aren't documented yet. The command dropped at 23:41 UTC on April 1 and the docs haven't caught up. The changelog plus Claude Code's architecture give us enough to sketch a good guess.

Claude Code's terminal UI sits on top of React + Ink, the framework that renders React components as terminal output. A v2.1.88 source leak the day before /powerup shipped exposed the full 512,000-line TypeScript codebase and confirmed it. Ink is what renders every rich UI element in the tool. The /model selector, the /config panels, the /diff viewer. Same pipeline.

Those "animated demos" are almost certainly Ink components playing back terminal animations. Keystrokes appearing one at a time. Command output streaming in. Feature walkthroughs running as if a human were driving. A screen recording in spirit, but rendered natively by the same React components Claude Code ships with.

Running /powerup probably opens a lesson picker you steer with arrow keys and Enter. Same shape as /model and /config. Pick a lesson, read a short explanation, watch the animation, move on. Keyboard only, no mouse, consistent with the rest of the tool.

What the Lessons Probably Teach

Anthropic hasn't published a curriculum. The command is described as teaching "Claude Code features," and there is a clear set of capabilities that benefit far more from watching than from reading:

Beginner territory: Basic context management, /clear and /compact, the CLAUDE.md memory system, plan mode toggling with Shift+Tab, model selection.

Intermediate territory: The skills and custom commands system, hooks (PreToolUse, PostToolUse), sub-agent orchestration, MCP server configuration, /rewind and checkpointing.

Advanced territory: Worktrees and parallel sessions, auto mode and permission management, /schedule and cloud tasks, the SDK and headless mode.

Animated demos earn their keep on workflows that resist prose. What does /rewind actually look like in use? What happens on screen when you flip plan mode on? How does a sub-agent spawn, run, and hand back results? Those are the moments where seeing beats reading by a wide margin.

What Else Shipped in v2.1.90

The lesson system wasn't alone in this release. v2.1.90 is one of the beefier Claude Code updates in months.

Critical fixes:

• Fixed an infinite loop where the rate-limit dialog would crash sessions by repeatedly auto-opening
• Fixed --resume causing a full prompt-cache miss (regression since v2.1.69), meaning users with deferred tools or MCP servers were paying a performance penalty every time they resumed a session
• Fixed auto mode ignoring explicit user boundaries like "don't push" or "wait for X before Y"
• Fixed Edit/Write failing with "File content has changed" when a PostToolUse format-on-save hook rewrites files between consecutive edits

Performance wins:

• Eliminated per-turn JSON.stringify of MCP tool schemas on cache-key lookup
• SSE transport now handles large streamed frames in linear time (was quadratic)
• SDK sessions with long conversations no longer slow down quadratically on transcript writes
• /resume now loads project sessions in parallel

Security hardening:

• Fixed PowerShell tool permission bypasses: trailing & background job bypass, -ErrorAction Break debugger hang, archive-extraction TOCTOU vulnerability
• Removed Get-DnsClientCache and ipconfig /displaydns from auto-allow for DNS cache privacy

The --resume prompt-cache fix by itself is a big deal. If MCP servers or deferred tools are part of your setup (and plenty of power users lean on both), every resume since v2.1.69 was silently paying a performance tax. That tax is gone. The SSE shift from quadratic to linear also lands hard on anyone running long sessions with heavy tool use.

What It Says About Anthropic's Plan

Adding a tutorial system to a CLI is a product maturity move. Early developer tools assume users will read the docs and figure the rest out. They ship a getting-started page and let the community cover what's missing. A tool that builds structured onboarding into itself is a tool whose makers have decided to optimize for the next wave of users, not just the early crowd who already know every trick.

Community-built learning has been stacking up for months. An interactive browser tutorial at claude.nagdy.me. A $20/month course at ccforeveryone.com. A Vanderbilt class on Coursera. GitHub repos with 41 diagrams and CLI quizzes. The demand for better onboarding is real and loud.

With /powerup, Anthropic is saying something plain: the learning curve is steep, and we'll own fixing it. Education isn't getting outsourced to the community anymore. It's being baked into the product.

The source leak sits in the background of all this. Developers had to reverse-engineer 512,000 lines of leaked TypeScript to surface hidden features like the buddy pet system and the internal kairos metrics. The new lesson system is the opposite stance: decompiling the binary shouldn't be how people learn the tool.

The Invisible-Feature Problem

New slash commands ship constantly. Voice mode. Interactive mode. Plan mode. The skills system. Hooks. Sub-agents. MCP integration. Every one of them is strong. Every one is underused because discovery runs through changelogs most people never open.

Even the official docs admit that "not all commands are visible to every user. Some depend on your platform, plan, or environment." Which means Claude Code carries a class of features that are effectively invisible unless you already know what to look for. /powerup is the first attempt to close that gap from inside the tool itself.

How well it lands depends on execution. The size of the initial lesson set. How often Anthropic pushes new ones alongside feature launches. Whether the animations actually beat reading prose. The direction is right either way. You want to learn a feature the moment you need it, not from a browser course you half-finished three weeks ago.

Try It

Update Claude Code to v2.1.90 or newer:

npm update -g @anthropic-ai/claude-code

Then run:

/powerup

If the command is missing, check your version with /stats or claude --version. There may also be plan restrictions like the ones on /buddy (Pro/Max tier). Anthropic hasn't confirmed plan availability yet.

For readers who want to go further than the lesson system covers, our configuration guide walks through a full setup, and the Claude Code introduction lays down the groundwork you need before touching the advanced features.

The release nobody talked about on April Fools' might be the one you come back to all year. A pet in your terminal is a one-day story. Knowing the tool you work in is a forever one.