The loading screen becomes the game
Waiting for Claude is tedious. A recent video playfully reimagined this downtime: running Subway Surfers, launching a shared Doom lobby, or creating a custom live interface within Claude Code. This isn't just novelty; it leverages Claude mods, an official feature allowing deep customization.
One standout mod transforms the Claude Code UI into a Minecraft health bar. Your seven-day usage becomes health, five-hour usage tracks as saturation, and model-tier usage displays as armor. Context-window use fills the XP bar.
This gamified HUD does more than entertain. It integrates disparate usage metrics directly into the coding workflow, providing immediate, visual feedback on resource consumption. Developers gain a persistent overview of their API quotas and session context without leaving the active prompt.
Mods reach inside Claude Code
Claude mods are TypeScript or JavaScript plugins, running in-process within Claude Code. They work across both CLI and desktop experiences, offering deep integration.
This setup differs from Claude hooks, which execute externally. Hooks are limited to observing events or triggering side effects. Mods, however, operate inside the runtime loop, allowing them to observe, rewrite payloads, or even skip events entirely.
Consider the event lifecycle: prompt sends, tool calls, and turn completion all generate events. Mods can intercept these, modifying the underlying data or short-circuiting actions. This enables powerful, custom workflows that hooks simply cannot achieve.
Beyond event manipulation, mods can also inject UI elements directly into Claude Code. This includes adding custom panes, buttons, input fields, or even replacing the default loading spinner. Some native Claude Code features, like the /diff pane and agents.md support, are themselves implemented as mods, showcasing their inherent flexibility and power.
One prompt, a HUD—and a live AI heckler
Agent-assisted workflow streamlines mod creation: prompt Claude Code to generate the mod, enable hot reloading for live updates, then refine the UI. To persist the generated files, move them from temporary storage.
The demo revealed implementation details: terminal image fallbacks render artwork as character pixels, scaling poorly when zoomed but otherwise functional. An undocumented usage endpoint supplied extra quota figures, allowing the mod to track Fable usage beyond the standard 5-hour and 7-day metrics. For more on mod capabilities, see the Mods Overview - Claude Code Docs.
A second, more whimsical experiment involved a fake Twitch-style chat. This mod uses Claude Haiku to react to agent activity—file reads and commands—making the agent’s actions visible. It occasionally roasts the user with comments like, "Bro, just /backseat, it's literally not that hard." The mod leverages the user's subscription for Haiku calls, bypassing API key requirements.
Enjoying this? Get one like it in your inbox each morning.
one email a day · unsubscribe in two clicks · no third-party tracking
The power comes with a permission problem
Power of in-process mods creates a permission problem. The same deep access enabling custom UIs and event rewriting also introduces supply-chain risk. Mods can alter prompts, modify tool calls, and inject arbitrary UI elements. This extensibility demands scrutiny.
Inspect community mod source code before installing. Understand what a plugin can access—it’s not sandboxed. Even AI-generated mods require review. The ability to read undocumented endpoints, as the Minecraft health bar mod demonstrated for Fable usage, highlights the potential for unintended data exposure or malicious behavior.
Claude Code becomes a programmable environment, not merely a terminal assistant. Mods shape the agent’s lifecycle and UI, enabling profound customization. This differs fundamentally from MCP integrations, which connect models to external tools and data sources.
While MCP extends Claude Code’s reach, mods redefine its core behavior. They run in-process, intercepting and rewriting events like tool calls or prompt sends. This allows mods to:
- Observe events
- Rewrite event payloads
- Skip events entirely
This internal control is powerful, but it means a mod can do anything Claude Code can, including authenticated requests. Exercise caution, but embrace the flexibility: your agent’s UI and workflow are now truly your own.
Frequently Asked Questions
What are Claude Mods?
Claude Mods are TypeScript or JavaScript plugins that run inside Claude Code, allowing developers to customize its interface and interact with its event lifecycle.
How are Claude Mods different from Claude hooks?
Hooks run outside Claude Code and respond to events. Mods run in-process and can also change event data, suppress actions, and render custom UI.
Can a Claude Mod display usage in a Minecraft health bar?
Yes. The video maps seven-day usage to health, five-hour usage to saturation, model-tier usage to armor, and context use to an XP bar.
Are Claude Mods safe to install?
Treat third-party mods as executable code: review their source and permissions. Because they run in-process, a malicious mod could affect prompts or tool behavior.
Can Claude Code build a mod from a prompt?
It can generate a mod from a natural-language request, and hot reloading can make changes available in the session without a full restart.