Pixel Agents for Claude Code

Three Claude Code terminals in, and you no longer remember which one is writing code and which has been silently waiting on your "yes" for five minutes. Pixel Agents turns your agents into a pixel office: you can see every status at a glance 🕹️

What it is and who it's for

Pixel Agents is a free open-source VS Code extension by Pablo De Lucca (8.5k+ GitHub stars, 75k+ marketplace installs). Every Claude Code terminal gets its own animated pixel character in a virtual office inside an editor panel: when the agent writes code, the character types at a desk; when it reads files, it "reads"; when it's waiting on your answer or permission, a speech bubble pops up above it. You don't have to touch Claude Code itself: the extension is a pure observer — through the official Hooks API, and if hooks aren't there, through the session JSONL transcripts.

Who it's for: vibe coders running several agents in parallel who are tired of flipping between terminals to figure out "who's stuck waiting over there". And honestly — anyone who just enjoys watching their AI team "work" in a pixel office.

Key features

• One agent, one character: every Claude Code terminal spawns its own hero (6 different characters out of the box).
• Live animations driven by real activity: typing, reading, running a command, waiting for input.
• Status bubbles and optional sound: you see and hear immediately when an agent finishes its turn or asks for permission.
• Subagents (the Task tool) show up as separate linked characters — the whole hierarchy is visible.
• An office editor: floors, walls, furniture; the layout persists across VS Code sessions.
• External asset packs: custom and third-party furniture, floors and walls can be loaded from any folder (loading your own character sprites is on the roadmap).
• Standalone mode: a local Fastify server opens the office in a browser — for tmux and remote sessions without desktop VS Code (right now you launch it from source with node dist/cli.js; there's no ready-made npx command on npm yet).

How to start in 5 minutes

1. Check the requirements: VS Code 1.105+ and the Claude Code CLI installed (Windows/Linux/macOS).
2. In VS Code open Quick Open (Ctrl+P) and run ext install pablodelucca.pixel-agents — or install it from the marketplace page.
3. Open the Pixel Agents panel and run claude in the terminal — a character shows up in the office and starts acting out what the agent is doing.
4. Fire up a second and third terminal with agents and turn on notification sounds — "who's waiting, who's working" is now visible without any flipping around. If you like, rearrange the office in the editor to suit yourself.

When to take it, when not to

✅ Take it if you run 2+ Claude Code agents in parallel and want to see all their statuses at a glance instead of poking through terminals. ✅ Take it if you keep missing the moments when an agent is waiting on a permission or a question — bubbles and sound solve that. ❌ Don't take it if you work with a single agent in a single terminal — Claude Code's built-in status is enough, and the office panel will just eat screen space. ❌ Don't take it → Claude Squad is better if you need orchestration rather than visualization: isolated workspaces on git worktrees, auto mode, change review. ❌ Don't take it if your main agent is Codex, Gemini or Cursor: today only Claude Code is supported, the rest are roadmap only.

The honest price

Completely free and open source (MIT license): the extension in the marketplace is free, there are no paid tiers and no "premium furniture", and the repo is open — you can fork it and extend it. You only pay for using Claude Code itself (a subscription or the Anthropic API is a separate line item). Tags: free / oss.

Gotchas

• Only Claude Code is supported: adapters for Codex, Gemini, Cursor and others are announced on the roadmap but don't exist yet.
• The project is young (the repo was created in February 2026) — features and behavior can shift from version to version; there are a few dozen open issues.
• It's a monitor, not an orchestrator: you still assign and untangle the agents' tasks yourself.
• The preferred mode is through Claude Code's Hooks API; the fallback mode based on JSONL transcripts runs on heuristics and is less precise about statuses.
• The characters live in a VS Code panel — on a small screen the office will feel cramped.
• The cuteness is addictive: it's easier to get lost watching the pixel office than to review the code the agents are "typing" in it.

🤖 Prompt accelerator

"Break the task into 2-4 independent subtasks with no shared files and run them in parallel as subagents via the Task tool: give each one a clear spec, a file list and an acceptance criterion. Before touching any shared file, stop and ask me. I'm watching progress in Pixel Agents, so after each subtask finishes, give me a short summary: what got done and what to check."