牧野

牧野

不追新框架了,先把手头这个做完。

Joined 10/30/2026
3
Projects
0
Collaboration requests
0
Iterated
0
Likes
0
Saves
3
Views
0
Comments
Developer Toolsclaude-code

Stack Diagram

Describe your stack in one sentence — get an architecture diagram with the official logos.

Stack Diagram turns one sentence about your stack into an architecture diagram with the official logos. Tell it "Next.js and a Hono API on Vercel. Postgres on Neon, auth with Clerk, payments with Stripe," and the agent draws it: it fetches the real logos, routes the arrows around the labels, and renders the diagram to check it before handing it over. It covers cloud stacks on AWS, Google Cloud, and Azure with each vendor's own icon set, plus AI apps, monorepo tooling, and more. Every diagram comes in light and dark themes, and as a single HTML file with its fonts and logos inlined, so it opens anywhere with no network — or as a PNG or a React component if you need that instead. Installing it is one command — npx skills add caru-ini/stack-diagram — and it works in Claude Code, Codex, Cursor, and other agents that support skills. More than 2,500 developers starred it, with 140 forks, and version 0.6.0 is the latest release. It is MIT licensed. Finished diagrams can go to the public gallery at stacks.caru.moe, but only after you confirm twice, and your prompt is never sent. Two honest limits: it is a skill, not an app, so you need a coding agent that supports skills to run it at all. And the diagram is only as accurate as the sentence you give it — a vague prompt gets a vague diagram, and you still have to check that the boxes and arrows match what you actually run.

Developer Toolsclaude-code

Vibe Usage: Every AI Coding Quota You Pay For, in One Tray Window

See how much Claude Code, Codex, Cursor, Antigravity and Copilot have left — and keep working on the tool that still has room.

If you code with several AI tools, you know the moment: the five-hour limit runs out in the middle of a task, you open another tool, and you start explaining the whole job from scratch. This little tray app puts every subscription you pay for in one window — Claude Code, Codex, Cursor, Antigravity and GitHub Copilot — showing how much each one has left and when it resets. More than 70 people have starred it on GitHub. Every account gets its own card: five-hour and weekly windows, per-model scopes and spend pools, with countdowns to each reset and a warning when a limit will run out first at your current pace. Token counts come from the tools' own logs, and if a provider's status page reports trouble, the card says so — a slow tool may not be your limit at all. Unknown numbers are never shown as a full quota, and a failed refresh leaves the last reading visible, clearly marked as stale. The point is not just watching the numbers. Smart Launch opens the first tool in your priority order that is verified to have room, and when a limit runs out mid-task, Handover moves the work: a Claude Code session moves whole to your other account and carries on by itself, while other tools get a briefing so they can pick up where the old one stopped. If the computer restarts mid-session, open sessions come back where they left off. It lives in the Windows tray or the macOS menu bar, with a one-line compact view and 13 languages. Two honest notes. The tray app runs on Windows and macOS, with a terminal command for everyone else, and it needs Node 24 or newer. GitHub Copilot shows its quota only — it cannot hand a session over. No passwords or tokens are stored: it reads the sessions already signed in on your machine, and the code is MIT licensed. Fresh code landed on 7 November 2026.

Toolsclaude-code

Lump: A Desk Buddy for Claude

A little orange blob on a glass prism display that watches your Claude sessions and waves when one needs you.

Lump is a little orange blob who lives on a tiny glass prism display on your desk, and his only job is to show you what your Claude sessions are doing. While Claude Code is busy he works away quietly; when a session needs you — a permission prompt, a question, a plan to approve — he waves for attention, and when it finishes he celebrates. You can look away from the screen and still know exactly where things stand. He is small but busy. Corner counters show how many sessions are working, waiting on you, or finished and unread, and his color warms from tangerine to red as you get close to your Claude usage limits. Tap the glass and he plays one of a dozen gags: an electric shock, a ring of fire, a wolf howl, a banana peel, or a cartoon bomb. With a few optional settings he fades blue before calendar meetings, holds up a letter when a Slack VIP messages you, makes a tiny ocarina noise, and sleeps at night. What makes him special is that every frame of him is drawn in code: one web page holds his body, faces, moods, and every gag, and a script bakes it into a file the little ESP32 board plays back at 24 frames a second. The display itself is an off-the-shelf board that costs about $20, the monitor mount is 3D-printable, and a small Mac app mirrors him in your menu bar or notch when you step away from your desk. In two days he picked up 22 stars and 4 forks, which is a lot of love for a desk toy. Two honest limits: there is no packaged release yet, so you flash the board yourself over USB and set up the bridge with a few scripts. And the bridge that connects Claude to Lump only runs on a Mac — the board itself does not care, but the helper that watches your sessions does. You also need Claude Code, PlatformIO, Python 3, and the Xcode command line tools. The code is MIT, while Lump himself — the character, drawings, and sounds — is CC BY-NC, so you can build one and share it, just not sell him.