Claude draws you a working machine, and a $7 board drives your iPhone
Ask Claude how a rate limiter works and Anatomy has it draw you one: an interactive machine built from dozens of shaded parts, with the numbers moving as the parts move. It's a free skill and the fastest-moving tool in this issue. The author also publishes what it costs, which I respect. A watch movement took Claude almost two hours of work. A turbofan took fourteen. Ask for a quick figure first, then decide whether the full one is worth the tokens. ESP32 iPhone Agent is the strangest hack this week. A $7 microcontroller plays two parts at once, an AirPlay screen and a USB keyboard and mouse, which is enough for an AI model to see a stock iPhone and tap around it. No jailbreak. It's slow, and there's a trust question: the one-click installer on the project's site flashes firmware built from code that isn't in the public repo. A device that types into your phone should be one you can read. Build it from source. Last week I mentioned that the ArtCraft team rebuilt Figma alongside PhotoCraft. FigCraft made the radar on the same rule that earned PhotoCraft its spot: the maintainers publish how ready it is, and right now that's about half. Test your real Figma files before you trust it. The last pick is DigUp, which searches your computer by what's inside your files, so typing "the clause about pets in the lease" opens the right page of the right PDF. The model runs on your machine and nothing you index leaves it. It's two weeks old, so update the Mac app to 0.6.2, which shipped a high-severity security fix this week.
An agent skill for creating detailed, interactive isometric illustrations of real-world machines and complex systems. Generates polished SVGs with optional WebGL shaders and 3D effects.
The Lens
Anatomy makes Claude explain a technical idea by drawing a machine that does it. Ask how a token-bucket rate limiter works and you get an interactive isometric figure built from dozens of shaded SVG parts, with live numbers from the same model that moves them. WebGL handles fire, water and glow; a beta 3D mode turns parts in true depth. MIT, free, no API keys.
Install is one npx skills add line, and it also loads in Codex, OpenCode, Antigravity and MiniMax Code. The kit uses only Node's built-in modules, but the self-checks need Node 20.10+ and Chrome, driven headless to screenshot and audit every joint. Eight examples ship in the repo, from a dial gauge to a Raptor engine. Solo writers, devrel folks and small docs teams: free. Larger teams with designers will still hand final art to an illustrator.
The catch is time, which means tokens. The author publishes his own numbers: a watch movement took 1 hour 54 minutes of Claude working, the Raptor engine 10 hours 33 minutes, a turbofan 14 hours. The skill demands at least 40 parts per figure and won't trade detail for speed unless you ask. Ask for a quick figure first.
FigCraft: interface design and prototyping; an open-source, clean-room reimplementation of collaborative design workflows in pure Rust
The Lens
FigCraft is a from-scratch Figma lookalike written in Rust: frames, auto layout, components and variants, vector networks, styles, variables and clickable prototypes, in a native, offline app. It's free under MIT or Apache-2.0, with no account and no telemetry, and your files are plain .figcraft archives on your own disk.
Grab a release for any desktop OS, or run it in the browser. Real-time collaboration is a preview: share through the default public server or run your own figcraft-server. There are no accounts, so whoever has the link has access, and the server stores files unencrypted. A CLI and MCP server expose 400+ commands to AI agents.
No paid tier. Solo designers and developers who want offline mockups: free, try it. Small teams living in Figma multiplayer: not yet; Penpot is the more mature open source pick. Larger teams: no accounts, so no SSO or permissions.
The catch: Figma's .fig files open best effort only, and the project says that path hasn't been checked against real-world files. The maintainers rate it about half ready for real work, and its first version was built by parallel AI agents in about 25 hours. Test your own files before trusting it.
Let an AI model use a real iPhone with a $7 ESP32: AirPlay eyes, USB HID hands, any OpenAI-compatible vision model. No Mac, no developer account, no jailbreak.
The Lens
ESP32 iPhone Agent (now branded Handsy AI) lets an AI model drive a stock iPhone through a $7 microcontroller board. The board is both an AirPlay screen, so the model sees the phone, and a USB mouse and keyboard, so it can tap, swipe and type. The agent app on the phone works with any OpenAI-compatible vision model. Free, open source, no jailbreak. The board is a Seeed XIAO ESP32S3, $7.49 at Seeed. Flash it with arduino-cli or handsy.tech's browser installer. The app isn't on the App Store, so you build it in Xcode on a Mac. A free Apple ID can sign it but must re-sign every 7 days, and the one-tap hotspot join needs a paid developer account. Solo tinkerers: free, a fun weekend build. Teams doing app QA: mobile-mcp or Midscene skip the hardware, at the cost of a Mac and code signing. The catch: it's slow. The public firmware mirrors the screen over Bluetooth at 480x1040 and 5 fps at most, so steps take seconds. It types ASCII only. And the browser installer flashes firmware built from a commit that isn't in the public repo, so build from source to know what's typing into your phone.
Searches your files by what is inside them, not what they are named, with a model that runs on your own Mac, Windows or Linux machine.
The Lens
DigUp searches your computer by what's inside your files, not what they're named. Type "a dog on the beach" and it finds the photo; type "the clause about pets in the lease" and it opens the PDF page. It runs Google DeepMind's EmbeddingGemma 2 model on your own machine, and it's free and MIT licensed. The Mac build (Apple Silicon, macOS 14+) is stable. Windows and Linux are in beta, and the Windows installer still isn't signed. First launch downloads an 865 MB model, then reads the folders you pick: minutes on a Mac, hours for a big photo library on a PC without a good graphics card. Nothing you index leaves the machine. Code search is opt-in. There's no paid tier, and the model weights are Apache 2.0 too. Solo and small teams: free. Larger teams: free per machine, but every index is personal. The catch: it's two weeks old. Update the Mac app to 0.6.2, which fixed the battery drain and a high-severity bug that let a copied repo's git settings run a program during code search. A memory runaway on certain PDFs is still open, and the Linux beta's setup estimate can read folders you never listed.
Get the next issue in your inbox
Free. No spam. Unsubscribe anytime.