The shortlist
7 projects
The wider reading list: projects that stood out across the week, in their own words. Positions reflect the shortlist, not GitHub rank.
Also worth watching
Fast risers from off the boards — worth a look anyway.
The wider reading list: projects that stood out across the week, in their own words. Positions reflect the shortlist, not GitHub rank.
Fast risers from off the boards — worth a look anyway.
The weekly Star
An unfamiliar codebase can take a long time to become a picture in your head. You find the routes, then the database calls, then the jobs that run outside either. The folders tell you where things live. You still have to work out what connects them.
archify gives a coding agent a way to make that picture. The agent writes a structured JSON description of a system, and a command-line tool turns it into an interactive HTML diagram. The finished file opens in a browser, works offline, and can travel with the repository or go to a teammate.
We tried it on Starlogs, our Next.js app with three scheduled jobs, then on archify itself. Both produced working diagrams. The useful lesson was in what it took to get there.
Stars at each capture
Rank on each board
Our first diagram needed six validation rounds. The agent wrote the
specification, ran validate, read the errors, and revised it. Each failure
identified a particular problem and suggested a correction, so the next
attempt had a clear purpose.
That loop worked best when we kept the changes small. In one round, a broad layout revision turned three errors into six: new routes crossed edges that had previously been clear. Fixing the issue named in the report was more productive than rearranging the whole diagram.
The first scope mattered, too. Eleven nodes were enough to need real iteration. A diagram of the main request path gave us something we could check before adding the rest of the system.
Each of our outputs was about 800 KB. Inside the HTML file were the diagram, a legend, and guided views that dimmed everything except a selected path. Those views were especially useful: they let one diagram explain several parts of the system without asking the reader to take in every connection at once.
The export panel offered PNG, WebM, and a share card. Architecture was the diagram type we tested; the project also covers workflows, sequences, data flows, and lifecycles.
Revisions go through the specification. There is no drawing canvas for moving a box by hand. You ask the agent to edit the description and render it again, which keeps the diagram reproducible and makes the description worth keeping alongside the output.
The most important limit showed up quickly: layout validation does not establish that the diagram describes the code correctly. A readable diagram can still contain an invented connection or omit a job that matters.
We checked our connections against the source. Asking an agent to work from the README alone risks getting a diagram of the project's advertised shape, with the implementation details smoothed away. The specification needs the same scrutiny you would give any other account of how your system works.
There was a separate presentation issue. One diagram passed all nine
validation checks and still needed scrolling on a normal laptop screen.
The visual-check command addresses that part of the review. A wide layout
is also worth requesting at the start if the diagram is meant for a screen
or a slide.
archify is worth a look when you need to explain a codebase and expect the
explanation to change. Install the skill with
npx skills add tt-a1i/archify -g; the CLI requires Node 18 or newer. Start
with one path through the system, check it against the code, and keep the
specification. The HTML is the thing you share. The checked description is
what makes it useful.
Open-source web app that turns a single prompt into full interactive courses, built and steered by a multi-agent workbench.
Days on board: 3 daily · 7 weekly · 5 monthly
An agent skill that turns codebases or system descriptions into interactive, shareable HTML/SVG architecture maps.
Days on board: 2 daily · 7 weekly · 7 monthly
A collection of 166 ready-to-use scientific and research skills that extend AI agents supporting the open Agent Skills standard.
Days on board: 3 daily · 7 weekly · 0 monthly
A plugin for AI coding agents that steers them to write minimal code via a ladder favoring reuse, stdlib, and native platform features.
Days on board: 4 daily · 0 weekly · 0 monthly
Open-source C++ formatting library providing a fast, safe alternative to C stdio and iostreams with Python-like format syntax.
Days on board: 4 daily · 3 weekly · 0 monthly
A Claude Code plugin billed as an agent harness operating system, with an optional hosted GitHub App for private repositories.
Days on board: 6 daily · 1 weekly · 0 monthly · specialist pick
Open source local inference server that profiles your hardware, recommends fitting models, and runs them for your AI agents.
Days on board: 3 daily · 1 weekly · 0 monthly · specialist pick
OpenAI's machine-checked Lean formalizations of its Navier-Stokes and Euler finite-time-blowup papers — Clay Millennium alternatives (C) and (D), the week's frontier-math news auditable line by line.
Sketch Material 3 Expressive screens in the browser and out come vibe-coding prompts — 6k stars without a day on our boards; our editor builds in the same lane, so this call comes from inside the arena.