August 28, 2026SkillsToolOpen Source

archify: The Diagram Skill That Refuses to Ship Unverified

archify added 4,260 GitHub stars today, the biggest agent-related jump on the trending board, and it does one narrow thing: turn a codebase or system description into a polished, interactive system map, directly in chat. It is an agent skill, built to be installed into Claude Code, Cursor, or Codex CLI, covering five diagram types, architecture, workflow, sequence, data flow, and lifecycle, with theme switching, animation, and PNG, SVG, and WebM export. MIT licensed, 22.9k stars total.

The design detail that explains the star count is the word verifiable. The agent does not draw pictures; it generates typed JSON, a validator checks it, and only verified output gets rendered into a self-contained HTML artifact. If the model hallucinates a component or an edge that does not parse, the validator catches it before you see a diagram. That is the same pattern that made the best skills of the past few months work, put a deterministic checker between the model and the deliverable, and it is the difference between a toy and a tool you can put in a design review.

It also does comparison, diffing two architectures to review what a change actually did, plus search and route tracing inside the rendered map. Which quietly repositions it from make me a pretty picture to an inspection layer for agent-written systems. As more codebases become things an agent modified while you were at lunch, a fast way to see what the structure is now stops being decoration.

Diagram skills are becoming their own genre, and the lesson from watching several trend is consistent: the ones that survive are the ones where output is checkable. archify is the most complete version of that idea so far. Repo at github.com/tt-a1i/archify.
← Previous
Small Models Have Arrived, Says the Essay. The Week Agrees.
Next β†’
JIT-Agent Makes the Harness Something the Model Writes at Runtime
← Back to all articles

Comments

Loading...
>_