archify for Claude Code and Codex: architecture diagrams that pass checks before delivery
Who this is forProduct planners and developers who need a service architecture diagram for a planning review or a flowchart for a presentation.
Series · Claude Code & Codex skills, one at a time part 22 of 22, expand to see all
- pm-skills for Claude Code and Codex: a product planning skill pack, installed and run
- distilly for Claude Code and Codex: turn a person's judgment into a skill
- wondelai/skills for Claude Code and Codex: business book frameworks as agent skills
- yc-office-hours for Claude Code: a business mentor skill that asks before you build
- research for Claude Code: a skill that saves web research as sourced documents
- flow-map for Claude Code: a skill that keeps dated versions of your flow diagrams
- meeting-agenda for Claude Code: a skill for agendas built around decisions
- marketingskills for Claude Code and Codex: a 50-skill marketing pack, installed and run
- humanizer for Claude Code and Codex: a skill that edits out machine-like prose
- claude-ads for Claude Code and Codex: an ad account audit skill pack, installed and run
- blog-post for Claude Code: a skill that runs one blog post from sources to deployment
- blog-topic-judge for Claude Code: a skill that decides whether a topic is worth a post
- technical-writer for Claude Code: a skill with writing steps for each document type
- korean-writing-style for Claude Code: a skill that filters translationese from Korean
- card-news for Claude Code: a skill that builds mobile card news decks from one JSON file
- brunch-apply for Claude Code and Codex: a skill that prepares a Brunch writer application
- last30days-skill for Claude Code and Codex: research 30 days of public reactions
- nature-skills for Claude Code and Codex: a skill pack for journal figures and statistics
- notebooklm-py for Claude Code and Codex: drive NotebookLM from the command line
- insane-search for Claude Code: a skill that reads blocked public pages another way
- apps-script-collector for Claude Code and Codex: log daily market prices in Google Sheets
- archify for Claude Code and Codex: architecture diagrams that pass checks before delivery
archify is a skill that turns a system description or a codebase into an interactive HTML architecture diagram. It ships as a single skill with a Node.js CLI, and the repository is public under the MIT license. This post covers the problem it solves, who it suits, how to install it and make the first request, and the details worth knowing before you put its output in front of a review meeting.
What problem it solves
Asking an agent for an architecture diagram usually gets you something that looks fine at first glance. Then you notice an arrow that crosses a label, a box that overlaps another, or a layout that shifts completely the next time you ask for a small edit. The diagram has no source you can reason about, so every revision starts over.
archify works from a typed JSON source instead. The agent first picks a diagram type with guide, which recommends one of five types the README lists: architecture, workflow, sequence, data flow, and lifecycle. It then writes the JSON and runs finalize, which chains validate and deliver. Validation checks layout, paths, and label overlaps, and when something fails it returns rule codes and the items to fix as JSON. Only HTML that passes every check is kept, and the on-screen export menu offers PNG and SVG. Because the layout compiles from the same source, it stays consistent after edits.
Who it fits
The chapter rates four situations. Three cover most decisions.
| Situation | Fit | Reason |
|---|---|---|
| A single architecture diagram for a planning review | Good fit | Creates a diagram from a description |
| A flowchart for a presentation | Good fit | Can export PNG |
| Creating brand posters | Poor fit | Not a general-purpose drawing tool |
If you expect to drag boxes around by hand afterward, look elsewhere. The README states that direct drag-and-drop editing is outside its scope.
Install and first run
One command installs it globally, and Codex can also run it once without installing.
# Install globally
npx skills add tt-a1i/archify -g
# Codex: use once without installing
npx skills use tt-a1i/archify@archify --agent codex
According to the README, the skill lands in ~/.claude/skills/ for Claude Code and ~/.agents/skills/ for Codex CLI. Rendering and validation need Node.js 18 or later. After that, the first request is a single sentence.
Turn [system description] into an architecture diagram
The generation rules live in archify/SKILL.md, and the CLI commands run after node bin/archify.mjs if you want to call them yourself.
What trips people up
A passing diagram is not a verified system. archify validates the drawing, not your infrastructure, so the boxes and arrows reflect exactly the relationships you described. If your description is missing a queue or gets a dependency backwards, the diagram will be clean and wrong. Treat it as a picture of your claim, and check runtime behavior separately. Two smaller notes: package.json requires Node 18 or later, and the CLI queries a fixed manifest about every 24 hours to check for new versions. Set ARCHIFY_UPDATE_CHECK_DISABLED=1 if you do not want that call.
Where this comes from
This post adapts part of chapter 04-02 of The Claude Code & Codex Skills Guidebook. The full chapter covers the complete command table, including preview for validated versions, and how compare reports additions, deletions, and moves between two architectures, and the book has 52 more skills like this one. Read the full chapter in The Claude Code & Codex Skills Guidebook on Gumroad.
Frequently asked questions
- Who is archify a good fit for?
- It fits people who need one architecture diagram for a planning review or a flowchart for a presentation, since it builds the diagram from a description and exports PNG. It is a poor fit for brand posters, and its README says drag-and-drop editing is out of scope.
- Does archify check my real infrastructure?
- No. The diagram shows the relationships you describe, and its validation checks layout, paths, and label overlaps in the JSON source. Whether the running system matches the diagram is something you still have to check separately.
BuildnWrite helps teams build AI agents that keep running. About BuildnWrite ›