Archify draws architecture your agent cannot make up
A diagram skill built on a typed JSON IR with deterministic checks, so nodes trace to real source and two snapshots diff into added, removed and rerouted.
- Stars
- 13K
- Language
- HTML
- License
- MIT
- Age
- 5 months old
- Last push
- August 14, 2026
- Latest release
- v2.14.0 · August 11, 2026
Figures as of . Project site: tt-a1i.github.io.
Ask an agent for an architecture diagram and you get something that looks right. Whether the arrows correspond to calls that exist is a separate question, and usually the answer is partly. Archify's design is aimed squarely at that gap.
What it is
An MIT-licensed agent skill for Claude Code, Codex CLI, Cursor, OpenCode and Raven that turns a repository or a system description into an interactive technical map: five diagram types — architecture, workflow, sequence, data flow and lifecycle — with four visual presets, dark and light themes and optional finite motion.
Output is one self-contained HTML file, plus PNG, SVG, WebM and a 1200×630 share card.
Why it showed up now
v2.14.0 on August 11, four months after the first commit, 13,000 stars. The project publishes a gallery it calls the Proof Lab, which is the appropriate name for the thing that has to convince you.
How it actually works
Between the agent and the picture sits a typed JSON intermediate representation with deterministic checks. The model proposes topology; the IR validates it; the renderer draws only what survives. That is the whole trick, and it is why the project can use the word verifiable about output that a language model helped produce.
The consequences show up in the interactions. You can search nodes, optionally open revision-verified source for a node, trace upstream and downstream reach along exact routes, compare roles, and play guided stories — all without the diagram inventing topology to fill a gap. A guided tour that cannot introduce an edge that does not exist is a different artefact from a generated picture.
The feature I would use first is the snapshot diff: compare two validated snapshots as Before / Delta / After and get exact added, removed, changed, moved and rerouted facts. Reviewing an architectural change as a diff of validated graphs is strictly better than reading two diagrams side by side and squinting.
Try it
npx skills add tt-a1i/archify -g
Then ask your agent to map the repository's runtime architecture. The Proof Lab shows generated output before you install anything.
Where it is weak
Verifiable means the graph is internally consistent and node references check out — not that the abstraction is the right one. Deciding which components matter and which detail to omit is the hard part of architecture drawing, and that judgement still comes from a model.
Self-contained HTML with motion and an interactive graph is a heavy artefact. For a large system, "one file you can share" becomes a file people wait for.
Fifteen open issues, v2.14.0 in four months. The README carries two sponsors, one of them an API reseller with a referral link and a discount code, the other a company whose own agent harness supports Archify as a skill — disclosed, and worth reading with that in mind.
Sources
- tt-a1i/archify · GitHub · April 15, 2026
- Release v2.14.0 · GitHub · August 11, 2026
- diagrams
- architecture
- agent skills
- visualization