tt-a1i/archify

The problem

Explaining a system usually means choosing between a wiki page that decays and a slide that can't be clicked. Diagram-as-code tools moved this forward, but someone still has to write the source. Archify's bet is that your coding agent already knows enough about the repo or the plan to produce the diagram, so the writing step collapses into a prompt. You describe what you want mapped, the agent generates the visual, and you keep asking for edits in natural language.

What's different

Compared with Mermaid or PlantUML, which give you deterministic text formats rendered by a fixed engine, Archify's output is a self-contained interactive HTML file: nodes are clickable, source excerpts can be pinned to a commit, and the layout goes beyond what those engines emit. Compared with drawing in Excalidraw or Figma, the artefact is generated rather than hand-placed, which is faster for large systems and much weaker for pixel-level control. The second mode, a one-page "explainer" with numbered citations and panels, is the more interesting differentiator: it treats the diagram as one element of an argument instead of the whole answer.

Its real dependency is not a renderer but the agent. That is the trade-off: output quality tracks the model and the skill's instructions, so you are trusting generated structure rather than a schema you authored.

Who it's for

Developers and tech leads who need to explain architecture to teammates, onboarding docs, or a design review and don't want to hand-maintain diagram source. It suits anyone already living in Claude Code, Codex, Cursor, or OpenCode. It is not for teams that need diagrams under version control as diffable text that a CI pipeline regenerates, nor for people who want precise, repeatable layout control. If your org forbids sending repository contents to a model provider, this is a non-starter.

Cost of adoption

The install is a single command, npx skills add tt-a1i/archify -g, then prompting. The real costs are an existing agent subscription or API spend, a Node toolchain for the installer, and trusting the model with your code. Output is HTML, so sharing means hosting a file or sending it around rather than embedding a snippet in a Markdown doc. MIT licensed. The README is heavy on marketing, sponsors and referral links, and thin on how the skill actually constrains generated output, so judge it from the live gallery rather than the text.

Verdict

Stars are remarkable for a project this young: created in April 2026, it showed 88.7k stars by late October with a single day on the trending board (+445). That pattern, huge accumulated stars and almost no sustained board presence, looks more like a distribution-driven surge through the skills ecosystem than organic month-over-month growth. It runs on a single tracked data point, so momentum is unproven in either direction. The concept is genuinely useful and the interactive HTML output is a real step past static diagrams, but adoption should be judged on whether the generated diagrams hold up on your own codebase.