AINEWS
Archify reads a codebase and emits a clickable HTML system map with structural checks. It just hit GitHub Trending 1, but does not watch live systems.
When the architecture diagram in a software team's slide deck goes stale, it usually happens within weeks, because the code underneath has moved on. A new open-source project called Archify tries to fix that by reading the code repository itself and emitting an interactive HTML system map after a structural check. The project has hit GitHub Trending #1 with roughly 31,400 stars and about 1,900 forks, a sign that AI coding agents are absorbing a developer-experience category that used to live in slide decks.
Archify is built as an Agent Skill, the plug-in format used by AI coding agents like Claude Code, Codex, Cursor, and OpenCode, rather than as a stand-alone editor. According to the project's README and its SKILL specification, it accepts either a code repository or a natural-language system description and walks the code to produce five diagram kinds: architecture, workflow, sequence, data-flow, and lifecycle. The output is a self-contained HTML file with PNG, SVG, WebM, and share-card exports and a light/dark theme toggle; a live example is on the project's site.
Archify validates a JSON intermediate representation, capturing schema, layout, and label/edge conflicts, before it emits the diagram, and it exposes a Before/Delta/After diff between two snapshots so the diagram can ride along with code review. Mermaid and D2 plugins still expect a human to hand-edit text to fix arrow tangles on a complex graph; Archify's design assumes the agent is doing the editing.
The README and the QbitAI profile both list limits the project does not try to hide. Archify is not a live runtime monitor. It cannot prove the diagram matches a running system; the JSON IR is a static analysis of the code, not of the deployed process. Some users report that the per-node detail panes are small, and that performance gets rough on lower-vision models, with DeepSeek-flash variants called out specifically in coverage. Model cost is non-trivial for a first-pass render of a large repository, and the project does not pretend otherwise.
QbitAI's aggregator-style framing of "X 多条10万赞正向测评" is the kind of social signal that does not survive a careful re-count, so the traction is best read as a Trending slot, roughly 31,400 stars, and roughly 1,900 forks, not as a quantified endorsement. An Agent Skill with a validated JSON IR and a snapshot-diff mode is now a recognizable pattern in developer tooling, large enough to top GitHub Trending in days.
The developer behind Archify, who goes by 也无风雨也雾晴 on X and posts on V2EX under the same handle, has been open about his vocational-college background in computer applications (2018), topping the transfer exam (the Chinese 专升本 pathway that moves a two-year college graduate into a four-year software-engineering program) at Chongqing University of Posts and Telecommunications, and a long stretch of rejected offers, including, he writes, named companies like ByteDance's Feishu and Yuanfudao, once the 专升本 record surfaced in employment background checks. He later received a Shanghai big-tech offer after Archify took off, and the project is, in his own framing, a public proof-of-skill artifact. The QbitAI profile paraphrases Manus COO Cz Chen as a personal encourager in the pivot, and QbitAI says it has a follow-up interview lined up; both the Manus quote and the broader personal claims are reported through the outlet and are not independently verified at the post level.
The next thing to watch is whether the same JSON IR + diff pattern shows up in adjacent developer-experience lanes: API surface documentation, schema migrations, infra-as-code dependency maps. If it does, the diagram is not the product; the validation pass is.