Archify is one skill that has an AI coding agent turn a description, Mermaid source or a code repository into a checked, standalone interactive HTML file. That file holds either one of ten diagram types or a one-page explainer that answers a question. It is meant for people using Claude Code, Codex, Cursor or OpenCode who want to understand, plan or share a system or process.
Why read it
The agent writes the complete candidate and then runs one `finalize` command. That command chains validation, delivery, a strict check and a real-browser check. The agent doesn't run a pre-validation pass of its own, and it may not count a failed exit as done. With one gate there is one clear finish line, which cuts out redundant steps and false claims of success.
When a gate fails, the agent reads the receipt, changes only the connected neighborhood of the problem, and reruns `finalize`. Applying layout hints gets exactly one try, with a rule to restore the backup copy if things get worse and an explicit ban on a second round. This keeps the agent from looping forever on fixes.
A table maps each kind of question (including everyday ones like 'where money or documents go') to a diagram type, an exact schema path and an exact example. A CLI `guide` command settles ambiguous cases. The agent picks a type and loads the right files without browsing directories.
The skill names the low-value moves agents tend to make and forbids them: help, doctor, starter validation, temporary diagrams, listing output paths, and planning coordinates in prose. It also warns that examples teach shape, not facts. Naming the specific bad moves works better than a general call for efficiency.
The agent must tell automated checks apart from actual perceptual review, and the skill gives a one-line model answer for 'was it verified?'. The output rules also make the agent disclose illustrative values, missing source evidence, or content dropped during repair. Without this, the agent could overstate what it checked.
How eagerly the skill makes explainer pages is set by a marker such as `[archify explainer: proactive]` in the user's global rules file. The skill offers this mode once per conversation, writes the line only if the user accepts, and offers `on-demand` to someone who declines so the question stops coming back. Proactive behavior stays under the user's control and can be undone by deleting one line.
The reply leads with a descriptive link to the HTML and is usually one line. Commands, repair rounds, hashes and check lists stay in the receipts. A worked example sets the target length, so the chat doesn't fill up with process narration.
The ERD guidance doesn't just order one domain per column. It explains how the viewer scales wide and tall canvases, and names the error code (`erd/domain-not-drawn`) the agent will hit if domains don't form rectangles. Knowing the reason lets the agent apply the rule to schemas the examples don't cover.
A model read each skill and found no calls or hand-offs between them.
| SKILL | KIND | STARTED BY | DESCRIPTION | LINKS | LICENCE |
|---|---|---|---|---|---|
| archify | skill | You or the agent | Create validated architecture, workflow, sequence, data-flow, lifecycle/state, ERD, class, tree, timeline, and waterfall diagrams as explorable standalone HTML with dark/light themes, motion and image/SVG/WebM export. Accept requirements or Mermaid; inspect repository evidence for code-backed diagrams. Use when visualizing systems, API calls, pipelines, schemas, histories, timings, or everyday steps, parts, relationships and states. Also turn an answer into a one-page explainer, with the conclusion first, structured panels, native diagrams and charts, when the user asks for a page, an explainer or an answer to share (做成一页, 用页面讲), or proactively when the user's rules opt in with an [archify explainer: ...] line. Not for dashboards. | 0 | Apache-2.0 |
From tt-a1i/archify at a406594, 2026-10-31. Skills read by Claude Sonnet 5.5 on 2026-11-07. Summary and techniques by Claude Opus 5.5 on 2026-11-07.