How to use the Archify agent skill: diagrams and explainer pages

What the Archify skill makes the agent do

Archify asks the agent to build a candidate diagram file, run a bundled command-line checker that validates it, render it in a browser check, and deliver a standalone HTML file with dark and light themes. Supported diagram types include architecture, workflow, sequence, data-flow, lifecycle/state, ERD, class, tree, timeline and waterfall. Diagrams can be exported as image, SVG or WebM.

The skill can also turn an answer into a single explainer page with the conclusion first, structured panels and native diagrams or charts. The skill description says it is not for dashboards or live monitoring.

• Diagram types: architecture, workflow, sequence, dataflow, lifecycle, erd, class, tree, timeline, waterfall
• Input can be a plain requirement or Mermaid; for code-backed diagrams the agent inspects repository evidence
• Each request gets its own folder named like .archify/<type>-<slug>-<YYYYMMDD-HHMMSS>/ in your working directory

Install the Archify skill

Option 1: download the ZIP from the Archify page in this directory and unzip it into ~/.claude/skills/ for a personal install, or into a project's .claude/skills/ folder to share it with one repository.

Option 2: in the Claude apps, open Settings, go to Capabilities, then Skills, and upload the skill there.

• Personal: ~/.claude/skills/archify
• Project: .claude/skills/archify inside your repository

How to trigger Archify with natural requests

You do not need a special command. Ask for the thing you want visualized in plain words. The skill description lists triggers such as visualizing systems, API calls, pipelines, schemas, histories, timings, and everyday steps, parts, relationships and states.

To get a page instead of a diagram, say so: ask for a page, an explainer, or an answer to share. The skill also recognises the Chinese phrases 做成一页 and 用页面讲. If your rules include an [archify explainer: ...] line, the skill's behaviour changes according to the mode set there (proactive, always-on or on-demand).

• Draw the request flow between the browser, API gateway and the orders service.
• Visualize the database schema for the billing tables in this repository.
• Make a timeline of the release history as a one-page explainer I can share.
• Explain this trade-off as a page with a conclusion first.

What you get from a run

The agent writes the candidate and the final <slug>.html into the request folder, and the delivery includes dark and light themes. Image, SVG and WebM export are supported. The skill asks the agent to review the result visually before claiming visual quality, so treat a passing check as proof of validity rather than a finished design asset.

Archify's delivery includes a bounded update check in its receipt. The skill's own files include examples for architecture, workflow, sequence, lifecycle, tree and timeline diagrams that the agent reads for shape.

What Archify needs to run

Archify runs scripts and uses the network. It needs Node.js to run the bundled .mjs scripts in bin/. For the browser visual check and image or WebM export it needs Chrome or a Chromium-based browser. To open the preview it uses the platform opener: open on macOS, xdg-open on Linux, or PowerShell on Windows.

• Node.js (runs bin/*.mjs)
• Chrome or a Chromium-based browser (visual check and image/WebM export)
• Platform opener: open, xdg-open or PowerShell

Safety notes: scripts, network and repository code

The skill runs bundled scripts and uses the network. Brand marks are optional: built-in brand logos are bundled, and an unknown site is only captured through an explicit two-stage command (brands capture <url>), not during normal rendering. Brand names and logos may be trademarks of their owners, so confirm your use is permitted.

For code-backed diagrams the agent reads repository source. If your code is confidential, do not ask for a repository-backed diagram. Read the scripts in bin/ before running them if your environment requires it.

• Do not use repository-backed diagrams on confidential code the agent should not read
• Brand capture of unknown sites is explicit and digest-pinned
• Files are written into the per-request folder inside your working directory

Tips for better results

State the scope in one sentence: which system, which flow, and which level of detail. For a database schema, say which tables matter. Ask for a repair if the first check reports a failure; the skill expects the agent to read the failure and fix the candidate rather than stop.

Use the page format when the answer needs a conclusion and supporting panels. Keep single-sentence answers and raw command output as plain text.

• Name the system and the scope in the first request
• Ask for a page when you want to share an answer
• Rerun with a new request to get a new folder rather than overwriting an old one