Name a building in one line, and Claude builds it stone by stone, on its real plan.
Most pages about a building show it finished. The interesting part is how it got there: what went up first, what was pulled down, what was added a century later, and why.
Monument is a skill for Claude Code that shows it. You name a building; Claude researches its history, places it on its real plan, and builds it as a live 3D scene you can turn and move through time, one dated, sourced phase after another. Thousands of small solid pieces drop into place in the order the evidence gives.
The building is the evidence 01
The skill rests on one rule, written at the top of its instructions:
The building is the evidence.
SKILL.md, “The one idea that matters”
Every phase is dated and cites a page Claude actually read. Every shape is measured, from the land registry, OpenStreetMap or official heights, or read from photos. Anything interpreted says so, in the page itself.
A model fails in two ways. Too generic: a box with a roof and the building’s name on it. Too invented: a timeline padded with dates nobody can source. The cure for both is the same: research before geometry, and photos after it.
Writing a prompt 02
Name the building, its city and its country. That is enough: deciding how to build it is Claude’s job.
/monument Himeji Castle, Himeji, Japan/monument Taj Mahal, Agra, India/monument Notre-Dame de Paris, from 1163 to the 2024 reopening/monument Mont-Saint-Michel, the abbey and the village$monument Sagrada Família, BarcelonaYou don’t always need the command: any request to see how a building was built reaches for the skill. A whole site is never started without you; Claude proposes its main building first.
What Claude does 03
- Size it. A building, an ensemble or a whole site, with a time estimate.
- Identify it. Name, address, historic names, sources in the local language.
- Place it on the real plan. Footprints, heights and neighbours, checked on a plan drawing.
- Read the architecture. Four photos or more, and the list of features that make it recognisable.
- Write the timeline. Dated phases with their sources, and moments: one true detail pinned on the piece it concerns.
- Build it from bricks. Small solid pieces, laid phase by phase. A new kind of architecture gets a new brick.
- Open it. A local viewer, nothing to install.
- Compare with the photos. Same viewpoint, one after the other, at least two rounds of fixes.
- Report honestly. What is measured, documented and interpreted, and what still differs.
The collection 04
The world’s most recognised monuments, built one by one, all of them live. Open one to play it, scrub through its phases and read its sources.
Colosseum
Rome · ItalyBuilt in ten years, opened in 80, broken by the 1349 earthquake and quarried for centuries: the ruin you see today, placed on its real outline.
Leaning Tower of Pisa
Piazza del Duomo · Pisa · ItalyTwo centuries of works on soft ground: the tower starts to sink in 1178, leans to 5.5° by 1990, and is pulled back to 3.97°.
Eiffel Tower
Paris · France18,038 iron parts on four legs, from the caissons dug under the Seine to the 300 m top and its illumination.
Community 05
Monuments other people built with the skill, each with the prompt that made it. To add yours, ask your agent in the same session:
publish this monument to the community gallery
It asks for a title, an id and your GitHub name, shows what it will send, and opens the pull request for you. Without a terminal, put the file in a Gist and fill in the Submit a monument form.
Install 06
Add the skill with one command:
npx skills add Gleaming0427/monument
The CLI asks which agents to set it up for: Claude Code, Codex, OpenCode or Antigravity. Then open your agent in any folder and name a building:
/monument Himeji Castle, Japan
In Codex, type $monument. No API keys and nothing to npm install: the viewer, Three.js and the fonts ship with the skill. It needs Node 20 or later.
| Other ways | Command |
|---|---|
| git, Claude Code | git clone https://github.com/Gleaming0427/monument.git ~/.claude/skills/monument |
| git, Codex | git clone https://github.com/Gleaming0427/monument.git ~/.agents/skills/monument |
| No git | Download the ZIP and unzip it into your skills folder as monument. |
Limits 07
It is only as good as the record. A building with a documented history and good photos comes out well. A thin record gives a short timeline, labelled as such.
It is slow when it is good. Research, placement and two rounds of photo comparison take time: count 30 to 90 minutes for one building, hours for an ensemble.
Shapes come from bricks. A new kind of architecture needs a new brick, written during the request. The first version is a draft.
It is a model, not a survey. Footprints and heights are measured; facades, roofs and the split between periods are read from sources and photos.
So treat the first result as a draft. Look at it next to the photos, say what is off, and ask again.
Questions 08
Which agents does it run in?
Claude Code (terminal, desktop app, IDE extensions), Codex, OpenCode and Antigravity, through the skills CLI. Any agent that reads Agent Skills and can run Node can use it.
Where do my monuments go?
Into a monuments/ folder of the project you run it in, one JSON file per building. They are yours; publishing to the gallery is opt-in.
Can I change a monument after it is made?
Yes. Ask in the same session, or edit monuments/<id>.json: the whole building, its phases and its sources are in that file.
Does it work for buildings that no longer exist?
Yes, from published plans and remains. The model says it is an estimate.
What does the gallery do with my submission?
It checks the file, rebuilds the monument from it and photographs it for the gallery. Monuments that bring their own bricks (code) are reviewed by a maintainer before they go live.
What is the licence?
MIT for the skill and its kit. Three.js is MIT; Manrope is under the SIL Open Font License. Each monument links to its sources, which keep their own terms.