frankee0920-rgb

Rhino Architectural Reverse Modeling — AI skill for Claude Code

AI community

A Claude Code skill for turning reference photographs of a building into a Rhino model that is editable and dimensionally defensible.

How to install Rhino Architectural Reverse Modeling

This entry records only its repository, not the path inside it, so there is no exact command to give. Open frankee0920-rgb/rhino-architectural-reverse-modeling and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Rhino Architectural Reverse Modeling does

A Claude Code skill for turning reference photographs of a building into a Rhino model that is editable and dimensionally defensible. A constraint set, not a staged procedure.

Alternatives in AI

Sentrux — Real-time architectural sensor that helps AI agents close the feedback loop, enabling recursive self-improveme 2.9k ★
Workflow Quick Reference — Model: Contractor (you direct, Claude orchestrates) 98 ★
Claude Harness — Deconstructing Claude Code — interactive source code analysis, guided wc/helpe walkthroughs, and deep architec 90 ★

README

Rhino Architectural Reverse Modeling — Mode A

A [Claude Code](https://claude.ai/code) skill for turning reference photographs of a real building into a Rhino model that is **editable** and **dimensionally defensible**.

Not photogrammetry. Not a pixel-perfect copy. Not a single carved solid.

What it optimises for

Most image-to-model workflows optimise for *looks like the photo*. That is easy to reach and hard to build on: the result is usually one boolean-carved blob whose dimensions are eyeballed, and which nobody — including the model that made it — can tell you is wrong.

This skill optimises for two different things:

Editable — one architectural decision equals one named, editable object. Changing the loggia depth must not require deleting the model and re-running a script.
Dimensionally defensible — every load-bearing dimension is either measured two independent ways, given by the user, or explicitly flagged as unevidenced and asked about. Inventing a principal dimension and labelling it low-confidence is forbidden.

Design

It is a **constraint set, not a procedure.**

Earlier versions prescribed staged workflows (`LEVEL_1/2/3`, then `Phase A/B/C`). Staging turned out to be scaffolding for a weaker model that becomes drag on a stronger one — it spends attention on process that should go into looking at the building. It was also ordered by geometric scale, which does not correlate with where the risk is: the largest dimension in a building can be the one with no evidence at all.

So the staging was deleted. What remains is what does not lose value as the underlying model improves:

Kept	Why
Domain facts	typological dimension ranges, construction logic — things a model will not reliably recall
Executable checks	each one can fail; each caught a real error
Output constraints	editability, naming, no fabricated dimensions
Tool traps	silent failure modes in the Rhino MCP layer