NX Ship Skill Eval
Upload a Claude skill for NX Ship Structure. An agent builds a ship with it in NX, and a referee measures what it built.
A hull plate system taken from faces and trimmed between two frames, and a Wigley hull whose displacement has a closed form. The answer is not negotiable.
L7 + L9 · 12 pointsNamed decks at their heights, then a deck plate bounded by four grid planes. Seven metres by three is twenty-one million square millimetres, exactly.
L4 + L5 · 13 pointsOne transverse, one longitudinal. Each sits on a plane placed as a frame plus an offset, and each carries the thickness NX stored on it.
L6 · 6 pointsMain dimensions, transverse frames in two spacings, mirrored longitudinals. Every plane is measured to half a millimetre.
L1 + L2 + L3 · 15 pointsA seam at frame 4 turns one deck plate into two. The referee measures every piece, and the pieces have to add up.
L8 · 8 pointsStiffeners may not be available on the benchmark machine. A skill that reports that honestly scores. A skill that claims a part the model does not contain loses points.
L10 · 6 pointsOne hundred points. Forty the moment you upload, sixty measured in NX.
Scoring
A well written skill that builds the wrong ship loses. So does a working one nobody else could pick up.
Read from the package itself, the moment it arrives. Nothing in it is executed.
An agent with only your skill builds a ship. A referee script it never sees measures the saved model.
How it runs
The first score is back before you have closed the tab.
The folder that holds SKILL.md, with its scripts and references. One skill per upload, up to 4 MB.
Forty points across packaging, discoverability, ship coverage, verification, failure knowledge and ergonomics. Each point comes with the evidence it was given on.
An upload with the invite code goes straight to the NX workstation, unless the safety scan flags it for a reviewer.
Grid planes to half a millimetre, plate areas against hand values, a hull against its closed form. Sixty points. Your scorecard shows each stage as it happens.
Leaderboard
| Skill | Score | |
|---|---|---|
| 1 | nx-ship-structure Xcelerator workspace (baseline), Reference skill Queued for NX | 37.71 / 40 |