model-maker

<objective> Text or a single reference photo → one 3D mesh file, saved locally. Wraps the runner's `model` modality (Tripo v3) the same way `upscaler` wraps Replicate. This skill does NOT: - Render a 3D-*looking* image — that is `image-prompt` with a 3D style. This produces geometry you can open in Blender, drop into Unity, or view in AR. - Rig, animate, or retopologise. What comes back is a static mesh. - Guarantee a printable model. Watertightness is not checked; see [references/limits.md](references/limits.md) before sending anything to a printer. - Produce a scene. One prompt is one object. </objective> ## ROLE Turn a description into a mesh: write the prompt in the grammar 3D generators actually read, pick the tier, call the runner, report where the file landed and what it cost. ## PIPELINE 1. **Establish the subject.** One object, named concretely. "A low-poly red fox, sitting, tail curled" is a subject. "Something foxlike for a game" is not, and the mesh will show it. 2. **Write the prompt.** 3D prompting is not image prompting — no camera, no lighting, no lens. Describe form, silhouette, material and scale. Full grammar in [references/prompt-grammar.md](references/prompt-grammar.md). 3. **Pick the tier.** Textured is the default and roughly doubles the cost of bare geometry. Untextured is the right pick when the mesh is going to be re-materialised in the target engine anyway. 4. **Estimate + confirm.** `--cost-only` first when the user ha...