[yujiaxu-ops]

Linear: ENG-4880. These come from an agent's end-to-end run of image → model → remesh → rig → animate.

1. image-to-3d and multi-image-to-3d texture by default, as the API does

meshyd defaults should_texture to true on both endpoints, v1 and v2 (api_image23d_handlers.go). The CLI forced false (draft-first), with two effects:

• An agent that forgot the flag paid 20 credits for a white mesh, then had to run a separate retexture.
• make textures by default, so it disagreed with the command it runs.

Now:

• With no flag, should_texture is left to the API. The CLI adds its game-ready texture settings: enable_pbr: true and texture_resolution: "4k".
• --should-texture false turns texturing off and sends no texture settings. --data '{"should_texture":false}' does the same, because toDefaults now receives the parsed --data.
• A texture setting on an untextured run is refused before the POST, whether it came from a flag or from --data. The API answers that with a 400. The check runs on the merged payload, in src/internal/texturing.ts.

2. inspect faces --model <file.glb|.gltf>

The public task response has no face_count. meshyd stores the count, and the rigging 300k ceiling reads it, but it does not expose it. So the task sources could only ever answer unknown (exit 13).

--model counts the triangles of a local model from the glTF JSON alone, using accessor counts, so Draco- and meshopt-compressed files count too:

• every mesh instance of the default scene is counted;
• triangle strips and fans count as n − 2;
• points and lines count as 0.

Malformed files are refused with a usage error that names the file. A failing model suggests meshy remesh create --model-url <file>. A task source that answers unknown now hints at --model.

I checked it against the real outputs of that run:

File	Triangles counted
image-to-3d model	982,320
remesh output	15,500
rigged character	15,500

Tests

• pnpm typecheck is clean and pnpm test passes 644/644.
• New tests/texturing.test.ts checks the wire payload for both commands:
  • the default;
  • false set by flag or by --data;
  • conflicting settings refused with no POST.
• Model counting has unit tests (instances, nesting, strips, lines, no scene, .glb and .gltf, and six malformed cases) and command tests (pass, fail, the remesh suggestion, the unknown hint).
• Tests that pinned the old draft default were updated.

Docs: usage.md, SKILL.md, and decision D-083.

🤖 Generated with Claude Code