This technical example authors a stylized skeleton's geometry, 23 joints, vertex weights and walk clip in one interpreted AdaScript recipe. It is intentionally outside the public examples catalog; it does not appear in Studio recommendations.
Open this directory in Ada Studio, open Assets/Scenes/Walk.ascn and press Play.
The committed Skeleton-AdaScript.glb already contains the rig and Walk clip.
The scene uses the existing native importer, animation player and GPU skinning.
Select BindPose instead of Walk on the subject's Model3DSource to inspect
its undeformed pose. Showcase.ascn retains the original static model.
The character has 7,348 vertices, 4,284 triangles, 11 materials and 306 vertices with two nonzero joint influences. Walk lasts 1.2 seconds with 49 keys per track. The leg IK runs while authoring keys; runtime playback evaluates a regular clip.
Use a Studio build containing the procedural rig builder from
Ada PR #206. Pass the full contents of
Generators/Skeleton-Rig.ada as source to editor.model.generate, with a fresh
destination, such as Assets/Models/Skeleton-v2.glb. The tool never overwrites
an existing asset. Validate the new file and point the scene at its asset reference.
The native builder ships precompiled in the app. The embedded compiler generates bytecode for the VM; recipe execution needs no JIT, Swift compiler, Python, shell, Blender or cloud 3D service. Character anatomy, weight rules and leg IK are in AdaScript. The language agent itself may still use an external model.
SkeletonRigRecipe.ada holds the rig and motion rules; Skeleton.ada contains
static geometry generated by author_recipe.py. The composed Skeleton-Rig.ada
is self-contained and does not load those files at runtime.
Windows VM execution, buffer validation and native GPU playback have been checked. Physical iPad execution and a complete rebuilt Studio bundle remain unverified. The installed validation Studio predates the new generation API: it proves output playback; the actual extended generator was exercised in the developer harness.
These commands require a compatible local Ada engine checkout and Swift, and are not steps performed on an iPad. From this example directory, set the engine path explicitly; no local package dependency is committed:
$env:ADAENGINE_ROOT = 'C:/path/to/Ada'
$env:ADAENGINE_GRAVITY_PACKAGE_PATH = 'C:/path/to/windows-compatible/gravity-lang'
python script_rig.py test
python script_rig.py generate
python check_geometry.py
python check_rig.py --asset Assets/Models/Skeleton-AdaScript.glb --output Evidence/ScriptRig/buffer-check.jsonUse the repository's Windows-compatible Gravity override on Windows. On other
hosts, the engine's Sources/gravity-lang is the default dependency. The harness
copies actual production sources into an isolated package under the selected
engine's .build-tools directory and uses a standalone resource-scope adapter.
It does not build the full engine or Studio. After building the runner once,
python script_rig.py run executes it directly without invoking a compiler.
With this project open in Studio and its MCP server on port 2516:
python rig_capture.py --asset Assets/Models/Skeleton-AdaScript.glb --evidence Evidence/ScriptRig --live-only
Captures, reports, logs, workspace state and auxiliary sampled GLBs are ignored.
The committed preview and four small GLBs are the playback assets and comparison
fixtures. The older Swift author in NativeRig/ is a developer comparison baseline;
it is not needed to execute the AdaScript recipe. Its tests run with
python native_rig.py test using the same ADAENGINE_ROOT.
This recipe assigns an explicit rig to known generated parts; it does not infer anatomy or edit arbitrary imported meshes. The builder supports up to 128 joints, two weights per vertex and linear joint translation/rotation clips. Rest-joint rotations, scale keys, fingers, facial rigging, cloth simulation and retargeting are not implemented here. Check actual poses visually; import and buffer checks do not establish the absence of every self-intersection.
The original user-supplied character reference was labeled KayKit. This is a procedural study based on that reference, not an official KayKit asset package.