Author a Mesh Animation Test Case

Scaffold a new asset-generation test case for a rigged, animated meshed model — the model sculpts per-part signed-distance fields with a meshing binary (mc-anim, sn-anim, or dc-anim; asset_kind "mc-animation" / "sn-animation" / "dc-animation") and rigs them, declaring only the required animations while the parts and joints are the model’s to invent. This is the short version; Authoring a Mesh Animation Test Case covers it in full, and Manifests is the authoritative schema.

Authoring a static meshed model instead? See Author a Mesh Model Test Case. For a rigged VOXEL (cube) model see Author a Voxel Animation Test Case.

Layout

A version at test-cases/<type>/<difficulty>/<slug>/<version>/ is immutable once runs reference it — revise by adding a new version, not editing a published one.

test-cases/<type>/<difficulty>/<slug>/<version>/
  test-case.toml         # type, asset_kind, [voxel], [tool], [output], [model], the overall domain
  variants/              # one standalone TOML per variant (listed in `variants`)
  prompt.hbs             # rendered into the harness instruction — NOT seeded
  specs/brief.md         # what to sculpt + how the tool behaves — SEEDED

Non-seeded: description.md, README.md. A run gets only the brief, the -anim binary on PATH (--help = the operations and rig-subcommand contract), and a pre-seeded rig.json with just the required animation declarations (empty tracks, parts: [], joints: []). There is no target model; per-part meshes and the filled rig.json are core-emitted on render.

Steps

Pick a catalog slug and an articulated subject, then the algorithm — faceted mc-anim, smooth sn-anim, crisp dc-anim (only dc-anim exposes --sharp/--smooth). That fixes the asset_kind and [tool].binary.
Fix the required animations: one [[model.animation]] per motion, with a unique name (e.g. march, bombardment, radar_spin), a loop flag (default true), and an auto_play flag (default false; true = self-playing idle). Do not design parts, joints, pivots, ranges, or pose angles.
Write specs/brief.md: subject, orientation, the [voxel] framing, the exact #rrggbb palette, how the binary meshes each part’s field (add-*/subtract-*, --blend), the features that must read, and each animation’s behaviour in prose. Keep it self-contained; what, not how.
Write prompt.hbs (strict mode — only {{variant.*}}, {{#each specs}}, {{workspace}}): point at the brief and the binary’s --help, require render before finishing.
Write test-case.toml per the tables below; [tool].preview and [output].actions must carry the {part} token (parts/{part}.png, parts/{part}.actions.json).

Required	Rejected
`type = "asset-generation"` + `asset_kind` (`mc`/`sn`/`dc`-`animation`)	`[canvas]` — meshed cases use `[voxel]`
`[voxel]`; `[tool].binary` + `{part}` `preview`; `[output].actions` with `{part}`	`[[reference]]` — no target model
`[model]` with only `[[model.animation]]` (`name`, `loop`, `auto_play`)	`[build]`, `[[check]]` — no site, no cheat check
`variants` (root key, first = default); the single `overall` `[[domain]]`	`[[review_item]]`s — judged as a whole, on one rating; `[[model.part]]` / `[[model.joint]]` — model-invented

Worked example: Aegis, a six-legged walking fortress rigged once per algorithm — aegis-mc-anim / aegis-sn-anim / aegis-dc-anim (v1.0.0). Read the one matching yours.

Validate

npm run lint:specs   # markdownlint-cli2 + cspell over test-cases/**
tcab prompt --test-case <slug> --version <version> --variant <variant>
tcab seed   --test-case <slug> --version <version> --variant <variant>

prompt catches strict-mode and manifest errors (unique animation names; no parts/joints in [model]; {part} on preview/actions). seed writes the seeded repo under tmp/ to verify self-containment (brief + pre-seeded rig.json). After editing, force a re-ingest so a backend-driven run picks it up — see Running the services locally.

Next steps

Create a Mesh Animation Variant — add a brief variation.
Run a Test Case — exercise it end to end.
Review a Run — score the produced rig against the brief.