tea-x-random/unity-game-skills / skills/unity-scene-composition
unity-scene-composition
VerifiedDefine and enforce Unity scene composition before asset generation or assembly. Use for composition.yaml, camera projection/yaw/pitch/lens/framing, visual layers, focal path, negative space, big-medium-small rhythm, prop density/distribution, color zones, occlusion budgets, contact shadows, BeautyCell/golden screens, registry asset placement, or scenes that are mechanically correct but visually weak. Covers sprite-on-3D/2.5D pixel density, Y-axis billboarding, feet pivots, blob shadows and quarter-view framing; diagnoses pasted-on, floating or mismatched-chunkiness sprites; and runs deterministic scene measurements plus visual critique. Pair with unity-game-layout, unity-asset-pipeline, unity-art-direction, unity-graphics and unity-ui-designer.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 12.6 KB |
| agents/openai.yaml | config | 213 B |
| references/composition-template.yaml | config | 4.5 KB |
| references/scene-measurement.md | markdown | 4.4 KB |
| references/screenshot-acceptance.md | markdown | 3.3 KB |
references/screenshot-acceptance.md
View on GitHub# Screenshot acceptance — measured gates + scored dimensions
Capture target-device screenshots from the actual Game View/camera, not only Scene View — at the target aspect plus at least one alternate aspect. Every composed-scene screenshot runs BOTH passes below; neither replaces the other.
## Pass 1 — MEASURED gates (deterministic, from Unity scene data via MCP)
Computed with `references/scene-measurement.md` — never eyeballed, never delegated to the VLM. Any violation is a **hard failure**: fix by moving/removing/rescaling (workflow step 9), not by re-generating art.
- **Density within budget** — sum of registry `density_cost` for renderers in the gameplay-camera frustum ≤ `density_budget.max_density_cost_per_screen`; interesting-object and family counts within their caps.
- **Occlusion within budget** — foreground renderers' projected viewport coverage of the gameplay area ≤ `occlusion_budget.gameplay_area_max_percent`; 0% over critical path and UI safe area.
- **Screen heights within role ranges** — each renderer's bounds-projected screen-height % inside `shape_rhythm.target_screen_height_percent` for its contract role.
- **Camera contract match** — active camera projection/yaw/pitch/ortho-size equal `camera_profile`; every visible asset's `camera_contract` matches (or is marked UI/background).
- **No unapproved assets** — every visible renderer resolves to a `registry.yaml` entry or is explicitly flagged as a gray-box placeholder in the ledger.
- **One pixel density** (sprite/2.5D scenes, when `pixel_density` is set) — ground `texels_per_unit` within `texel_tolerance_percent` of the project PPU (`art-spec craft.pixels_per_unit`); filter modes uniform (pixel track: Point everywhere, mipmaps off); one light model across sprites and ground.
- **UI safe area** — HUD RectTransforms inside `Screen.safeArea` at the target resolution.
## Pass 2 — SCORED dimensions (VLM, scene mode)
Run `critique_image.py <screenshot> --scene-mode --subject "<scene intent>" --reference <golden screen> --art-spec <spec>` (`unity-image-generator`). These are qualitative reads the numbers can't capture. **Calibration policy:** a low scene score triggers a bounded re-roll or manual review — it does not hard-block the scene.
- **Focal read** — first read is the gameplay objective within three seconds; first → second → third read order is clear; forbidden first reads (background detail, filler, decorative VFX) don't win.
- **Layer contrast** — interactables out-contrast decoration; background/ground spends less contrast/outline/saturation than gameplay; big/medium/small rhythm visible; ≥30% quiet area (unless genre forbids).
- **Grounding** — shadow direction/contact darkening agree across assets; no floating/pasted-on sprites; no baked-in PNG shadows disagreeing with the scene light.
- **Cohesion** — camera angle and asset perspective agree; palette zones match (quiet gameplay path, accented rewards); UI related to the world but more legible; no chunky-vs-smooth pixel mismatch, alpha fringe/halo, or obvious tiling wallpaper.
## Evidence
Record: screenshot paths, scene name, device resolution, camera profile id, the measurement JSON (pass 1), the scene-mode critique JSON (pass 2), registry commit/hash if available, and pass/fail notes. Store with the BeautyCell or scene QA artifact so session N+1 can resume from it.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |