mobile-glb-budget

Purpose

Translate "make it work on mobile" intent into a concrete numeric planning ceiling table (triangles, vertices, GLB file size, texture pixels, materials, draw calls, animation channels, bones, morph targets) per device tier, scene scope, and target context. Pin this budget before scene mutation so downstream scene planners consume it, polycount-budget-checker and material-quality-checker validate counts and textures against it, render-export-evidence measures compliance, glb-web-handoff consumes it for web delivery, and glb-animation-handoff owns animation contracts.

Quick start

• pick the device tier (low-end Android, mid-mobile, high-mobile, desktop-fallback)
• read the budget row for that tier
• pin the budget into the scene plan before mutation
• treat any later violation as a blocker, not a warning

When to use

• before scene plan is locked
• before any GLB export attempt
• when user says "for mobile" / "for the web" / "phone" / "low-end devices"
• when designing a scene that must run alongside other web content

When not to use

• desktop-only renders without web handoff → use ../polycount-budget-checker/SKILL.md (Delegates out-of-scope requests to polycount-budget-checker)
• offline pre-rendered video (no real-time budget needed) → use ../material-quality-checker/SKILL.md (no real-time budget needed)
• replacing engineering load testing (this skill sets a planning budget, not a real-world benchmark) → use ../glb-web-handoff/SKILL.md (this skill sets a planning budget, not a real-world benchmark)

Trigger phrases

• "what's a good polycount for mobile"
• "how big can the GLB be"
• "will this load on iPhone"
• "what should I aim for on a phone"

Prerequisites / readiness

• intended device tier known (or user willing to pick one)
• expected scene scope known (single product / multi-product / environment)
• GLB target context known (hero card / e-commerce viewer / AR / configurator)

Input schema

Required inputs

Optional inputs

Assumptions to confirm

• Budget is a planning ceiling, not a guaranteed runtime performance result.
• Field-of-view, post-processing, and shader complexity affect real GPU cost beyond polycount.
• Mobile GPU memory varies wildly; budgets here favor the lower end of the tier.

Output schema

Primary output

Numeric budget table sourced from references/budget-tiers.md for the chosen device tier, scene scope, and target context covering all 9 metrics: triangles, vertices, GLB file size, texture pixel budget, distinct material count, draw calls, animation channel count, bones, and morph targets.

Secondary output

• a "violations block handoff" rule list (sourced from references/violation-rules.md)
• a fallback degradation plan specifying what to drop first if a violation appears (sourced from references/degradation-plan.md)
• explicit assumptions about device tier, network, and FOV
• a named handoff to the next skill in the workflow

Evidence / caveat output

Because this skill produces a planning artifact before mutation or export, truth states default to Runtime status: Not Run and Artifact status: Not Produced. Truth states are downgrade-only, never upgraded without measured evidence, and never marked Verified by this skill.

Runtime status: Not Run
Artifact status: Not Produced
Evidence used: none (planning ceiling only; compliance measured later by render-export-evidence, polycount-budget-checker, and material-quality-checker)
Limitations: Planning ceiling only, not a runtime benchmark or compliance claim.

Required laws

• ../../laws/evidence-before-done.md
• ../../laws/non-blender-user-language.md
• ../../laws/no-arbitrary-python-interface.md

Official runtime boundary

BlendOps runtime guidance uses 2 MCP execution paths plus a CLI fallback appendix (see ../../docs/runtime-stack-strategy.md):

1. Path 1 — Official Blender Lab MCP (Lab add-on + Lab server in Blender 5.1+, hosted from Anthropic Connector or any other MCP client manually).
2. Path 2 — Community mcp-for-blender (formerly blender-mcp) (Blender 3.0+, third-party prior art).
3. CLI fallback (appendix) — blender --background --python .... Documented upstream as a first-class Blender CLI surface; no in-repo evidence file yet.

This skill is a planning skill. It sets numeric budgets but never installs Blender, runs Blender, or claims that an exported GLB satisfies the budget without measured evidence.

Operating procedure

1. Confirm device tier with the user; default to "mid-mobile" if not stated and record the assumption.
2. Confirm scene scope and target context.
3. Look up the exact 9-metric budget row from references/budget-tiers.md matching the device tier, scene scope, and target context (never invent budget numbers).
4. Pin the numeric budget table into the scene plan before any mutation.
5. Attach the violation rules from references/violation-rules.md stating that any measured violation blocks handoff.
6. Attach the ordered degradation plan from references/degradation-plan.md specifying what to drop first if a ceiling is exceeded.
7. Set Runtime status: Not Run and Artifact status: Not Produced.
8. Name the explicit downstream handoff: scene planners (product-hero-scene-planner, character-portrait, environment-shot, interior-scene, product-grid-scene-planner) to build within the budget, glb-animation-handoff for animation contracts, or polycount-budget-checker and material-quality-checker to validate an existing scene.

Decision tree

Tier unknown?
  → Default to "mid-mobile" + document assumption
Scene scope unknown?
  → Ask once, fall back to "single product" + document assumption
Multiple competing contexts (e.g., AR + hero card)?
  → Pick the strictest ceiling row from references/budget-tiers.md, document the override
Animation present in scope?
  → Pin animation channels, bones, and morph target ceilings from references/budget-tiers.md, then hand off animation contracts to glb-animation-handoff
Existing GLB or scene exceeds budget?
  → Block handoff; apply references/degradation-plan.md and route to polycount-budget-checker and material-quality-checker

Playbooks

Playbook A: Single product hero on mobile

Select the mid-mobile + single product + hero card (or AR if AR viewer is required) row from references/budget-tiers.md. Pin all 9 numeric ceilings (triangles, vertices, GLB file size, texture pixels, materials, draw calls, animation channels, bones, morph targets), attach references/degradation-plan.md, and hand off to product-hero-scene-planner.

Playbook B: Configurator on mid-mobile

Select the mid-mobile + configurator row from references/budget-tiers.md to account for material variants and draw-call ceilings. Pin all 9 numeric ceilings, require material-quality-checker and polycount-budget-checker validation before export, and route final delivery through render-export-evidence and glb-web-handoff.

Playbook C: Low-end Android

Select the low-end Android row from references/budget-tiers.md. Pin strict numeric ceilings across all 9 metrics (including bone and morph target caps from the reference table), attach references/violation-rules.md so any overrun blocks handoff, and hand off animation constraints to glb-animation-handoff.

Mode handling

Text-only mode

Produce the 9-metric budget table (from references/budget-tiers.md), violation rules, degradation plan, and named handoff. Pin Runtime status: Not Run and Artifact status: Not Produced. Do not claim compliance.

Runtime-ready mode

The budget table remains a planning artifact (Runtime status: Not Run, Artifact status: Not Produced at budget-pinning time). Actual compliance is measured later by render-export-evidence alongside polycount-budget-checker and material-quality-checker after export. Truth states are downgrade-only and never upgraded to Verified without measured evidence.

Blocked runtime mode

Produce the same planning budget table, violation rules, degradation plan, and named handoff. Pin Runtime status: Not Run and Artifact status: Not Produced, and record the runtime blocker under Limitations.

Validation checklist

• Device tier chosen explicitly
• Scene scope chosen explicitly
• Target context chosen explicitly
• Budget table pinned with concrete numbers (no "low" / "high" without numbers)
• Violations rule list present
• Degradation plan present
• Assumptions section present
• No "guaranteed performance" claim made
• Handoff to next skill named

Pass / Warn / Fail rubric

Failure handling

• If user pushes for a "yes, it'll run on iPhone X" guarantee → restate that this skill produces a planning budget, not a benchmark.
• If competing constraints exist → pick the strictest and document the override in writing.
• If user has no device data → use mid-mobile as default and write the assumption.

Troubleshooting

Best practices

• Always pin all 9 budget metrics together from references/budget-tiers.md — polycount alone is misleading without texture pixel, draw-call, material, file-size, and animation ceilings.
• Treat every ceiling in the pinned table as a hard handoff blocker enforced downstream by polycount-budget-checker, material-quality-checker, and render-export-evidence.
• Treat AR contexts as a distinct stricter context in references/budget-tiers.md, not an informal tweak.
• Keep role boundaries strict: pin numeric ceilings here, let glb-animation-handoff own animation contracts, and let glb-web-handoff own web delivery packaging.

Good examples

• "Tier: mid-mobile | Scope: single product | Context: hero card. All 9 numeric ceilings pinned from references/budget-tiers.md (triangles, vertices, GLB file size, texture pixels, materials, draw calls, animation channels, bones, morph targets). Violations block handoff per references/violation-rules.md. Degradation order attached from references/degradation-plan.md. Runtime status: Not Run, Artifact status: Not Produced. Next skill handoff: product-hero-scene-planner."

Bad examples

• "Should be fine on mobile." — no numbers, no tier, no rules.
• "GLB will be fast." — no measurement, no caveat.

User-facing response template

Device tier: <low-end Android / mid-mobile / high-mobile / desktop-fallback>
Scope: <single product / multi-product / environment>
Context: <hero card / e-commerce viewer / AR / configurator>
Budget (sourced from references/budget-tiers.md):
  Triangles: <N>
  Vertices: <N>
  GLB file size: <N MB>
  Texture pixel budget total: <N MP>
  Distinct materials: <N>
  Draw calls: <N>
  Animation channels: <N>
  Bones: <N>
  Morph targets: <N>
Violations block handoff: yes (enforced by polycount-budget-checker, material-quality-checker, and render-export-evidence)
Degradation order if violated: <list 1-3 things to drop first from references/degradation-plan.md>
Assumptions: <device tier, network, and FOV assumptions>
Next skill handoff: <product-hero-scene-planner | character-portrait | environment-shot | interior-scene | product-grid-scene-planner | polycount-budget-checker | material-quality-checker | render-export-evidence | glb-web-handoff | glb-animation-handoff>
Runtime status: Not Run
Artifact status: Not Produced
Evidence used: none (planning budget only)
Limitations: Planning ceiling only; not a benchmark or compliance measurement.

Anti-patterns

• Picking polycount in isolation.
• Treating the budget as a benchmark.
• Promising performance without measurement.
• Mixing tiers ("kind of mobile but desktop too").

Cross-skill handoff

• [progression] Scene planning (consumes pinned budget) (camera-composition) → ../camera-composition/SKILL.md
• [progression] Scene planning (consumes pinned budget) (lighting-material-plan) → ../lighting-material-plan/SKILL.md
• [alternative] Scene planning (consumes pinned budget) (character-portrait) → ../character-portrait/SKILL.md
• [alternative] Scene planning (consumes pinned budget) (environment-shot) → ../environment-shot/SKILL.md
• [alternative] Scene planning (consumes pinned budget) (interior-scene) → ../interior-scene/SKILL.md
• [alternative] Scene planning (consumes pinned budget) (product-grid-scene-planner) → ../product-grid-scene-planner/SKILL.md
• [alternative] Scene planning (consumes pinned budget) (product-hero-scene-planner) → ../product-hero-scene-planner/SKILL.md
• [progression] Count and texture budget validation (scene-readiness) → ../scene-readiness/SKILL.md
• [progression] Count and texture budget validation (material-quality-checker) → ../material-quality-checker/SKILL.md
• [progression] Count and texture budget validation (polycount-budget-checker) → ../polycount-budget-checker/SKILL.md
• [progression] Measured export compliance → ../render-export-evidence/SKILL.md
• [progression] Web delivery handoff (consumes pinned budget) → ../glb-web-handoff/SKILL.md
• [progression] Animation contracts (owns animation contract details within pinned channel/bone/morph ceilings) → ../glb-animation-handoff/SKILL.md

Non-goals

• Run a real-world benchmark.
• Install Blender.
• Run Blender.
• Choose a render engine.
• Write web-side optimization code.

References

• references/budget-tiers.md
• references/violation-rules.md
• references/degradation-plan.md
• ../../docs/skill-system.md
• ../../docs/runtime-stack-strategy.md