asset-style-check

Purpose

Catch style mismatches between assets before they end up in the same scene where the inconsistency reads as compositing error.

Quick start

• list every asset planned for the scene
• evaluate each on style axes (realism / detail / color / surface)
• flag drift between assets
• recommend remediation (drop / restyle / replace)

When to use

• after asset-license-checker clears the assets
• before scene build
• when a scene preview shows visible style mismatch
• when assets come from multiple sources

When not to use

• single isolated asset where no multi-asset or scene-brief consistency check is requested → use ../asset-discovery/SKILL.md (Delegates out-of-scope requests to asset-discovery)
• pure planning before candidate assets are picked → use ../asset-discovery/SKILL.md (use asset-discovery)
• intentionally mixed-style art direction, ONLY when the locked brief explicitly specifies mixed styles → use ../asset-license-checker/SKILL.md (Delegates out-of-scope requests to asset-license-checker)
• checking asset licenses (asset-license-checker), polygon/performance budgets (polycount-budget-checker), topology, or planning materials/lighting (blender-light → use ../asset-license-checker/SKILL.md (lighting-material-plan)

Trigger phrases

• "do these assets fit together"
• "style check the assets"
• "scene feels like a collage"

Prerequisites / readiness

• candidate asset list with previews sourced via asset-discovery
• all candidate assets cleared by asset-license-checker
• scene brief visual style locked (via intent-to-3d-brief-writer if needed)
• target scene quality bar set

Input schema

Required inputs

Optional inputs

Assumptions to confirm

• Brief style is unambiguous.
• Asset previews are representative of final use.

Output schema

Primary output

A per-asset row table containing: asset name, realism axis, detail axis, color axis, surface axis, drift verdict (None | Minor | Major), validation verdict (Pass | Warn | Fail), and remediation hint (none | drop | restyle | replace).

Secondary output

• top-line scene consistency verdict (Pass | Warn | Fail)
• summary list of assets to drop, restyle, or replace
• downstream handoff routing
• evidence and truth-state block (kept strictly separate from Pass | Warn | Fail validation verdicts)

Evidence / caveat output

Runtime status: Not Run | Attempted | Produced | Verified | Failed
Artifact status: Not Produced | Produced | Verified | Failed (default: Not Produced)
Evidence used: <links, paths, logs, or "none">
Limitations: <known gaps>

Note: Truth states are downgrade-only (never upgrade without verified evidence) and remain strictly separate from per-asset Pass | Warn | Fail validation verdicts.

Required laws

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

Official runtime boundary

Validation-only skill — never modifies or re-styles assets, and never executes Blender or Python. Per-asset verdicts (Pass | Warn | Fail) and drift classifications (None | Minor | Major) are derived solely from documented asset previews and the locked style brief, and are tracked separately from runtime/artifact truth states (which default to Not Produced for artifacts and are downgrade-only).

Operating procedure

1. Confirm the asset list is sourced from asset-discovery and cleared by asset-license-checker.
2. Read the locked brief style baseline (if missing or ambiguous, hand off to intent-to-3d-brief-writer).
3. For each asset, evaluate all 4 style axes (realism, detail, color, surface) against the locked brief (see references/style-axes.md).
4. Assign a drift verdict (None | Minor | Major) and a validation verdict (Pass | Warn | Fail) per asset row.
5. Provide remediation hints only (drop | restyle | replace) for any asset with Minor or Major drift — never modify or re-style the asset here.
6. Route approved rows (Pass) to asset-library-plan for organization and scene-readiness for broad scene review; route rejected rows (Fail) to asset-fallback-strategy or back to asset-discovery.

Decision tree

Brief explicitly locks mixed-style art direction?
  → Evaluate each asset against its assigned role in the brief; allowed contrast = None drift (Pass row)
Asset matches brief baseline across all 4 axes (realism / detail / color / surface)?
  → Drift: None | Verdict: Pass | Remediation: none
Asset off on 1 axis?
  → Drift: Minor | Verdict: Warn | Remediation: restyle or drop (verify intent)
Asset off on 2+ axes, or clashes severely with hero asset?
  → Drift: Major | Verdict: Fail | Remediation: replace or drop (hand off to asset-fallback-strategy)
Two or more assets in the same scene have Major drift between them without explicit mixed-style brief?
  → Top-line Verdict: Fail; cannot combine without breaking visual hierarchy

Playbooks

Playbook A: All-realistic product hero

All assets realistic; same detail density; PBR materials. One stylized prop → drop or restyle.

Playbook B: Stylized character + flat background

Character is stylized; background must be equally simplified. Detailed background asset → drop.

Playbook C: Mixed-style intentional (collage / editorial)

Brief explicitly mixed-style; verify with brief; drift becomes feature, not bug.

Mode handling

Text-only mode

Evaluate rows from documented previews and metadata. Set Runtime status: Not Run and Artifact status: Not Produced.

Runtime-ready mode

Evaluate rows using rendered scene preview snapshots alongside asset previews. Downgrade truth states if any preview is incomplete; never upgrade truth states beyond verified evidence.

Blocked runtime mode

Skip rows that depend on a live rendered preview and mark limitations clearly. Keep Runtime status: Not Run (or Failed if attempted and blocked) and Artifact status: Not Produced.

Validation checklist

• Every asset has a 4-axis row (realism, detail, color, surface) evaluated against the locked brief
• Every asset row includes both a drift verdict (None / Minor / Major) and a validation verdict (Pass / Warn / Fail)
• Top-line consistency verdict (Pass / Warn / Fail) is stated
• Remediation hints (drop / restyle / replace) are provided for any drift without modifying assets
• No license, polycount budget, or topology checks are mixed into the evaluation
• Runtime and artifact truth states (default Not Produced, downgrade-only) are reported separately from Pass / Warn / Fail verdicts

Pass / Warn / Fail rubric

Note: Pass / Warn / Fail validation verdicts measure style consistency only and must never be conflated with Runtime or Artifact truth states.

Failure handling

• Major drift (Fail row) → recommend replace or drop; hand off rejected assets to asset-fallback-strategy or asset-discovery.
• Brief unclear or unlocked regarding visual style → stop evaluation and hand off to intent-to-3d-brief-writer.
• Claim that "mixed style is intentional" → accept ONLY if the locked brief explicitly specifies mixed styles; otherwise mark drift and hand off to intent-to-3d-brief-writer or scope-guard.

Troubleshooting

Best practices

• Lock style early; revisions are cheap pre-build.
• Always evaluate against the brief, not against vibe.
• Prefer 1 source for related assets to reduce drift risk.
• Keep this skill narrow — do not check license / budget / topology here.

Good examples

• "shoe-base.glb: realism=realistic, detail=high, color=neutral, surface=PBR | Drift: None | Verdict: Pass | Remediation: none. shoe-laces.glb: realism=realistic, detail=high, color=neutral, surface=PBR | Drift: None | Verdict: Pass | Remediation: none. shoe-environment-stylized.glb: realism=stylized, detail=low, color=saturated, surface=flat | Drift: Major | Verdict: Fail | Remediation: replace via asset-fallback-strategy."

Bad examples

• "Assets look like they go together." — no axis-by-axis verdict.

User-facing response template

Brief style: <one line locked baseline>

| Asset | Realism | Detail | Color | Surface | Drift (None/Minor/Major) | Verdict (Pass/Warn/Fail) | Remediation (none/drop/restyle/replace) |
|---|---|---|---|---|---|---|---|

Top-line consistency: Pass / Warn / Fail
Assets to drop / restyle / replace: <list or "none">
Runtime status: Not Run | Attempted | Produced | Verified | Failed
Artifact status: Not Produced | Produced | Verified | Failed (default: Not Produced)
Evidence used: <links, paths, logs, or "none">
Limitations: <known gaps>
Next: asset-library-plan & scene-readiness (Pass) / asset-fallback-strategy (Fail)

Anti-patterns

• "Vibe match" verdicts.
• Skipping the brief reference.
• Mixing license / budget / topology checks here.
• Inventing axis values.

Cross-skill handoff

• [upstream_reference] Upstream discovery → ../asset-discovery/SKILL.md
• [upstream_reference] Upstream license gate → ../asset-license-checker/SKILL.md
• [upstream_reference] Upstream brief lock → ../intent-to-3d-brief-writer/SKILL.md
• [boundary_delegation] Scope enforcement → ../scope-guard/SKILL.md
• [fallback] Rejected asset fallback → ../asset-fallback-strategy/SKILL.md
• [progression] Approved asset organization → ../asset-library-plan/SKILL.md
• [progression] Broad scene review → ../scene-readiness/SKILL.md
• [progression] Material & lookdev boundary (lighting-material-plan) → ../lighting-material-plan/SKILL.md
• [progression] Material & lookdev boundary (material-quality-checker) → ../material-quality-checker/SKILL.md
• [progression] Budget & topology boundary → ../polycount-budget-checker/SKILL.md

Non-goals

• Never modify or re-style assets.
• Never check asset licenses (belongs to asset-license-checker).
• Never check polycount budgets or mesh topology (belongs to polycount-budget-checker).
• Never execute Blender or Python code.

References

• references/style-axes.md
• references/drift-classification.md
• references/intentional-mixed-style.md
• ../../docs/skill-system.md