asset-discovery

Purpose

Decide where the assets in a scene will come from before the scene planner commits to a layout. Asset acquisition decisions affect poly budget, web GLB size, licensing, and time-to-handoff. This skill produces a per-asset plan with explicit confidence labels and licensing notes — without installing tools, running Blender, or fetching anything.

This skill is BlendOps-specific (no direct analog in Anthropic Superpowers or BMad), driven by a real Blender domain need.

Quick start

• read the 8-slot brief from intent-to-3d-brief-writer
• enumerate every asset class needed (subjects, props, environment, materials, HDRIs)
• pick a strategy per asset class (procedural / library / generative / photogrammetry)
• record license + confidence label per choice
• hand off to scene planner

When to use

• after the 3D brief is locked, before any scene plan
• when scene assets aren't trivially given (e.g. user says "shoe" but doesn't supply a shoe model)
• when poly budget / GLB size / brand-licensing constraints exist
• when planner needs to know upfront whether assets are procedural or external

When not to use

• the user supplied all assets in the brief (skip; route directly to planner) → use ../intent-to-3d-brief-writer/SKILL.md (skip; route directly to planner)
• runtime is already executing and assets are already in the scene → use ../product-hero-scene-planner/SKILL.md (route to evidence skill)
• the request is iteration on an existing scene (assets are already chosen) → use ../render-export-evidence/SKILL.md (assets are already chosen)

Trigger phrases

• "what assets do we need"
• "where do the models come from"
• "asset plan before scene planning"
• "do we generate or import"
• "surface asset license risks before planning"

Prerequisites / readiness

• 8-slot brief from intent-to-3d-brief-writer (or equivalent)
• list of constraint slots (performance, brand, time)
• no Blender runtime required
• works in text-only mode

Input schema

Required inputs

• 8-slot brief from intent-to-3d-brief-writer (subject, mood, deliverables, constraints, evidence, acceptance, audience, owner)
• list of asset classes implied by the brief (e.g., hero subject, lighting, floor, environment backdrop)

Optional inputs

• known asset libraries the user has access to (approved brand library, internal library, user-supplied files)
• license preferences (public domain only, commercial allowed, brand-restricted)
• prior similar scenes to match asset choices

Assumptions to confirm

• the operator can defer fetching or generation to a later runtime step
• license restrictions matter for the deliverable (risks surfaced here; verdicts handled by asset-license-checker)
• truth states start at Runtime Not Run and Artifact Not Produced for plan-only work (downgrade-only, never upgrade)

Output schema

Primary output

• per-asset-class plan with strategy + source candidate + license + confidence + risk notes

Secondary output

• budget reconciliation summary compared against mobile-glb-budget and polycount-budget-checker
• explicit list of assets the operator must obtain before any runtime step
• license risk callouts for asset-license-checker to adjudicate

Evidence / caveat output

• Runtime status: Not Run
• Artifact status: Not Produced
• Truth states are downgrade-only, never upgrade
• No asset is downloaded, generated, fetched, or installed by this skill

Required laws

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

Official runtime boundary

This skill does not download, generate, fetch, or install any asset. It does not run Blender, configure any MCP path, or touch the network. It produces a plain-language plan in markdown.

If the chosen strategy is generative, it is Path 2-only (community mcp-for-blender, formerly ahujasid/blender-mcp, Hyper3D / Hunyuan3D integrations) with per-path caveats (see docs/unofficial-runtime-bridges.md). Actual generation happens in a runtime step, never here.

Operating procedure

1. Read the 8-slot brief from intent-to-3d-brief-writer.
2. Enumerate every asset class implied (subjects, props, environment, materials, lighting backdrops, effects).
3. For each asset class, pick one acquisition strategy:
   • procedural — created inside Blender via built-in procedural rules; no external asset file.
   • library — existing approved library or user-supplied asset.
   • generative — Hyper3D / Hunyuan3D (Path 2-only via community mcp-for-blender, formerly ahujasid/blender-mcp, with per-path caveats); actual generation happens in a runtime step, never here.
   • photogrammetry — user-supplied scan; cleanup required before scene assembly.
4. For each pick: record source candidate, license, confidence label (verified-read / linked-only / unknown), and risk notes.
5. Reconcile per-class polycount and file-size estimates against mobile-glb-budget (and polycount-budget-checker). Flag any overage risk explicitly.
6. List every asset that must be obtained before runtime; surface license risks for asset-license-checker (do not issue license verdicts here) and flag multi-source combinations for asset-style-check.
7. Hand off the asset plan + brief to the target scene planner (product-hero-scene-planner, character-portrait, environment-shot, interior-scene, or product-grid-scene-planner).

Decision tree

• subject is brand-specific (e.g., exact branded product) -> strategy must be library (user-supplied / brand library); never generative without explicit brand authorization
• mood is highly specific -> backdrop = library, color and surface treatments = procedural
• tight web or mobile budget constraint -> reconcile against mobile-glb-budget; favor procedural over heavy external imports
• iteration-loop deliverable -> favor procedural (fast and reproducible to adjust)
• one-shot marketing render -> library or generative (Path 2-only with caveats) is acceptable once license risk is cleared by asset-license-checker

Playbooks

• Playbook A: "Cyberpunk shoe web hero" — shoe = generative or user-supplied; floor + props = procedural; HDRI = library; budget = web-tight
• Playbook B: "Brand product render" — subject = brand library only; environment = procedural; license = brand-restricted
• Playbook C: "Iteration sketch" — everything procedural; library only for HDRI

Mode handling

Text-only mode

• plan only, no fetch / generate / install
• Runtime status: Not Run
• Artifact status: Not Produced
• Truth states are downgrade-only, never upgrade

Runtime-ready mode

• plan stays text-only; downstream runtime steps consume the plan
• if generative strategy is chosen, route to runtime-readiness first to confirm Path 2 (mcp-for-blender, formerly ahujasid/blender-mcp, Hyper3D / Hunyuan3D integrations) readiness and caveats

Blocked runtime mode

• plan is still produced with Runtime Not Run and Artifact Not Produced
• mark generative options blocked; favor procedural or library via asset-fallback-strategy

Validation checklist

• every implied asset class enumerated
• canonical strategy chosen per asset class (procedural / library / generative / photogrammetry)
• source candidate named per choice
• license recorded per choice with risks surfaced for asset-license-checker (no final license verdicts claimed here)
• confidence label (verified-read / linked-only / unknown) assigned per choice
• budget reconciled against mobile-glb-budget and brief constraints
• obtain-before-runtime asset list included
• generative strategy flagged as Path 2-only (mcp-for-blender, formerly ahujasid/blender-mcp, Hyper3D / Hunyuan3D) with per-path caveats
• no asset downloaded, fetched, installed, or generated by this skill
• truth states set to Runtime Not Run and Artifact Not Produced (downgrade-only, never upgrade)

Pass / Warn / Fail rubric

Failure handling

• license unknown: mark TBD, surface the risk note, and route to asset-license-checker before the scene planner commits
• generative strategy requested but Path 2 readiness unknown or unavailable: route to runtime-readiness (and asset-fallback-strategy if blocked)
• conflicting constraints (e.g., high-detail scan vs. tight mobile budget): surface conflict against mobile-glb-budget and route back to intent-to-3d-brief-writer if constraints must be relaxed

Troubleshooting

• if the asset list is too long, group by class and propose tiered approach (essentials vs. nice-to-have)
• if the user has no preference, default to procedural-first per BlendOps evidence-before-done (procedural is reproducible)

Best practices

• procedural-first when license / budget pressure is high
• library-second when matching mood matters
• generative as a last resort, always with explicit caveats
• never fetch from this skill

Good examples

• "Cyberpunk shoe brief. Asset plan: hero shoe = generative via Path 2 mcp-for-blender (formerly ahujasid/blender-mcp) Hyper3D/Hunyuan3D integration (license = TBD, risk surfaced for asset-license-checker; confidence = linked-only); floor + props = procedural (license = N/A; confidence = verified-read); environment lighting = library (approved studio backdrop; license = commercial-OK; confidence = verified-read). Budget reconciliation: estimated footprint checked against mobile-glb-budget. Obtain-before-runtime list: hero shoe mesh and environment backdrop. Runtime: Not Run, Artifact: Not Produced."

Bad examples

• "Just use any free shoe model from the internet, we'll figure out license later." (no source named, no license, no confidence label, license risk ignored)

User-facing response template

• Brief recap (one line)
• Asset table (markdown)
• Budget summary
• License risks called out
• Generative caveats if applicable
• Handoff: scene planner with this asset plan + brief

Anti-patterns

• downloading, fetching, generating, or installing any asset from this skill
• issuing final license verdicts instead of surfacing license risk for asset-license-checker
• skipping style consistency handoffs (asset-style-check) when mixing procedural, library, and generative sources
• claiming generative strategy works in Path 1 or without checking Path 2 caveats via runtime-readiness
• referring to the community bridge as bare blender-mcp instead of mcp-for-blender (formerly ahujasid/blender-mcp)
• upgrading truth states above Runtime Not Run or Artifact Not Produced

Cross-skill handoff

• [upstream_reference] Before this skill → ../intent-to-3d-brief-writer/SKILL.md
• [alternative] Scene planners (primary handoff with brief + asset plan) (character-portrait) → ../character-portrait/SKILL.md
• [alternative] Scene planners (primary handoff with brief + asset plan) (environment-shot) → ../environment-shot/SKILL.md
• [alternative] Scene planners (primary handoff with brief + asset plan) (interior-scene) → ../interior-scene/SKILL.md
• [alternative] Scene planners (primary handoff with brief + asset plan) (product-grid-scene-planner) → ../product-grid-scene-planner/SKILL.md
• [alternative] Scene planners (primary handoff with brief + asset plan) (product-hero-scene-planner) → ../product-hero-scene-planner/SKILL.md
• [progression] License verdicts → ../asset-license-checker/SKILL.md
• [progression] Style re-checks → ../asset-style-check/SKILL.md
• [progression] Budget reconciliation (mobile-glb-budget) → ../mobile-glb-budget/SKILL.md
• [progression] Budget reconciliation (polycount-budget-checker) → ../polycount-budget-checker/SKILL.md
• [upstream_reference] Generative readiness → ../runtime-readiness/SKILL.md
• [fallback] Fallback / constraint updates (asset-fallback-strategy) → ../asset-fallback-strategy/SKILL.md

Non-goals

• fetching / downloading / installing assets
• running Blender or generative tools
• making the scene plan
• license adjudication beyond surfacing risk

References

• BlendOps law: ../../laws/official-runtime-only.md
• BlendOps law: ../../laws/evidence-before-done.md
• BlendOps doc on Path 2 generative integrations: ../../docs/unofficial-runtime-bridges.md
• Skill system: ../../docs/skill-system.md