Blender Recipe Decomposer

blender-recipe-decomposer

Purpose

When user intent is too big for one recipe (multiple deliverables, mixed modes, conflicting briefs), split it into smaller recipes that each fit the BlendOps recipe shape (brief → plan → assets → runtime → handoff).

Quick start

• enumerate the deliverables in the intent
• group them into 2-5 cohesive recipe-sized chunks
• name each new recipe + assign a brief + stop condition
• pin recipe order (linear or parallel-OK)

When to use

• intent has multiple distinct deliverables (e.g., hero render + animation + GLB + variant grid)
• intent crosses multiple scene types (portrait + environment in one ask)
• previous attempts collapsed under their own scope
• before any planner runs

When not to use

• single deliverable; recipe is already right-sized
• pure exploration / brainstorming
• after recipes have started running (use scope-boundary-enforcer or stop-condition-decider instead)

Trigger phrases

• "I want all of these"
• "this is a lot"
• "we'll need several outputs"
• "across these scenes"

Prerequisites / readiness

• intent stated by user (or extractable)
• user accepts that splitting will yield more reliable output

Input schema

Required inputs

| Input | Why it is required | |---|---| | Full user intent | Source for decomposition | | Acceptable recipe count target (2-5 typical) | Constrains over-splitting |

Optional inputs

| Input | Use | |---|---| | Time / budget per recipe | Affects sizing | | Deliverable priority | Drives recipe order |

Assumptions to confirm

• Splitting is acceptable (not all users want this).
• Each new recipe will need its own brief.
• Some shared assets across recipes is OK; design re-use early.

Output schema

Primary output

A recipe table: name, brief seed (one line), stop condition (one line), order, dependencies, handoff target.

Secondary output

• shared assets list (used by 2+ recipes)
• recommended start recipe
• explicit out-of-scope reminder for each recipe

Evidence / caveat output

Runtime status: Not Run | Attempted | Produced | Verified | Failed | Blocked / Not Run
Artifact status: Not Run | Not Produced | Produced | Verified | Failed
Evidence used: 
Limitations: 

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

Process / discipline skill — does not run Blender. Splits intent into recipe specs only.

Operating procedure

1. Read full intent.
2. Enumerate distinct deliverables.
3. Group deliverables into 2-5 cohesive recipes (see references/recipe-grouping-rules.md).
4. For each recipe, write: name, brief seed, stop condition, order, dependencies.
5. Identify shared assets across recipes.
6. Recommend start recipe.
7. Hand each recipe off to intent-to-3d-brief-writer for full brief.

Decision tree

Single deliverable?
  → No decomposition needed; pass to brief writer directly
2-5 distinct deliverables?
  → Group + decompose
6+ deliverables?
  → Strongly recommend product backlog approach; do not handle as single intent
Conflicting briefs (e.g., realistic + stylized in one ask)?
  → Split by style; do not merge
Shared assets across all deliverables?
  → Plan asset recipe first; downstream recipes consume it

Playbooks

Playbook A: Hero + variants + animation (3 recipes)

1. Hero static render (base recipe).
2. Color variants grid (consumes hero scene + asset).
3. Animation loop on hero (consumes hero scene; new export).

Playbook B: Portrait + environment + product (3 recipes, parallel-OK)

1. Character portrait scene.
2. Environment establishing shot.
3. Product hero render.

Each independent; parallel execution acceptable.

Playbook C: Multi-pack catalog (5 recipes)

1. Asset library prep (shared).
2. Product 1 hero.
3. Product 2 hero.
4. Product 3 hero.
5. Catalog grid composite.

Recipes 2-4 can run parallel; recipe 5 depends on 2-4.

Mode handling

Text-only mode

Decomposition into recipe specs; no runtime claims.

Runtime-ready mode

Recipe specs include runtime path attribution per recipe (each recipe may use a different path).

Blocked runtime mode

Decomposition still useful; runtime per-recipe marked Blocked until paths available.

Validation checklist

• [ ] Distinct deliverables enumerated
• [ ] Recipe count 2-5
• [ ] Each recipe has name + brief seed + stop condition
• [ ] Order + dependencies explicit
• [ ] Shared assets identified
• [ ] Start recipe recommended
• [ ] Out-of-scope per recipe stated

Pass / Warn / Fail rubric

| Verdict | Criteria | |---|---| | Pass | 2-5 cohesive recipes + name + brief + stop + order + handoff. | | Warn | Recipes present but stop conditions soft, or order unclear. | | Fail | One mega-recipe still; > 5 over-split fragments; missing stop conditions; no order. |

Failure handling

• Single mega-recipe → forced to split per recipe-grouping-rules.md.
• > 5 fragments → consolidate adjacent ones.
• User refuses split → document as risk; warn that gates and stop conditions will be hard to maintain.

Troubleshooting

| Problem | Response | |---|---| | Decomposed recipes have heavy dependencies | Re-group to reduce coupling; consider shared-asset pre-recipe. | | Recipes overlap in scope | Re-draw boundaries; eliminate overlap. | | Recipe order ambiguous | Pick one (typically asset-first → main render → variants → composite); justify. |

Best practices

• Keep recipes 2-5; outside that range usually wrong-sized.
• Plan shared assets early; reduces rework.
• Name recipes after the deliverable, not the process.
• Hand off each recipe to intent-to-3d-brief-writer for full brief.

Good examples

• "Recipes (3): (1) shoe-hero-render — out/hero.png, (2) shoe-color-variants-grid — out/grid.png, (3) shoe-orbit-animation — out/orbit.glb. Order: 1 → 2 (variant uses hero scene) → 3 (animation uses hero scene). Shared assets: shoe-base-mesh, hero-lighting-rig. Start: recipe 1."

Bad examples

• "We'll do it all in one go." — no decomposition.

User-facing response template

User intent (full): 
Deliverables enumerated: 

Recipes:
| # | Name | Brief seed | Stop condition | Order | Dependencies |
|---|---|---|---|---|---|

Shared assets: 
Recommended start: 
Each recipe hands off to: intent-to-3d-brief-writer
Limitations: 

Anti-patterns

• Single mega-recipe.
• > 5 fragmented recipes.
• Recipes without stop conditions.
• Skipping the dependency analysis.

Cross-skill handoff

• After decomposition, each recipe → ../intent-to-3d-brief-writer/SKILL.md
• Workflow gates per recipe → ../blender-checklist-driven-workflow/SKILL.md
• Stop condition per recipe → ../blender-stop-condition-decider/SKILL.md

Non-goals

• Run Blender.
• Author full briefs (that is for intent-to-3d-brief-writer).
• Render or export anything.

References

• references/recipe-grouping-rules.md
• references/dependency-patterns.md
• references/recipe-naming-rules.md
• ../../docs/skill-system.md

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: ThanhNguyxnOrg
• Source: ThanhNguyxnOrg/blendops
• License: MIT
• Homepage: https://github.com/ThanhNguyxnOrg/blendops

Install and usage instructions live in the source repository linked above.