text-to-blender

这段内容是审核时保存的快照。外部来源才是完整且最新的版本。

---
name: text-to-blender
description: Drive Blender from natural language. Converts plain-English requests ("model a sword and render it with cinematic lighting", "make this glass look frosted", "set up three-point lighting", "export this scene as glTF for the web") into Blender Python code executed via the Blender MCP server. Acts as the orchestrator that picks and chain-loads specialised sub-skills (blender-modeling, blender-materials, blender-lighting, blender-cameras, blender-rendering, blender-animation, blender-export, wireframe-to-3d, blender-pro-workflow, blender-skill-harmonizer, quality-refinement-autoloop). Use this skill whenever the user wants Claude to do anything in Blender, including creating geometry, applying materials, lighting a scene, framing a camera, rendering, animating, or exporting. Make sure to invoke this skill even if the user does not say "Blender" — also covers requests like "create a 3D model of...", "render this...", "make a glTF from...", "set up a scene with...", or any 3D-creation task. Requires the Blender MCP addon (ahujasid/blender-mcp) running on port 9876.
when_to_use: User asks for any 3D creation, modification, lighting, rendering, animation, or export task. Anything involving Blender or that should reasonably be done in Blender.
allowed-tools: Read Bash Glob Grep mcp__blender__execute_blender_code mcp__blender__get_scene_info mcp__blender__get_object_info mcp__blender__get_viewport_screenshot
---

# Text-to-Blender Orchestrator

Turn plain-English requests into Blender work. You are the conductor: read the request, decide which sub-skills to chain-load, sequence them in the right order, and execute via the Blender MCP.

## Multi-skill harmonization

For complex tasks that trigger multiple Blender skills, especially reference/template/brand work, load `blender-skill-harmonizer` before choosing the execution order. It owns precedence, handoff artifacts, and conflict policy.

If the user rejects an output as sub-par, says the same issue is recurring, or asks to improve the skill stack before retrying, load `quality-refinement-autoloop` before further product work. It turns evidence into a sanitized reusable lesson, patches generic skills/docs/versioning when explicitly requested, and only then resumes the Blender repair loop.

## How this skill works

The user speaks in tasks ("render a hero shot of a sword on a stone"); you:

1. **Identify intent** → which capabilities are needed (modeling? materials? lighting? rendering? export?).
2. **Check prerequisites** → MCP server reachable, scene state.
3. **Reset world** to a known-good baseline (see *World reset* below) — prevents leftover broken HDRI / Environment Texture nodes from previous runs corrupting the render.
4. **Look up real-world dimensions** for the subject (`references/common-object-dimensions.md`) BEFORE generating Python. A "sword" has specific proportions; a "chair" has specific proportions; do not guess.
5. **Route to sub-skills** → load the relevant ones via `Read` and follow their instructions.
6. **Sequence** the work in the order pros use (see `references/assembly-order.md`).
7. **Execute** generated Python via `mcp__blender__execute_blender_code`.
8. **Validate** with `mcp__blender__get_scene_info`, `get_object_info`, AND **`get_viewport_screenshot`** — see *Visual validation checkpoint* below. Numerical validation alone is not enough: the API can report success while geometry is grossly wrong.
9. **Iterate** — if the visual check shows an obvious problem (subject barely visible, wrong proportions, wrong orientation), fix and re-render BEFORE reporting success.
10. **Report** to the user with concrete numbers AND a path to the proof render they can inspect.

### World reset (always step 3, before any composition)

Previous tests can leave the scene's world in a broken state — particularly an `Environment Texture` node with `image=None` cascading into Background, which produces a magenta-flooded render. Reset world to