Skip to content
Back to search

cc-blender-skill

SkillWorks with Claude CodeGitHub

Claude Code skill plugin: drive Blender 5.x like a senior 3D artist via natural language. 10 chain-loadable skills (modeling, materials, lighting, cameras, rendering, animation, export, wireframe-to-3d, pro-workflow). Validated on 6 scene classes.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/roble3-cc-blender-skill/badge.svg)](/listings/roble3-cc-blender-skill)

HTML

<a href="/listings/roble3-cc-blender-skill"><img src="/api/listings/roble3-cc-blender-skill/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Prerequisites: Blender ≥ 4.0 with [BlenderMCP addon](https://github.com/ahujasid/blender-mcp) running on port 9876, Claude Code, Python 3.9+ with `opencv-python numpy scipy Pillow`.

git clone git@github.com:RobLe3/cc-blender-skill.git
cd cc-blender-skill

# Symlink all 30 skills into ~/.claude/skills/
for skill in plugin/skills/*/; do
    name=$(basename "$skill")
    ln -sfn "$(pwd)/$skill" "$HOME/.claude/skills/$name"
done

# Restart Claude Code to pick up the new top-level skills directory entries.

Then ask Claude something like:

Make a 3D model of a teapot and render it with three-point lighting.

Or invoke a specific skill:

/wireframe-to-3d ./glasses_front.png

Full install + verification: [`plugin/README.md`](./plugin/README.md).

---

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 345 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. # cc-blender-skill
  2. A portable agent skill pack that lets Codex, Claude Code, and compatible skill hosts use Blender like a senior 3D artist through natural language.
  3. **Version**: **1.4.0** ([CHANGELOG](./CHANGELOG.md)) · adds official-first dual MCP support and checked Unreal collision / Multires cage-export workflows. The existing 30-skill stack, legacy asset integrations, and source-locked validation gates remain. See the [integration test report](./docs/test-results/v1.4.0-integration.md) for verified behavior and known validation gaps.
  4. **Quick links**: [Install](#quick-install) · [What works (honestly)](#what-works-honestly) · [What doesn't](#what-doesnt-work-yet-honestly) · [Architecture](#architecture-in-one-paragraph) · [Contributing](.github/CONTRIBUTING.md) · [Releases](https://github.com/RobLe3/cc-blender-skill/releases)
  5. ---
  6. ## What this is
  7. Thirty chain-loadable skills turn requests like *"model a sword and render a hero shot with three-point lighting"* into Blender Python executed via either [official Blender Lab MCP](https://www.blender.org/lab/mcp-server/) or [ahujasid Blender MCP](https://github.com/ahujasid/blender-mcp). The shared skill content is agent-agnostic; installation and MCP configuration remain host-specific.
  8. ```
  9. User prompt
  10. ↓
  11. Agent detects intent → loads text-to-blender (orchestrator)
  12. ↓
  13. Orchestrator chain-loads relevant sub-skills
  14. ↓ ↓ ↓
  15. modeling, materials, lighting, cameras, rendering, animation, export, wireframe-to-3d, pro-workflow, reference-to-3d, UV/atlas fitting, validation, fit repair, look calibration
  16. ↓
  17. Generated Python → mcp__blender__execute_blender_code → Blender → output
  18. ```
  19. The plugin is the actual installable thing. It lives at [`plugin/`](./plugin/). Knowledge research that produced it lives at [`knowledge/`](./knowledge/) and [`docs/`](./docs/).
  20. ---
  21. ## What's new in v1.4.0
  22. - **Both MCP backends are supported.** Official Blender Lab MCP is preferred for shared capabilities; existing ahujasid installations still work. Tasks requiring its asset-download/generation tools use ahujasid from the start. The chosen backend stays fixed throughout the task.
  23. - **Official installation and validation are explicit.** Official MCP requires Blender 5.1+, its add-on and a separate MCP process. Render validation reads the returned file path, not an assumed destination. Existing legacy Blender 4.x caveats remain.
  24. - **Unreal collision export has a preflight.** Collider names, evaluated geometry, closedness, winding, convexity, primitive restrictions, and same-file selection are checked before the new static FBX route.
  25. - **Multires baking targets the exported cage.** Level 0 is set before baking and kept through export; temporary scene/context settings are restored afterward. Existing general export defaults are unchanged.
  26. These are additive integrations, not a claim that every Blender version or Unreal import has been validated. [Testing status and limitations](./docs/test-results/v1.4.0-integration.md).
  27. ### Thanks, and sorry for the slow response
  28. Thanks to **[fatalerrorist](https://github.com/fatalerrorist)** for
  29. [PR #1: official Blender MCP support](https://github.com/RobLe3/cc-blender-skill/pull/1),
  30. and **[NicoCaltha](https://github.com/NicoCaltha)** for
  31. [PR #2: Unreal collision meshes and Multires baking](https://github.com/RobLe3/cc-blender-skill/pull/2).
  32. Both contributions surfaced useful gaps and supplied concrete fixes.
  33. Sorry it took me a while to respond. I've been head-down in a gazillion other
  34. projects. I hope these updates make the skill more useful for everyone using it.
  35. ### v1.4.0 testing status and feedback wanted
  36. The release candidate has been exercised on Blender 5.1.1 through both MCP
  37. backends. All 30 skill folders pass Codex metadata validation; 6 host checks,
  38. 12 Blender export checks, and all 35 core recipe checks pass. Native Codex runs
  39. also covered skill discovery, materials, rendering, animation, GLB export, and
  40. checked collision FBX export. See the
  41. [full evidence and exact limits](./docs/test-results/v1.4.0-integration.md).
  42. The following still needs community testing:
  43. - Native **Claude Code** discovery, automatic routing, and explicit skill invocation
  44. - The complete 225-query trigger set and ten backend-routing scenarios in both clients
  45. - Unreal Engine import behavior for collision FBX and baked cage normals
  46. - Blender 4.x and versions newer than the tested Blender 5.1.1 environment
  47. - Legacy provider downloads/generation and a wider range of real production scenes
  48. Claude Code users are especially invited to try v1.4.0 and report what actually
  49. happens. Open a [bug report](https://github.com/RobLe3/cc-blender-skill/issues/new?template=bug_report.yml)
  50. or submit a PR. Include the agent/client version, Blender version, selected MCP
  51. backend and versions, installation scope, exact prompt, whether invocation was
  52. automatic or explicit, expected versus actual behavior, and sanitized logs or
  53. renders. Please remove credentials, private paths, and proprietary scene data.
  54. ---
  55. ## What's new in v1.3.0
  56. This release adds a generic self-refinement layer for cases where output quality is below expectation. Instead of blindly retrying, the stack now freezes the artifact, diagnoses the failure dimension, decides whether existing skills are sufficient, sanitizes the lesson into publishable generic guidance, validates the skill graph, and only then repairs the product.
  57. ### Added
  58. - `quality-refinement-autoloop` — a RALPH-style loop for subpar outputs, repeated failures, skill-gap diagnosis, sanitization, docs/version prep, and release handoff.
  59. - `closed-surface-uv-coverage` — a hard gate for closed/extruded assets so front caps, back caps, and sidewalls each have real surface coverage rather than overlay-only detail.
  60. - Animation/look motion skills: `texture-state-animation`, `orbital-hud-motion`, and `animation-quality-gate` for registered texture states, source-derived HUD motion, and contact-sheet QA.
  61. - Additional source-driven repair skills: `source-part-segmentation`, `texture-driven-mesh-fitting`, `landmark-fit-repair`, and `multiview-constraint-solver`.
  62. ### Changed
  63. - `blender-skill-harmonizer` now routes rejected/subpar outputs through the quality-refinement autoloop before further artifact work.
  64. - UV/texturing guidance now includes closed-surface front/back/side coverage validation.
  65. - README/plugin docs now describe the expanded 30-skill stack and the publishable self-refinement workflow.
  66. ---
  67. ## What's new in v1.2.9
  68. This release adds a generic, source-driven reconstruction layer on top of the existing natural-language Blender workflow. The important changes are:
  69. ### Added
  70. - `blender-skill-harmonizer` — a meta-orchestrator for multi-skill precedence, source-of-truth policy, handoff artifacts, and source-conflict gates.
  71. - `reference-to-3d` — a source-locked reconstruction workflow for templates, reference sheets, texture packs, and orthographic views.
  72. - `reference-analysis-validator` — source manifests, part-count gates, masks, overlays, IoU/SSIM/bbox/centroid validation, and fail-before-export checks.
  73. - `contour-to-mesh` and `orthographic-registration` — silhouette-first mesh construction plus front/side/back/top coordinate contracts.
  74. - `blender-uv-texturing` and `atlas-uv-fitting` — UV projection, atlas-region mapping, decals, lightmaps, and supplemental-map sanity checks.
  75. - `multiview-fit-loop` and `fit-repair-optimizer` — render/compare/adjust loops and dependency-aware repair queues.
  76. - `reference-look-calibration` — measurable source-image look matching for crop, brightness, saturation, hue, accent/glow, materials, lights, and render settings.
  77. - `mascot-logo-reconstruction` — generic brand mascot/logo orchestration driven by `reference_manifest.json`, not hardcoded project assumptions.
  78. ### Changed
  79. - Existing production skills now hand off to the reference-locked stack when the user provides source templates, texture packs, or repeated visual mismatch feedback.
  80. - The wireframe workflow now has a correction mode for source/texture-driven subjects instead of continuing primitive-first rebuilds.
  81. - Lighting, material, and rendering skills now defer source-image look matching to `reference-look-calibration` when exact visual match matters.
  82. - Examples and scripts were sanitized so structural counts, accent hues, guide masks, and validation gates are manifest/report-driven and reusable across projects.
  83. ---
  84. ## What works (honestly)
  85. | Capability | Status | Validation |
  86. |------------|--------|------------|
  87. | End-to-end scene build from natural language | ✅ Works first-try on most common subjects | 6 validated scene classes: sword, bottle, chair, aviator, desk lamp, broadcaster avatar |
  88. | Real-world dimension lookup | ✅ Reference covers 7 categories | swords, chairs, bottles, mugs, tables, lamps, eyewear, characters — `references/common-object-dimensions.md` |
  89. | Multi-skill chaining | ✅ Validated | wireframe-to-3d → modeling → materials → lighting → camera → render → export |
  90. | Blender 5.x compat | ⚠️ Bounded live checks on 5.1.1 | Both MCP routes pass all 35 core checks; broader 5.x coverage remains open |
  91. | Blender 4.x compat | ⚠️ Compat code present, not directly tested | See `plugin/skills/text-to-blender/references/blender-version-compat.md` |
  92. | Subject-class lighting (metal / glass / wood / fabric / skin / product) | ✅ Subject-aware profiles | Recipe 0a in `blender-lighting/SKILL.md` |
  93. | Emission material + practical lighting | ✅ Lamp scene validated | Recipe 11b in `blender-materials`, Recipe 0c in `blender-lighting` |
  94. | Volume-absorption coloured glass | ✅ 5 colour types tuned | Recipe 6b in `blender-materials/SKILL.md` |
  95. | Trigger-eval description tuning | ✅ 200 starter queries shipped | Each skill has `evals/evals.json`; aggregate 100% TP / 4% FP |
  96. | Wireframe-to-3d auto-extraction | ⚠️ Works for simple line-art; now has reference-locked correction handoff | Aviator validated; complex designed objects should use the reference-to-3d / contour / registration / validation stack |
  97. | Reference-locked reconstruction | ✅ Workflow added | Source manifest, contour-to-mesh, orthographic registration, multiview fit loop, repair queue, and overlay validation |
  98. | Texture-pack / atlas fitting | ✅ Workflow added | UV projection, per-region atlas mapping, alpha decals, supplemental-map sanity checks, lightmap handling |
  99. | Reference look calibration | ✅ Workflow added | Camera crop, brightness/saturation/hue, accent/glow masks, material/light/render handoff |
  100. | Skill harmonization | ✅ Workflow added | Meta-orchestrator defines precedence, handoff artifacts, source-conflict gates, and sequential/parallel repair lanes |
  101. ## What doesn't work yet (honestly)
  102. - **Design quality ≠ build correctness.** Plugin produces functionally correct objects. Aesthetic refinement (curved chair backs, profile-cut legs, consciously-composed silhouettes) is human-driven — out of scope for automatic generation.
  103. - **Human faces from primitives.** A sphere + nose + ears + mouth + brows reads as "abstract avatar," not "human." Real human faces require subtractive sculpting. v1.2.1 documents three escape paths: import via `download_polyhaven_asset` / `download_sketchfab_model` / `generate_hyper3d_model_via_text` (then chain), sculpt mode, or commission an artist (`docs/avatar-design-kit/prompts/04-blender-workflow.md`).
  104. - **Thin-metal specular flare.** Hero shots of thin metal (eyewear arms, jewellery) catch side lighting as bright streaks. Workaround: top-down softbox lighting or crop temple arms out of the frame.
  105. - **Subjective quality.** Numerical validation passing ≠ render looks right. The orchestrator's mandatory visual-validation checkpoint exists, but the user remains the final oracle.
  106. - **External user feedback.** Core validation is internal; the newer reference-locked stack still needs broader external examples, and real-use feedback drives v1.x patches.
  107. ---
  108. ## Quick install
  109. Prerequisites: a compatible skill host, a configured Blender MCP backend, and Blender. Official MCP requires Blender ≥5.1; the existing ahujasid route retains Blender ≥4.0 (4.x compatibility code is not live-tested here). Image-processing helpers need Python 3.9+ with `opencv-python numpy scipy Pillow`. See the [agent-host setup](./plugin/skills/text-to-blender/references/agent-hosts.md) and [backend setup](./plugin/skills/text-to-blender/references/mcp-backends.md); enabling an add-on alone is not enough.
  110. ```bash
  111. git clone git@github.com:RobLe3/cc-blender-skill.git
  112. cd cc-blender-skill
  113. # Codex personal install. Use .agents/skills in a project for project scope.
  114. mkdir -p "$HOME/.agents/skills"
  115. for skill in plugin/skills/*/; do
  116. name=$(basename "$skill")
  117. test ! -e "$HOME/.agents/skills/$name" && test ! -L "$HOME/.agents/skills/$name" || {
  118. echo "Refusing to replace $name" >&2; exit 1;
  119. }
  120. done
  121. for skill in plugin/skills/*/; do
  122. name=$(basename "$skill")
  123. ln -s "$(pwd)/$skill" "$HOME/.agents/skills/$name"
  124. done
  125. ```
  126. Claude Code installation and project-scoped examples are in the [agent-host guide](./plugin/skills/text-to-blender/references/agent-hosts.md).
  127. Then ask the agent something like:
  128. ```
  129. Make a 3D model of a teapot and render it with three-point lighting.
  130. ```
  131. Or invoke a specific skill:
  132. ```
  133. $wireframe-to-3d ./glasses_front.png
  134. ```
  135. Codex supports `$skill-name` and `/skills`; Claude Code uses its own explicit skill syntax.
  136. Full install + verification: [`plugin/README.md`](./plugin/README.md).
  137. ---
  138. ## Repository structure
  139. ```
  140. cc-blender-skill/
  141. ├── README.md # this file
  142. ├── CHANGELOG.md # release notes per version
  143. ├── VERSIONING.md # full per-version rationale
  144. ├── LICENSE # MIT
  145. ├── requirements.txt
  146. ├── .github/ # CONTRIBUTING + issue templates
  147. │
  148. ├── plugin/ # ← the installable skill plugin
  149. │ ├── README.md
  150. │ ├── manifest.json
  151. │ └── skills/
  152. │ ├── text-to-blender/ # orchestrator
  153. │ ├── blender-skill-harmonizer/ # multi-skill precedence + handoff contracts
  154. │ ├── quality-refinement-autoloop/ # RALPH-style quality loop + release prep
  155. │ ├── blender-pro-workflow/ # multi-phase guidance
  156. │ ├── blender-modeling/ # geometry creation
  157. │ ├── blender-materials/ # PBR via Principled BSDF
  158. │ ├── blender-lighting/ # 3-point, HDRI, studio
  159. │ ├── blender-cameras/ # framing, DoF, animated cameras
  160. │ ├── blender-rendering/ # Cycles/EEVEE
  161. │ ├── blender-animation/ # keyframes, F-curves, shape keys
  162. │ ├── blender-export/ # glTF/FBX/OBJ/USD/STL
  163. │ ├── wireframe-to-3d/ # specialty: 2D wireframe → 3D
  164. │ ├── reference-to-3d/ # source-locked reconstruction
  165. │ ├── reference-analysis-validator/ # masks, metrics, overlays
  166. │ ├── contour-to-mesh/ # contour-derived mesh generation
  167. │ ├── orthographic-registration/ # front/side/back/top coordinate contract
  168. │ ├── blender-uv-texturing/ # UV, projection, baking, lightmaps
  169. │ ├── atlas-uv-fitting/ # per-part atlas/decal mapping
  170. │ ├── closed-surface-uv-coverage/ # front/back/side surface coverage audits
  171. │ ├── mascot-logo-reconstruction/ # generic brand mascot/logo workflow
  172. │ ├── multiview-fit-loop/ # render/compare/adjust validation loop
  173. │ ├── fit-repair-optimizer/ # dependency-aware repair queues
  174. │ ├── reference-look-calibration/ # source-image look matching
  175. │ ├── source-part-segmentation/ # per-part source masks
  176. │ ├── texture-driven-mesh-fitting/ # mesh boundaries fit texture/source contours
  177. │ ├── landmark-fit-repair/ # named feature repair gates
  178. │ ├── texture-state-animation/ # layered registered texture motion
  179. │ ├── orbital-hud-motion/ # source-derived HUD/circle motion
  180. │ └── animation-quality-gate/ # contact-sheet animation QA
  181. │
  182. ├── knowledge/ # raw research aggregation (16 domains)
  183. │ ├── README.md
  184. │ ├── 01-modeling/00-overview.md
  185. │ ├── 02-curves-surfaces/00-overview.md
  186. │ ├── 03-sculpting-retopo/00-overview.md
  187. │ ├── 04-geometry-nodes/00-overview.md
  188. │ ├── 05-materials-shading/00-overview.md
  189. │ ├── 06-uv-texturing/00-overview.md
  190. │ ├── 07-lighting/00-overview.md
  191. │ ├── 08-cameras-composition/00-overview.md
  192. │ ├── 09-animation/00-overview.md
  193. │ ├── 10-rigging/00-overview.md
  194. │ ├── 11-rendering/00-overview.md
  195. │ ├── 12-compositing/00-overview.md
  196. │ ├── 13-physics-particles/00-overview.md
  197. │ ├── 14-import-export/00-overview.md
  198. │ ├── 15-cross-cutting/00-overview.md
  199. │ └── 16-pro-workflows/00-overview.md
  200. │
  201. ├── docs/ # documentation
  202. │ ├── SKILL_FOUNDATION.md # original research
  203. │ ├── BLENDER_BEST_PRACTICES.md
  204. │ ├── BLENDER_INTEGRATION_GUIDE.md
  205. │ ├── BLENDER_MCP_ALIGNMENT.md
  206. │ ├── WIREFRAME_SKILL.md
  207. │ ├── SKILL_RESEARCH_SUMMARY.md
  208. │ ├── process/ # dev journals (planning, tests, mcp coverage)
  209. │ │ ├── PLAN.md
  210. │ │ ├── DEVELOPMENT.md
  211. │ │ ├── TESTING_PLAN.md
  212. │ │ ├── IMPLEMENTATION_LOG.md
  213. │ │ ├── VERIFICATION_REPORT.md
  214. │ │ ├── MCP_COVERAGE_ASSESSMENT.md
  215. │ │ ├── BLENDER_TOOLKIT_COMPARISON.md
  216. │ │ └── INSTALL_BLENDER_MCP.md
  217. │ └── test-results/ # per-round eval logs
  218. │ ├── test.md # round 1: 30-test smoke
  219. │ ├── test_round2.md # round 2: scene-build feedback loop
  220. │ └── test_round3.md # round 3: trigger-eval self-assessment
  221. │
  222. ├── src/ # original wireframe analyzer (still used by skill)
  223. │ └── wireframe_analyzer.py
  224. │
  225. └── skill/ # initial single-skill prototype (now superseded by plugin/)
  226. └── wireframe-to-3d/
  227. ```
  228. ---
  229. ## Architecture in one paragraph
  230. Pure-skill design — no Python wrapper class, no custom MCP. The host agent reads the user's intent, loads the relevant skill files with its file-reading capability, generates Blender Python, calls the selected backend's `execute_blender_code`, checks response status and stdout, and reports results. Both backends use a separate MCP process connected to an add-on socket; their protocols and result shapes differ. State lives in `bpy.data` (Blender's global state, persists between calls); Python variables don't (each `execute_blender_code` is a fresh namespace, so we identify objects by stable name like `bpy.data.objects['GEO-sword']`). Naming follows Blender Studio conventions (`GEO-`, `MAT-`, `LGT-` prefixes). Host mappings are documented in [`agent-hosts.md`](./plugin/skills/text-to-blender/references/agent-hosts.md).
  231. ---
  232. ## Why this exists
  233. There are already two Claude+Blender skills:
  234. - **[ra100/blender-claude-plugin](https://github.com/ra100/blender-claude-plugin)** — 8 generalist Blender API reference skills (geometry nodes, shader nodes, compositor, etc.). Teaches Claude *how* Blender works.
  235. - **[Dev-GOM/blender-toolkit](https://mcpmarket.com/tools/skills/blender-toolkit)** — Mixamo retargeting via custom WebSocket addon.
  236. Neither tackles the **task-level orchestration**: "given a natural-language request, produce a finished 3D output." That's what this plugin does. It supports both official Blender Lab MCP and `ahujasid/blender-mcp`, and adds:
  237. - Pro-workflow sequencing (block-out → camera → lighting → forms → materials → detail → render → composite → export)
  238. - Recipe libraries per domain with copy-paste-ready Python
  239. - Decision trees mapped to natural-language intent
  240. - Naming and validation conventions throughout
  241. See [`docs/process/BLENDER_TOOLKIT_COMPARISON.md`](./docs/process/BLENDER_TOOLKIT_COMPARISON.md) for the full landscape comparison.
  242. ---
  243. ## Honest status
  244. The plugin shipped through **15+ versions of validation and patches** since the v0.3.0 scaffolding, then expanded with a generic reference-locked reconstruction stack for source/template/texture-driven work. Each version's commit summary in [`CHANGELOG.md`](./CHANGELOG.md) records concrete bugs found and fixed; each user-driven feedback iteration is in [`VERSIONING.md`](./VERSIONING.md) with the patch it produced. The proof renders in [`plugin/skills/text-to-blender/assets/v0.X.0-validation/`](./plugin/skills/text-to-blender/assets/) are honest evidence — no cherry-picking, including failure-state renders.
  245. The validation pattern is documented in [`docs/process/TESTING_PLAN.md`](./docs/process/TESTING_PLAN.md): cheap-Haiku tester runs deterministic tests + writes structured results, expensive-Opus patcher reads them and applies fixes. ~10× cheaper than running the full loop on a frontier model throughout. Three round logs live in [`docs/test-results/`](./docs/test-results/).
  246. ### What v1.2.9 adds
  247. - Generic `blender-skill-harmonizer` meta-layer for multi-skill precedence, source-of-truth policy, and handoff artifacts.
  248. - Source-locked reconstruction stack: `reference-to-3d`, `reference-analysis-validator`, `orthographic-registration`, `contour-to-mesh`, `multiview-fit-loop`, and `fit-repair-optimizer`.
  249. - Texture workflow expansion: `blender-uv-texturing` plus `atlas-uv-fitting` for texture packs, decals, UV regions, and supplemental map sanity checks.
  250. - `reference-look-calibration` for measurable source-image look matching after geometry/UV gates pass.
  251. - Generic mascot/logo workflow that derives structural counts and visual constraints from `reference_manifest.json` instead of hardcoded project assumptions.
  252. What v1.2.9 means here: **the stable core remains intact**, scope boundaries (design quality, human faces, thin-metal flare) are documented with escape paths, and the new reference-locked stack is generic, manifest-driven, and ready for broader examples. Real external use will surface edge cases that incremental v1.x patches will address.
  253. ---
  254. ## Contributing
  255. Open issues / PRs at https://github.com/RobLe3/cc-blender-skill — especially welcome:
  256. - Validation runs ("I tried prompt X, got error Y")
  257. - Recipe contributions for the long tail (specific materials, lighting setups, camera moves)
  258. - Trigger-eval JSON files for any sub-skill (helps tune description triggering)
  259. - Worked example scenes with proof-renders
  260. ---
  261. ## License
  262. MIT.
  263. ## Author
  264. [RobLe3](https://github.com/RobLe3) with extensive collaboration with Claude (Sonnet 4.6, Opus 4.7, Haiku 4.5).

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.