Image: GitHub
Why it mattersA game studio or 3D artist can now hand a coding agent a brief like "ship a game-ready crate with baked normals and LODs" and get the file back without clicking through Blender by hand.
blender-skills is a new MIT-licensed collection of 22 Agent Skills that cover the main parts of a Blender 5.2 pipeline, so a coding agent can run modelling, UVs, PBR texturing, baking, rigging, animation, export and QA on its own. The repository is from a developer called EL4CTEO on GitHub and has drawn 273 stars in its first five days. It works with Claude Code, Codex, Gemini CLI, Copilot and Cursor.
What a skill contains
An Agent Skill, in the pattern Anthropic published and other agents have picked up, is a short folder that teaches a coding agent to do a specific job. It carries the written procedure, the specific Blender Python API calls that back it, and the checks the agent should run before and after. The agent reads the skill instead of writing the code from scratch, which means it does not have to rediscover a working rig setup or a baking configuration on every job.
The 22 skills here cover the shape of a real asset pipeline, not a tutorial dataset. There are skills for the retopology step, for the UV unwrap, for a full PBR texture setup, for baking normals and ambient occlusion into textures, for the rigging and skinning passes, for animation and the exports to glTF, FBX and USD. The QA skill is for checking a model after the fact, which is where most of the failures caught in a game studio pipeline actually sit.
Who it is for
The audience the project describes is 3D artists and small studios running Blender day to day, and the pitch is that a coding agent can carry a request from "ship a game-ready crate with baked normals and LODs for Unreal" to a working file. The agent calls Blender through its Python API, following the skill, and reports back when the output is ready.
That is a change in how an artist works. The traditional path is to open Blender and perform each step by hand, with the time cost sitting in the clicking rather than the thinking. A skill-driven agent does not help with the creative choices, which still have to be the artist's, and it does not know what a model should look like. It helps with the standard, repetitive pipeline moves between a model and a shipped asset.
Twenty-two skills released at once
Agent Skills are a new interface and 22 of them in one repository is a lot of surface. The skills in this collection have been released together and the repository is five days old, so a studio adopting them should expect to read each skill and test it against a known-good asset before trusting it on real work. The project is MIT-licensed, which means it can be forked and modified in place.
One subtle point. The skills tell the agent what to do, but they do not change the model doing the work. A weaker model will still make weaker choices inside a skill, so a studio switching from a frontier coding model to a cheaper one should expect the output to drop in the same way it drops on a coding task. For a shop that already runs a coding agent on code every day, adding Blender to the same agent is a small change.
Source
The code is at EL4CTEO/blender-skills on GitHub.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.