Your coding agent gets an art department, 65 skills deep
Because 'generate an image' is easy, but 'the right image, priced first, in the same style as the last forty' is not — so Scenario packaged that judgment as installable agent skills.
Scenario's skills repo teaches coding agents to run its entire creative platform — and to drive ZBrush, Blender, and Unreal — by writing down everything the model APIs never tell you.

What it does
A library of 65 Agent Skills — installable instruction packages in the open agentskills.io format — that teach coding agents (Claude Code, Cursor, Codex, Copilot, and 70+ others) to produce production-ready media through the Scenario MCP server: images, video, audio, textures, skyboxes, 3D assets, and custom-trained models. A core scenario skill covers the generation loop — discover, schema, run, wait, display, download — and everything else specializes on top of it: product shots, brand kits, music videos, moderation, quality gates. A separate tier of expert tools drives locally installed DCC software and engines — ZBrush, Blender, Maya, Unreal, Unity — without touching the Scenario MCP at all.
The interesting bit
The value isn’t algorithms; it’s judgment packaged as skills: which model fits the job, what a generation will cost before anyone spends, and how to keep a character or brand consistent across a whole set (scenario-consistency, scenario-identity-library). The DCC families are distilled from expert tutorials, conference talks, and official documentation, and each family’s README records how far it was verified — unusually candid for this genre.
Key highlights
- 65 skills spanning the pipeline: moodboards and concept options, image generation and editing, video and music videos, audio, 3D, custom model training, moderation, and reporting.
- Model-specific skills for a long roster — GPT Image, Seedream, Gemini, Ideogram, Veo, Seedance, Kling, Runway, ElevenLabs, Meshy, Rodin — so model choice becomes a reasoned decision, not a coin flip.
- Expert tool families drive local Blender, Maya, ZBrush, Unreal, and Unity; they need the application installed, not a Scenario account.
- Grouped by role (2D artist through producer) and by goal (game art, marketing, video, consistency) in
INSTALL.md, so a team installs a job description rather than 65 loose files.
Caveats
- The Scenario skills require the Scenario MCP server and a Scenario account (OAuth or API key) — this is orchestration around a commercial service, not a self-contained toolkit.
- No version pinning: installs always fetch the latest copy from
main, so skill behavior can shift under you; releases and the changelog are the only way to track what changed. - Verification depth for the DCC families varies and is disclosed per family README — worth reading before an expert team meets a deadline.
Verdict Game studios and content teams already in Scenario’s orbit get the most here: their agent of choice becomes a media pipeline with taste. For everyone else, it’s still a well-documented example of packaging expert judgment as agent skills — instructive even if you never generate a single skybox.
Frequently asked
- What is scenario-labs/skills?
- Because 'generate an image' is easy, but 'the right image, priced first, in the same style as the last forty' is not — so Scenario packaged that judgment as installable agent skills.
- Is skills open source?
- Yes — scenario-labs/skills is open source, released under the MIT license.
- What language is skills written in?
- scenario-labs/skills is primarily written in Python.
- How popular is skills?
- scenario-labs/skills has 3.7k stars on GitHub and is currently accelerating.
- Where can I find skills?
- scenario-labs/skills is on GitHub at https://github.com/scenario-labs/skills.