Generate Production Game Art With Scenario's 167-Skill Agent Library

Game art is a pipeline problem: one potion icon is easy, but forty style-matched icons, a tiling texture, and a skybox that survives engine export are where ad-hoc prompting falls apart. scenario-labs/skills is Scenario's official skill library for exactly that pipeline — it teaches your coding agent to drive generation with model choice, sizing contracts, and consistency rules instead of vibes.

Why This Skill Matters

The library targets Claude Code, Cursor, Codex, Copilot, and 70+ other agents, and every skill follows the open Agent Skills format. Counting the README's own tables gives 167 skills in two halves that work differently. The first 80 run through the Scenario MCP server and cover the full content pipeline: images, video, audio, textures, skyboxes, 3D assets, custom-trained models, workflows, and team administration. The other 87 are expert tools: teams of skills that drive DCC software and game engines installed on your machine — ZBrush, Blender, Maya, Unreal Engine, Unity, Godot, Roblox Studio — distilled from expert tutorials, conference talks, and official documentation. Those need the application, not the MCP server.

The model-family skills are the quiet differentiator. Beyond generic image generation, there are dedicated skills for GPT Image, MAI Image, Seedream, Gemini, Reve, Ideogram, Grok Imagine, and Luma on the image side, ten video model-family skills (MiniMax, Gemini Omni, Grok Imagine, Veo, Seedance, Kling, Vidu, Wan, Runway, and Luma), four audio families (MiniMax, ElevenLabs, ACE-Step, and Sonilo), and four 3D families (Meshy, Rodin, Sparc3D, and 3D worlds) — each documenting the member selection, caps, and cost traps that differ per model.

Installation

Pick from the installer menu, or install a single skill:

# Pick from the installer menu, grouped by topic (it opens with nothing preselected: select before pressing enter)
npx skills add scenario-labs/skills

# A single skill
npx skills add scenario-labs/skills --skill scenario

The README also documents one long command that names all 80 Scenario skills explicitly, plus pinned installs of the form npx skills@1.7.4 add scenario-labs/skills#<commit> --skill <name> when you want a reviewed release instead of the latest main. The menu's topic groups cannot be selected from the command line — the CLI matches --skill against exact skill names only.

The Scenario skills need the MCP server connected:

claude mcp add --transport http scenario https://mcp.scenario.com/mcp

Sign in with a Scenario account through OAuth, or use an API key. Expert tool families install as one set, since the specialists import the lead skill's scripts.

Real Workflow: Style-Matched RPG Icons

1. Install the scenario skill and connect the MCP server.
2. Ask your agent for the icons. The README's own example is a good start:

Generate four style-matched potion icons with transparent backgrounds for my RPG inventory

3. The core scenario skill runs a loop it documents as discover, schema, run, wait, display, download: the agent finds the right model, checks the request against the API schema, submits, waits for the job, shows you the result, and saves the files locally.
4. When the batch must hold one style across dozens of assets, scenario-consistency adds baseline-plus-delta prompts, reference images, and control maps; scenario-game-assets covers sprites, tilesets, and transparent-background batches.

The second workflow moves to your desktop while staying inside the Scenario skills: scenario-blender, one of the 80 MCP skills, drives an open Blender scene through the Scenario for Blender extension's local MCP server — it reads the scene, quotes and runs generations, captures viewport references, and places results. Separate from the 80, the Blender expert family is its own set: a lead skill plus fourteen specialists that drive Blender 5.2 through Python, headless or in a live session, with twelve test suites passing headless on Blender 5.2.1.

Tips

• Verification depth varies by family, and the README says so per family: the ZBrush agent bridge "was proven live" with full live tests of every skill still pending, while Unity's and Godot's procedures were each run live in the current versions (Unity 6.3, Godot 4.7.2).
• Changes land through releases: watch the repo's releases or the Scenario changelog, and pull with npx skills update where you installed.
• Every script shipped with a skill has a test suite in tests/<name>/; in a clone of the repo, pnpm test runs them (Python 3.11+ required, plus ffmpeg where a suite uses it).
• Blocked by a content filter? scenario-moderation covers provider-side filters, model switching, and rewording; scenario-report turns a stuck session into a reproducible issue.
• For role-based installs, INSTALL.md groups the skills by job, from 2D artist to producer, with one command each.

When Not to Use This

The Scenario half runs generation against Scenario's cloud — you need a Scenario account, and model access and costs follow that account. If you never use Scenario's models, only the expert-tool families apply, and they need the actual applications installed on your machine (several families document macOS-driven verification).

See the leaderboard for more skills.