scenario-labs/skills

What this is

This repository packages Agent Skills — SKILL.md files following the open Agent Skills specification — that teach AI coding agents how to work with Scenario through its MCP server. Instead of the agent guessing at prompts or model parameters, a skill supplies procedural knowledge: which model fits a job, how to size a request, how to price it before spending, and how to keep a character or brand consistent across a set. Coverage spans images, video, audio, textures, skyboxes, 3D assets, and custom-trained models, aimed at games, entertainment, and other creative work.

A second group, the "expert tools," works differently: these are skill teams that drive DCC applications and game engines installed locally — ZBrush, Blender, Maya, Unreal Engine, Unity, Godot, and Roblox Studio. They require the application, not the Scenario MCP server. Each family folder's README documents how it was built and how far it was verified.

The skills install into Claude Code, Cursor, Codex, Copilot, and other compatible agents via the skills CLI.

Install and first run

You need Node.js with npx available, and for the Scenario skills an agent connected to the Scenario MCP server. Start with the core skill that covers connecting to the MCP and the discover → schema → run → wait → display → download loop:

npx skills add scenario-labs/skills --skill scenario

To install every Scenario skill except the expert tools in one command, use the full --skill list shown in the README's Install section install commands. Running npx skills add scenario-labs/skills without --skill opens an interactive menu that starts with nothing preselected — select before pressing enter installer menu.

For a role or outcome-based selection, INSTALL.md groups skills by job and by goal with one command each.

Once installed and connected, prompt your agent normally, for example: "Generate four style-matched potion icons with transparent backgrounds for my RPG inventory." The skill handles model choice, request fields, and the generation loop.

Expert tool families install as a set because specialists import the lead skill's scripts; use the installer's "Expert tools" group or the command in the family README (Blender, Maya, Unreal, Unity, Godot, Roblox, ZBrush).

Installing only what you need

Skills are modular and can be installed individually. Groups cover: direction finding and concept options; image generation and editing; game art and 3D; video and audio; social formats; consistency and custom model training; output review and quality gates; placement formats; workflows and apps; troubleshooting; team administration; and per-family deep dives for image, video, audio, and 3D models. Pick the skill matching your task rather than installing everything at once.

Constraints and requirements

• Scenario skills need the MCP server: they drive Scenario through its MCP server, so an authenticated connection is required before generation works.
• Expert tools need the application: ZBrush, Blender, Maya, Unreal, Unity, Godot, or Roblox Studio must be installed locally. Verification varies by family — the Blender, Unity, Godot, and Roblox families report live-tested procedures, while the ZBrush bridge was proven live but full live tests of every skill are still pending.
• Host platform matters for expert tools: the Unreal, Unity, Godot, and Roblox families are documented for macOS, with specific version numbers noted.
• Spending: skills are designed to price requests before spending and include dry-run cost matrices, but actual generation consumes Scenario credits.
• Updates: the install commands fetch the latest copy from main. Pin a reviewed release if you need stability. Run npx skills update to pull changes update notes.
• License: MIT.

HotRepo's take

On 2026-11-01 this repository entered HotRepo's Python daily board at position 10 with 98 stars gained that day and 1,418 total stars. That is a strong single-day signal for a skills package, and the reason is visible in the structure: the repository does not just wrap one model API. It encodes per-family model knowledge — GPT Image tiers, Veo modes, Meshy parameter dialects — alongside expert-tool teams for seven DCC applications and game engines. The Blender family alone claims twelve test suites passing headless on Blender 5.2.1, and the Unity and Godot families claim every procedure was run live in their target versions. Those are the project's own verification statements, but they indicate a higher testing bar than typical prompt collections.

The weak spot is the ZBrush family, where the README itself says full live tests of every skill are still pending. For teams considering the expert tools, check the per-family README for its verification status before relying on it in production. The single snapshot does not establish a trend, and the Python daily board position reflects one day's growth, not sustained ranking.