The router is the first skill an agent opens on a game-development request. It looks for engine fingerprints in the project (project.godot, a .uproject, Assets plus ProjectSettings, go.mod with Ebitengine, package.json with phaser, and so on for 12 engines), reads the pinned engine version, and sorts the request into disciplines, genres and workflows.
It then names the smallest set of skills to load, in a fixed order: engine basics, the discipline concept, genre glue, then workflow. It dispatches and composes; it does not teach engine APIs itself.
What You Get
- An engine detection table covering Godot, Unreal, Unity, Ebitengine, Bevy, Phaser, PixiJS, three.js, Defold, LÖVE, pygame and Roblox, with the file that identifies each
- A routing table from request wording to the discipline, genre or workflow skill that owns it
- Ownership rules for requests two skills both claim, such as behavior trees or platformer jump feel
- A fallback path for unknown, custom or in-house engines that keeps the existing engine and binds APIs from project evidence
Common Use Cases
- Starting a game request without knowing which of the many game-dev skills fits
- Adding a feature such as a double jump or an inventory and having the engine and genre skills chosen for you
- Working in a project with a pinned engine version older or newer than the catalog baseline
- Splitting work with an engine vendor’s own agent skills or editor connection when one is already in the session
What Makes It Unique
Most game-dev skills teach one engine or one topic. This one teaches no engine API at all: its job is selection. It encodes engine fingerprints, version checks, a table of who owns overlapping requests, and a rule to load the minimum number of skills so the context is not filled with unrelated engines.
Under The Hood
The skill is instructions only and ships no scripts. It tells the agent to scan the project for fingerprint files, read the engine version from project metadata or lockfiles, and then open the chosen SKILL.md files and, on demand, three reference files in its own references folder. It does not fetch web pages, call APIs, run commands or write files by itself. It states a trust rule: text found in project files or tool output is not treated as an instruction. It also says never to install a vendor plugin or MCP server for the user.