The AI skills library22 Claude agent skills turn tribal knowledge into something another tech artist can run: the QA method, the tutorial format, the asset process, the rules for touching code we don't own. A selection:
Process
The two-phase QA method: static checks (filesystem, git, YAML, GUIDs) before Unity opens, then in-editor test passes; also writes the Figma requirements audit and files the tickets.
Content
Tutorial authors and updates artist-facing guides with editor screenshots. Tutor answers an artist’s “how do I…” as numbered clicks read straight out of those guides — never code, never a doc dump.
Process
The standing rule for code we don’t own: never edit a shared module in place. Worktree isolation, a feature branch, a dev-facing note, a pull request and the feedback task that carries the link.
Content
Renders a studio-lit thumbnail of every prefab and model in a package into a filterable HTML guide, and lays the same assets out in one review scene by category — problems on a red strip.
Content
Audits a new game’s asset dump against the asset-manager checklist: what survives import, duplicates, folder weight, a must-have / nice-to-have split and a contact sheet ready to share.
Module brain
One skill per game module — units, stats, merging, physics, cameras, audio — so “why does this unit deal no damage?” is answered from the Inspector, without reading the code.
Process
Package plumbing: local checkouts without ever committing local paths, the Samples~ link Unity needs to see hidden samples, and the scaffold for a new samples repo.
Process
Captures Unity Profiler data from Play mode and grades a playground against the handover budget — 60 FPS median, no frame over 100 ms, zero logged errors.
Ops
Daily brief, who runs dry on tasks, asset deadlines, 1:1 canvases and calendar blocking — the routines that keep a distributed tech-art team visible without status meetings.