// the find
Donchitos/Claude-Code-Game-Studios
Turn Claude Code into a full game dev studio — 49 AI agents, 72 workflow skills, and a complete coordination system mirroring real studio hierarchy.
A Claude Code configuration template that imposes a studio-style hierarchy on solo AI-assisted game development — 49 agent definitions, 73 slash commands, and a set of shell hooks that enforce naming conventions and commit standards. Aimed at indie developers who want process structure without hiring a team. Requires Claude Code and your own Anthropic API key.
The three-tier agent hierarchy (directors, leads, specialists) maps to how real studios actually divide responsibility, which means the escalation paths make sense rather than being arbitrary. Path-scoped coding rules are genuinely useful — prototype code in `prototypes/**` gets relaxed standards while `src/networking/**` enforces server-authority and message versioning, so Claude can't accidentally apply the wrong standard to the wrong layer. The hooks do real work: `validate-commit.sh` catches hardcoded magic numbers and malformed TODO comments before they land, and `detect-gaps.sh` warns on session open when code exists but design docs don't. The explicit collaboration protocol (ask first, show options, user decides, then draft) is an honest acknowledgment of where LLMs actually fail when given free rein.
The 49 agents don't actually coordinate — they're sequential prompt templates that share a working directory. The studio hierarchy is a UX fiction; if Claude skips a director gate, nothing enforces it, and there's no real escalation happening between context windows. The `src/` tree is completely empty in the template, which means the whole thing is process scaffolding with zero evidence it ships games; 22k stars reflect Claude Code hype more than validated workflow. Keeping 49 agent definitions and 73 skill prompts in sync with Claude Code API changes and model behavior shifts is a serious ongoing maintenance burden the README quietly admits to. And the tier assignments ('directors use Opus, specialists use Haiku') are hardcoded assumptions that will drift as pricing and model capabilities change.