Claude Code's single session becomes a game studio: 49 agents and 74 skills, gated from directors down to specialists
Synopsis
Claude Code Game Studios turns one Claude Code session into a game-development studio with a hierarchy: 49 specialized agents in three tiers of directors, department leads and specialists, plus 74 slash skills, each with defined responsibilities, escalation paths and quality gates. It targets the missing structure of a single chat session, where nothing stops hardcoded magic numbers, skipped design docs or spaghetti code. Four games built from one brief at different rigor levels all produced a running, tested game; two blind reviewers and a human playtest ranked the standard build last, so the extra 29 documents bought traceability rather than a better game.
Interpretation
A single Claude Code session can now be organized like a real studio hierarchy rather than one general-purpose assistant: 49 specialized agents sit in three tiers, with directors guarding creative and technical vision, department leads owning design, programming, art, audio, narrative, QA, release and localization domains, and specialists doing the hands-on work. The prior condition was building a game solo with AI inside one chat session, which has no structure: no one stops hardcoded magic numbers, skipped design documents or spaghetti code, and there is no QA pass, no design review, and no one asking whether a decision fits the game's vision. Each agent declares a model in its frontmatter: the three directors ask for Opus, lead-programmer and fifteen specialists name Sonnet, and the remaining thirty inherit the session's model; the template also includes agent sets for all three major engines.
Beyond the hierarchy, the studio exposes callable entry points: typing slash in Claude Code reaches 74 skills spanning onboarding and navigation, game design, art and assets, UX and interface, architecture, stories and sprints, reviews and analysis, QA and testing, production, release, creative and content, and team orchestration that coordinates multiple agents on a single feature. These skills turn checks that previously lived loosely in the session into named steps, such as design review, code review, balance check, scope check, performance profiling, tech-debt review, security audit, smoke check, soak test, regression suite and release checklist. The team-orchestration skills include team-combat, team-narrative, team-ui, team-release, team-polish, team-audio, team-level, team-live-ops and team-qa, used to coordinate several agents on one feature.
Whether the structure pays off is measured by a build experiment: four games start from one brief and are built at different rigor levels, counting the documents each build writes before its first line of game code. All four builds produced a running, tested game, so the difference is not whether a game can be made but how much traceable record the process leaves behind. Two blind reviewers and a human playtest then ranked the standard build last, indicating that the extra 29 documents bought traceability rather than a better game.
Perspective
The studio template targets solo game development inside Claude Code and provides agent sets for all three major engines; the build experiment covers only four games built from one brief at different rigor levels.
The extra 29 documents bought traceability rather than a better game, and the standard build ranked last in blind review and human playtest; a passing test says nothing about whether something like a menu is drawn correctly, so the result still needs someone to look at it.
