# whatsinagame A Claude Code starter kit that bootstraps multi-agent team infrastructure for any project. Clone once, install the `/kit-install` skill globally, run it in any project. Claude detects your tech stack, asks about team composition, and deploys adapted templates — agents, skills, database, decisions, and workflows — tuned to your project. Run `/kit-update` later to pull new skills and improved patterns from the source repo. ## Features - **Agent archetypes with tunable personalities** — architect, PM, QA, designer, developer, content author, consultant, and more. Each archetype has a personality spectrum (precise, creative, organized, holistic, empathic) that maps to concrete voice, priorities, and constraints. - **Stakeholder persona panel** — built-in personas (power-user, casual-user, product-advocate, marketer) plus project-specific ones. Personas participate in workshops and reviews to represent real user perspectives. - **Sprint lifecycle** — plan sprints, assign tickets, write briefings, run standups, track progress, run retros, cut release notes. - **Workshop orchestration** — multi-agent design workshops with structured rounds, facilitators, and decision capture. - **Decision tracking** — domain-organized decision files with IDs, status tracking, and cross-references. Never lose a design decision again. - **Ticketing database** — SQLite-backed project management with CLI tools. No external services required. - **PR review pipelines** — spawn reviewers matched to branch type (code, copy, visual, audio) in parallel. - **Worktree workflows** — multi-branch development with sync skills, or single-branch if you prefer. - **Git host agnostic** — patterns for GitHub (`gh`), Gitea (`tea`), and GitLab (`glab`). ## Profiles | Profile | Agents | Stakeholders | Skills | Infrastructure | |---------|--------|--------------|--------|----------------| | **minimal** | 3 | none | 2 | CLAUDE.md, settings.json, Makefile | | **standard** | 6 | 2 personas | 8 | + DB, CLIs, decisions, hooks, TEAM.md | | **full** | 11 | 4 personas | 17 | + Qdrant, briefings, sprints, workshops, DEVOPS.md | See `skill/references/profile-manifests.md` for exact file lists per profile. ## Installation ```bash # Clone the kit git clone https://github.com/yourname/whatsinagame.git # Install the skills globally cp -r whatsinagame/skill/ ~/.claude/skills/kit-install/ ``` ## Usage ```bash # Navigate to your project cd my-project # Start Claude Code claude # Run the installer /kit-install ``` Claude will: 1. Detect your tech stack (package.json, Cargo.toml, project.godot, pyproject.toml, etc.) 2. Ask you about profile, team composition, naming theme, and services 3. Deploy adapted templates — not copies, but project-specific rewrites 4. Initialize the database, verify connectivity, and print a getting-started summary ### Updating After initial install, check for kit updates periodically: ```bash /kit-update ``` This compares your installed files against the source repo, shows what's new (new skills, bug fixes, improved patterns), and lets you cherry-pick updates without losing your customizations. ## Directory Structure ``` whatsinagame/ README.md # This file LICENSE # MIT skill/ SKILL.md # The /kit-install installer skill references/ profile-manifests.md # What each profile includes archetype-gallery.md # Agent archetypes and stakeholder personas personality-guide.md # Tuning personality to role git-host-patterns.md # CLI patterns per git host customization-examples.md # Inspiration from The Settled Reach templates/ .claude/ agents/ # Agent personality templates skills/ # Skill templates (git-commit, ticket, pr-review, etc.) .config/ hooks/ # Git hook templates decisions/ # Decision tracking templates static/ db/ schema.sql # Ticketing database schema connectors/ # CLI tools (ticket, sprint, sqlite, qdrant) tests/ test-install.sh # Smoke test for all three profiles test-completeness.sh # Kit self-check ``` ## How It Works The `/init-team` skill is the heart of the kit. It doesn't just copy files — it reads the reference templates, understands your project context, and writes adapted versions. An architect agent for a game project gets different priorities than one for a SaaS API. A PM for a solo developer gets a lighter process than one for a 5-person team. The templates in `templates/` are examples of what good output looks like. Claude uses them as reference material, not as copy-paste sources. ## Credits Extracted from [The Settled Reach](https://github.com/jpmschweitzer/settled-reach), a top-down immersive sim built with a 20-agent Claude Code team. The patterns in this kit emerged from building that game — sprint planning, design workshops, decision tracking, PR reviews, and all the coordination infrastructure that makes multi-agent development actually work.