Contains Studio Agents: Building a Software Studio with 40+ Specialized AI Personas
Hook
What if instead of one AI coding assistant, you had an entire virtual software studio—complete with a backend architect who disagrees with the DevOps automator, a whimsy injector who makes loading screens fun, and a Reddit community builder who knows how not to get banned?
Context
AI coding assistants have evolved rapidly from autocomplete-on-steroids to context-aware pair programmers. But they’ve remained fundamentally generalist: one assistant trying to be good at everything from React hooks to database indexing to App Store optimization. Contains Studio takes a radically different approach inspired by how actual software studios work—specialized roles, not generalists.
The repository contains 40 AI agent definitions designed for Claude Code. Each agent is a domain specialist: the rapid-prototyper builds MVPs in days, the feedback-synthesizer transforms user complaints into feature specs, the legal-compliance-checker keeps you from accidentally violating GDPR. Rather than asking a single AI to context-switch between writing backend APIs and crafting TikTok strategies, you invoke the right specialist for each task. The philosophy mirrors the 6-day sprint methodology—ship fast, delegate to experts, iterate relentlessly.
Technical Insight
Each agent is defined as a markdown file with YAML frontmatter in ~/.claude/agents/. The structure is deceptively simple but remarkably effective:
---
name: rapid-prototyper
description: Build MVPs in days, not weeks. Examples - "Create a meditation tracking app", "Build a receipt scanner prototype"
color: orange
tools:
- read
- write
- multiedit
- bash
---
You are the Rapid Prototyper for Contains Studio...
[Detailed system prompt defining expertise, philosophy, and behavior]The architecture appears to leverage Claude Code’s sub-agent system. When you describe a task, Claude Code matches it against agent descriptions and invokes the appropriate specialist. Say “Create a meditation habit tracker” and the rapid-prototyper activates. Mention “Our app reviews are dropping” and the feedback-synthesizer takes over. You can also explicitly request agents: “Hey brand-guardian, does this color scheme match our identity?”
What makes this powerful is the specialist delegation pattern. The backend-architect agent has a system prompt focused entirely on API design, scalability, and data modeling. It doesn’t waste tokens on UI concerns. Meanwhile, the ui-designer thinks in components, design systems, and accessibility—it won’t get distracted suggesting database optimizations. Each agent is a tightly scoped expert.
The repository organizes agents into departmental folders that mirror real studio structure:
~/.claude/agents/
├── engineering/
│ ├── rapid-prototyper.md
│ ├── backend-architect.md
│ ├── frontend-developer.md
│ └── test-writer-fixer.md
├── marketing/
│ ├── tiktok-strategist.md
│ ├── reddit-community-builder.md
│ └── growth-hacker.md
├── product/
│ ├── trend-researcher.md
│ └── feedback-synthesizer.md
└── bonus/
├── studio-coach.md
└── joker.mdInstallation is straightforward:
git clone https://github.com/contains-studio/agents.git
cp -r agents/* ~/.claude/agents/
# Restart Claude CodeSome agents are proactive—they trigger automatically in specific contexts. The test-writer-fixer is designed to activate after you implement features or fix bugs. The whimsy-injector jumps in after UI changes to add delightful micro-interactions. The studio-coach meta-agent coordinates multi-agent workflows when tasks require multiple specialists.
This creates emergent collaboration patterns. Building a new feature might invoke the rapid-prototyper for the initial implementation, then the test-writer-fixer for coverage, then the ui-designer for polish, and finally the app-store-optimizer to write the release notes. Each agent contributes its specialized expertise to the pipeline.
The system prompt engineering is where the magic happens. Each agent embodies not just technical knowledge but a philosophy. The rapid-prototyper is explicitly instructed to prioritize shipping over perfection, to use battle-tested tools, to build for today’s scale. The legal-compliance-checker balances speed with regulatory reality, suggesting pragmatic compliance strategies rather than blocking launches. The joker agent exists purely to lighten the mood with tech humor during stressful deploys.
The 6-day sprint philosophy permeates every agent. The sprint-prioritizer helps you ruthlessly cut scope to ship maximum value in under a week. The project-shipper focuses on launch-critical tasks, deferring polish. The studio-producer keeps teams shipping instead of meeting. This isn’t a neutral tool collection—it’s an opinionated framework for rapid development.
Gotcha
The biggest limitation is Claude Code lock-in. These agents are completely dependent on Claude Code. You can’t use them with GitHub Copilot, Cursor, or any other AI coding assistant. If you’re committed to VS Code with Copilot or JetBrains with their AI assistant, this entire repository is useless to you. There’s no abstraction layer, no compatibility shim—it’s Claude Code or nothing.
Agent quality relies entirely on prompt engineering. There are no fine-tuned models, no specialized training, no actual code logic behind each agent. They’re system prompts that guide behavior. This means agent behavior can be inconsistent, context limits still apply, and you’re trusting that a well-crafted prompt is sufficient to create true specialization.
With 40 agents, there’s cognitive overhead in understanding the taxonomy. Do you want the content-creator or the visual-storyteller? Is your task better suited for the trend-researcher or the growth-hacker? The repository assumes you’re working in the specific “rapid startup studio” context these agents were designed for. If you’re building enterprise software with 6-month release cycles, or working on embedded systems, or doing academic research, most of these agents (TikTok strategist, Reddit community builder, 6-day sprint prioritizer) won’t match your workflow. The opinionated nature that makes them powerful for their target audience makes them awkward outside that context.
Verdict
Use if: You’re already using Claude Code and building consumer apps or MVPs rapidly. The agent collection excels at the full startup lifecycle—prototype to launch to growth. If you’re a solo developer or small team shipping frequently and wearing multiple hats (code + marketing + ops), having specialized AI teammates for each role is genuinely valuable. The proactive agents and multi-agent coordination reduce context-switching overhead. Use if you embrace the 6-day sprint philosophy and want AI assistance that reinforces shipping over perfecting. Skip if: You’re committed to a different AI coding tool (Copilot, Cursor, etc.), you prefer a single general-purpose assistant over managing a roster of specialists, or you work outside the rapid consumer app development context. Skip if you’re building enterprise software, working in regulated industries with strict compliance requirements, or if your development cycle doesn’t align with the “move fast and ship” mentality baked into every agent. The value proposition is narrow but strong—this is a power tool for a specific type of developer, not a general-purpose solution.
