PlatformPlatform Multi-Agent Development Platform
An enterprise-grade, multi-tenant SaaS blueprint on Azure, .NET 10 and React, with a Claude Code multi-agent team workflow that delivers complete features from requirements to deployable code.
Evidence: passwordless OIDC CI/CD logins, secrets via GitHub Secrets and .NET user secrets, paired firewall open/close with if:always, health-check-gated deploys — supporting moderate scores for sensitive data handling and external effects. Deducted: the Agent layer (Claude Code Agent Teams, 30+ rules/skills) shows no evidence of its permission scope, confirmation mechanisms, or data flow; least_privilege/user_confirmation/data_flow_transparency/source_attribution rest on one README sentence.
Evidence: concurrency groups, health-state polling with fail-fast, well-structured e2e tests including a security boundary check (404 on cross-host requests). Deducted: the Agent workflow's own failure handling and self-consistency are README assertions without supporting files; pinned action versions (actions@v7, not yet released) raise a self-consistency doubt, so not full marks.
Evidence: per-OS prerequisites, clear audience (B2B/B2C SaaS builders), optional Aspire/Stripe/Google OAuth setup — good environment fit. Deducted: the agents' capability boundaries and trigger precision (which agent handles what, when triggered) have no concrete description in the supplied files.
Evidence: exceptionally detailed install notes (step-by-step for Windows/Mac/Linux, 11-step Stripe sandbox guide), complete MIT license, clear top-level structure, CI badges and roadmap links, visible maintenance path. Deducted: no CHANGELOG/version file, no known-limitations section, and the AI rules/agent definition files are absent from the evidence, so information architecture is only partially verifiable.
Evidence: high output usability (full SaaS platform plus agent dev flow), one-command deploy and Aspire local stack lower cost, clear marginal value versus assembling parts. Deducted: the Agent portion's actual output quality and cost/benefit rest on README claims ("battle-tested over a year") with no file-level evidence, so not full marks.
Evidence: README claims on deployment, secrets, staging-to-prod image import, and DNS prompting cross-corroborate with the workflow file contents. Deducted: promotional claims ("30+ rules", "refined over a year") are not traceable in the evidence, and facts versus inferences are not cleanly separated; both scores capped at 1.
- This is a static source review with low confidence; no code or Agent workflows were executed.
- Core Agent-layer files (Claude Code Agent Teams, 30+ rules/skills) are absent from the evidence; their permission scope, confirmation mechanisms, and data flows are entirely unverified.
- README claims such as "battle-tested over a year" and "30+ rules" are not traceable; inspect the rules files and commit history yourself before adoption.
- Referenced action versions (e.g., actions/checkout@v7, download-artifact@v8) appear ahead of actual releases — possibly placeholders or mispins; verify before deploying.
- The deploy command creates service principals and modifies the Azure subscription; review its change list and required permissions carefully before running.
What does this agent do, and when should you use it?
PlatformPlatform is an MIT-licensed monorepo providing a complete blueprint for building enterprise-grade, multi-tenant B2B and B2C SaaS products. The backend uses .NET 10 and C# 14 with vertical slice architecture, DDD, and CQRS; the frontend uses React 19, TypeScript, TanStack Router/Query, and ShadCN 2.0. Infrastructure is deployed as code (Bicep) to Azure Container Apps and PostgreSQL, with passwordless CI/CD via GitHub Actions. Its core differentiator is a multi-agent development workflow built on Claude Code's native Agent Teams, where a team lead, engineers, reviewers, a Guardian, and an architect collaborate to deliver complete features. The repo ships production-ready product capabilities out of the box: login (Google OAuth/email OTP), Stripe subscription billing, feature flags, and a back-office dashboard. Note that the multi-agent workflow explicitly requires Claude Code and will not work with other AI coding assistants.
The repository contains application (the main and account self-contained systems, shared-kernel, shared-webapp), cloud-infrastructure (Bicep scripts), developer-cli (a .NET CLI), and a .claude directory (agent definitions, hooks, 30+ rules and skills). The flow: a user runs pp claude-agent team-lead to start the team lead agent; the /create-prd skill interviews the user and generates a feature specification with tasks in the configured product management tool (Linear, Azure DevOps, Jira, or Markdown). The team lead starts in plan mode and, after user approval, spawns persistent agents — guardian (sole owner of git commits), architect, regression-tester, researcher — plus fresh backend/frontend/qa engineer and zero-tolerance reviewer pairs per task set. Agents coordinate through Agent Teams primitives (TeamCreate, Task, SendMessage) and a custom MCP command, SendInterruptSignal, that enables real-time interrupts; backend, frontend, and E2E tracks run in parallel, and the Guardian runs a pre-commit pipeline (build, test, format, lint, Aspire restart, smoke tests) before committing in dependency order. On the deployment side, pp deploy (developer-cli dotnet run deploy) automates Azure subscription configuration and GitHub Workflows that deploy Docker containers to Azure Container Apps, achieving a 100% security score in Microsoft Defender for Cloud.
- A startup launching a multi-tenant B2B SaaS product: clone the repo and immediately get signup/login, Stripe subscriptions, account settings, and other complete user-facing features
- Engineering teams with a Claude Code subscription who want a multi-agent team to autonomously implement complete features — backend logic, database migrations, APIs, frontend UI, localization, and E2E tests
- Teams using Linear, Azure DevOps, or Jira who want agents to track features and task statuses directly inside their existing product management system
- Enterprises that need deployment to Azure with zero secrets (Managed Identities, federated credentials) and auto-scaling
- .NET developers learning vertical slice architecture, DDD, CQRS, and self-contained systems, using the preserved commit history as a learning resource
- Solo developers or small teams needing a low-cost entry point (under 2 USD/day per cluster) that scales to millions of users
What are this agent's strengths and limitations?
- Key differentiator: a multi-agent workflow on Claude Code Agent Teams featuring zero-tolerance code reviews, a dedicated Guardian agent owning all git commits, and a custom SendInterruptSignal MCP command that solves real-time agent collaboration
- Complete SaaS capabilities out of the box: Stripe subscription and billing (prorated upgrades/downgrades, dunning, invoices), Google OAuth login, feature flags (A/B rollout with four override levels), and a back office with MRR dashboards
- Zero-secret Azure architecture: Managed Identities and federated credentials throughout, a 100% security score in Microsoft Defender for Cloud, fully automated deployments, and cluster costs under 2 USD/day
- 30+ AI rules and skills refined over a year of daily use, encoding opinionated architecture patterns so agents produce consistent code
- The multi-agent workflow explicitly requires Claude Code and will not work with other AI coding assistants — a hard dependency on the Anthropic ecosystem
- Cloud infrastructure is deeply bound to Azure (Container Apps, Service Bus, Key Vault, Entra ID, etc.); migrating to AWS/GCP means rewriting all Bicep infrastructure
- Heavy prerequisites: .NET 10, Node, Docker, Azure CLI, GitHub CLI; Stripe and Google OAuth require involved dashboard configuration (each needing two Aspire restarts)
- A full multi-agent feature implementation can take several hours, and the README offers no benchmarks or independent evaluation of output quality
- The stack is opinionated and enforced (C# 14/React 19/ShadCN); teams off this stack face substantial adaptation costs
How do you install or deploy this agent?
Prerequisites: .NET 10 SDK, Docker, Node.js (version must match .node-version; fnm recommended), plus GitHub CLI and Azure CLI for CI/CD. Steps:
- git clone https://github.com/platformplatform/PlatformPlatform.git
- (Optional) Install the Developer CLI: cd developer-cli && dotnet run install, making the pp command available globally
- Start the local stack: pp run (or cd application/AppHost && dotnet run) — Aspire pulls Docker containers for PostgreSQL, a Blob Storage emulator, and a development mail server
- Optional: install the Aspire CLI (irm https://aspire.dev/install.ps1 | iex or curl -sSL https://aspire.dev/install.sh | bash) to enable the Aspire MCP server; and npm install -g typescript-language-server typescript plus dotnet tool install -g csharp-ls for enhanced Claude Code support
- The multi-agent workflow requires Claude Code itself to be installed (the README does not provide its installation command)
How do you use this agent?
- Create a feature branch: git checkout -b feature-name
- Set PRODUCT_MANAGEMENT_TOOL in AGENTS.md (options: Linear, AzureDevOps, Jira, Markdown) and configure the MCP server per the matching file in .claude/reference/product-management/
- Run pp claude-agent team-lead to start the team lead, then use the /create-prd skill to generate a feature spec and tasks via a guided interview
- Tell the team lead which feature to implement (title or ID); it automatically spawns guardian, architect, regression tester, and per-task-set engineer/reviewer agents that implement in parallel, review line-by-line, and let the Guardian commit
- For smaller tasks, run pp claude-agent pair-programmer instead of the full team
- CI/CD setup: cd developer-cli && dotnet run deploy — requires ownership of the GitHub repo and Azure subscription plus permissions to create service principals; store Google OAuth and Stripe credentials via pp github-config into GitHub secrets and Azure Key Vault
How does this agent compare with similar options?
The README positions itself as 'a ready-made blueprint, not a pile of parts to assemble,' in contrast to conventional SaaS starters/templates that require assembling components yourself. Architecturally it uses Self-Contained Systems (SCS) rather than the popular Backend-for-Frontend (BFF) pattern, allowing independent development and deployment per system. No specific competitors are named in the source.