A Two-Tier Agent Architecture
PhoenixCode's AI isn't a single model. It's an orchestrated team of 21 specialized agents wired into the migration pipeline — 7 masters and 14 specialists — each with a specific role and prompt template.
Tier 1: 7 Master Agents
Master agents coordinate the pipeline. Each owns a phase of the migration, makes its high-level decisions, and dispatches focused work to specialists.
- Analysis Master: Legacy code understanding, dependency mapping, business rule extraction
- Architecture Master: Clean Architecture design, project structure and patterns
- Backend Master: .NET 10 code generation across controllers, services, repositories
- Frontend Master: Angular 22 components, routing, state management
- Quality Master: Test generation, security scanning, functional equivalence
- Documentation Master: API docs, migration reports, deployment guides
- Deployment Master: CI/CD, Docker, Kubernetes, and Infrastructure as Code
Tier 2: 14 Specialist Agents
Specialists are cross-cutting experts the masters invoke for focused tasks — performance, security, API-contract validation, architecture review, SRE, test planning, correction, pattern extraction, and more. They do the individual transformation work: converting a Delphi form, generating a unit test, or migrating a stored procedure.
Why This Architecture Works
Masters use Claude Opus for reasoning; specialists use Claude Sonnet for speed. 106 prompt templates ensure consistency across all 21 agents. A separate toolchain of 22 Claude Code agents handles development-time code review and quality — bringing the full ecosystem to 43 agents.