team-arch-opt

Multi-agent architecture optimization is coordinated through a pipeline of Analyzer, Designer, Refactorer, Validator, and Reviewer phases. This pipeline systematically analyzes code structure, designs prioritized refactoring steps, executes changes, and validates outcomes with structured outputs at each stage.

A fan-out branch refactor is a parallel refactoring strategy where independent changes are executed across separate branches. You should use this approach, or let auto mode select it, when you have multiple refactoring targets to improve coupling, layering, duplication, or dead code simultaneously.

The review-fix safety loop handles validation failures by automatically creating targeted fix tasks when validation or review reports issues. This enforces safety checks during architecture optimization and escalates fix cycles until the refactoring outcomes meet the required validation verdicts.

Yes, you need a YAML-defined skill entrypoint to use this architecture optimization skill. The YAML entrypoint is required to orchestrate role-based phases with dependency-aware task spawning for analyzing code structure and executing coordinated multi-agent refactoring plans.

Evidence-based deliverables produced after reducing code coupling and layering include baseline metrics, ranked architecture issues, a prioritized refactoring plan, validation results, and a review report. These structured outputs document the entire architecture optimization pipeline from analysis to verification.