super-human

Multi-file project planning and execution for an AI agent is coordinated by decomposing work into a dependency graph. This allows parallel wave-based execution with TDD verification across complex features, refactors, and migrations.

Wave-based execution is a mechanism that runs decomposed tasks in parallel waves based on a dependency graph. It includes cross-task verification, rollback points, and conflict resolution to safely manage complex multi-file development.

AI-generated code changes during multi-step refactoring are verified using TDD verification. The lifecycle mandates tail tasks like Self Code Review, Requirements Validation, and Full Project Verification to ensure correctness.

Dependency graph decomposition works for greenfield projects, major refactors, and migrations. It supports a seven-phase lifecycle from intake through converge, maintaining a persistent board at the project root for resume and audit.

A stalled AI agent task is resumed using a persistent board maintained at the project root. This board captures intake, plan, execution, and verification states, allowing the lifecycle to recover and continue execution waves.

The seven-phase lifecycle requires explicit user guidance and plan approval before execution. It is designed for complex feature work, refactors, and migrations, and may be excessive for simple single-file modifications.