How do I structure my monorepo and workflow to keep Claude Code consistent as a project grows?

I'm developing a project using Claude Code, and I'm starting to run into a few problems as the codebase grows. First, I increasingly have to remind Claude of the same rules and constraints over and over again. Despite having project instructions and architectural decisions documented through ADRs, it gradually starts violating those rules, taking shortcuts, or making changes that don't align with the intended architecture. Second, my context window is being consumed much faster than I'd expect. I suspect my repository structure and the way I've organized project documentation and instructions may be contributing to the problem. This has made me question whether my current monorepo structure and development workflow are actually sustainable in the long run. I'm looking for advice on how to address this systematically rather than continuing to patch the problem with more instructions. Some areas I'm particularly interested in: Repository architecture: Best practices for structuring a maintainable monorepo, organizing modules, defining boundaries, and preventing architectural drift. I read about DDD (Domain-Driven Design) and Hexagonal Design, but not sure they are the right fit for me. Agent instructions and enforcement: How should CLAUDE.md, AGENTS.md, ADRs, skills, hooks, and automated checks work together? Which rules should be documented, and which should be enforced mechanically? Context management: How can I reduce unnecessary context consumption without losing important architectural knowledge or forcing Claude to reread the entire codebase? Workflow and methodology: Are there established practices, tools, or development workflows that make Claude Code more reliable as a project grows? Other blind spots: Am I misdiagnosing the problem? Could the main issue be how I delegate tasks, scope changes, manage architectural decisions, or validate the agent's work rather than the repository structure itself? I'm not necessarily looking for a specific skill or tool recommendation (they are also welcomed). I'd appreciate advice on the underlying principles and practices that experienced developers have found effective when building larger projects with Claude Code. How would you approach this from the ground up if you wanted a codebase that remained maintainable and an AI coding agent that followed its architectural constraints consistently over time?