Context Engineering Collection

If you're building agent systems that keep running into context limits or losing track of state across conversations, this collection covers the architecture patterns that actually matter. The filesystem-as-memory approach is smarter than you'd expect, multi-agent coordination patterns focus on context isolation rather than role-playing, and the evaluation guidance keeps you honest about what's actually working. The latent briefing section on KV cache sharing between orchestrators and workers is legitimately novel. It's opinionated about things like tool consolidation and when to use compression, which saves you from rediscovering these lessons the hard way. Treat it as a reference when you're past the prototype stage and need production reliability.