The context drift problem you mention is underappreciated. In production agent systems, I've seen prompts that work perfectly in isolation gradually degrade across long conversations—not because the model is failing, but because earlier turns implicitly narrow the solution space in ways that compound.
What's interesting about MCP standardization is that it forces teams to think about context as infrastructure rather than an afterthought. Most shops were rebuilding the same plumbing anyway: connection management, capability discovery, session state. Making it a shared layer means context engineering can focus on the actual hard problem: maintaining coherence and intent alignment over time.
The Skills-as-context-packages framing is sharp. It mirrors what we're seeing in production: teams struggling with agents are usually the ones treating prompts as configuration files. The ones succeeding treat them as behavioral contracts with versioning, rollback semantics, and integration tests.