I think it's a matter of implementing different aspects of outcomes into a project/codebase.

It's one thing to add skills and system prompts to a project; it's another to accurately frame and orchestrate the various parts of a context window to maximize the inputs in an efficient way to optimize the outputs of these agents/models we've all begun using.

Practically: explore and influence definitions of context to refine your ability to contribute to the documents that steer the agents.

Or better yet, focus on a part of the context window and figure out how to expose resources responsibly.

Personally, I think local models are the future...but that's just a hunch