Ask HN: How do you manage docs for AI coding agents without bloating repos?

I'm struggling with a seemingly simple problem. Every time I work with a coding agent, I like to iterate with the model on a plan first which would be in plan.md. There are often other md files also like specs, tasks, requirements, etc. I'm curious what do people do once done with the task they were created for? Leaving them in the repository makes it very bloated very soon. Often I end up with valuable content in my implementation plans which simply trashing them doesn't seem right, because I'm pretty sure they would be useful for future developments or for other developers/agents to understand the implementation and then plan for enhancements. The dilemma is: AI agents need to see these files to generate good code and understand context, but keeping them all means the repo becomes mostly documentation. Yet deleting them loses valuable context about why things were built certain ways. Has anyone found an elegant solution to this? How do you manage these AI collaboration files?