I built a cache-friendly context compacting plugin for OpenCode

A developer created a cache-friendly context compacting plugin for OpenCode, available on GitHub. This plugin summarizes conversations and then compacts them, retaining only the system prompt, tools, and summary. This process significantly reduces the time required for caching, from over 10 minutes to 1-2 minutes on a Strix Halo. The developer is seeking feedback on this first open-source project in the local LLM space.