fast-dev-compaction

What it does

It ports tamaratran's fast-jev-compaction approach to Codex lifecycle hooks.

How you can use it

Read this experiment to compare two ways of keeping details from a long AI chat. One uses a summary alone. The other adds back some old messages. Compare whether the coding assistant can still find the file names and errors it needs.

This is an idea to study, not a tool to use for real work. The author recommends the normal summary instead. When old chat history changes, the AI may need to read it again, costing you more.