Why not caveman:compress auto compress ?

[pauldx]

we understand : Compression uses LLM call itself → costs tokens to save tokens. Only worth it if file changed since last compress.
but its inconvenient to run compress after every claude.md and memory file changes.

Realistic fix would be: Hook on UserPromptSubmit that checks if CLAUDE.md modified since last compress → if yes, compress → then proceed. Checksum comparison, no LLM needed for detection part.

Would that help to avoid on demand compress ?

PR advice below : for automation