Zandereins/schliff ? reverse-engineered prompt

Build me a small Python tool called Schliff that scores AI instruction files like AGENTS.md, SKILL.md, CLAUDE.md, and .cursorrules in a deterministic way.

I want it to give the same score every time for the same file, with a clear 0 to 100 result and a simple breakdown of what is good or weak, like structure, coverage, efficiency, composability, and clarity. It should also spot obvious attempts to game the score, like padding, junk text, or keyword stuffing, and keep the core version free of extra dependencies.

Please include a command line interface that can score a file, run on a demo example, and be usable in CI or pre commit checks. If it makes sense, also make it work nicely inside Claude Code with a few commands. I’d like the project to feel polished, with good docs, tests or benchmarks, and a clean install experience. If you need to check current docs for any integration details, look them up online first.