Has anyone actually run controlled A/B tests on Claude "skills" and prompt plugins? Or are we all just tweaking configs instead of shipping things?
Seriously asking. There's a growing ecosystem of prompt frameworks, "skill" injections, superpowers packs, Obsidian-flavored system prompts, and whatnot that all promise to make Claude significantly smarter, more structured, or more capabl…