Alvaro-Manzo/promptshift

PromptShift is a thoughtful and genuinely useful idea: a lightweight prompt adapter focused on preserving user intent instead of inflating prompts into over-engineered instructions. The project has a clear point of view, and that is its biggest strength. I like that it explicitly pushes back against common “prompt optimizer” behavior where the tool adds roles, requirements, audiences, and constraints the user never asked for. The README explains the philosophy well, and the SKILL.md gives enough structure that someone could realistically use it as a Claude Skill without needing a full software package around it.

The strongest part of the repo is its restraint. The principles of “clarity first,” “minimal change,” and “model-aware, not model-dependent” are exactly the right framing for this kind of tool. The examples and benchmark categories also show that the author is thinking beyond a single use case and wants PromptShift to work across writing, coding, reasoning, UI, and RAG-style prompts.

That said, the repo would be stronger with tighter evidence and consistency. The benchmark section currently feels more like a benchmark plan than a completed benchmark. Adding actual scored results, model versions, dates, before/after outputs, and evaluation criteria would make the project far more credible. Some examples also seem to violate the project’s own “do not invent requirements” rule by adding roles or extra requested outputs. That should be cleaned up, because intent preservation is the core promise.

I’d also recommend fixing small polish issues: broken contribution filename references, empty result files, inconsistent output formats like SCORE vs CONFIDENCE, and visible encoding problems in Markdown. These are not fatal, but they make the repo feel less production-ready than the idea deserves.

Overall, PromptShift is a strong early-stage concept with a clear identity. With reproducible benchmarks, cleaner examples, and stricter adherence to its own rules, it could become a very credible prompt-quality skill rather than just a well-written prompt philosophy.