A local tool that checks AI drafts against your editorial voice
AI text is easy to demo and hard to judge. writing-eval flips that: you build a reusable style profile from approved prose, then check any Markdown draft against it for clarity, readability, rhythm, and vocabulary, all on-device with deterministic results and no hosted models. Reports surface current versus target values plus editing instructions, so it works as a regression gate, not just a vibe check. There's even an agent skill so LLM tooling can run the checks itself.