Paste a prompt and get instant, rule-based feedback on the things that quietly make prompts unreliable, vague wording, conflicting instructions, and missing structure. Runs in your browser.
This is a rule-based linter, not a model. It looks for the patterns that reliably make prompts weaker: vague quality words the model can’t optimize (“good”, “appropriate”), conflicting instructions (“detailed” and “concise”), asking for JSON without specifying the shape, prohibitions instead of positive instructions, and missing role, format, or audience. It runs entirely in your browser and stays out of your way, treat it as a checklist, not a grade from an oracle.
A high score is not a guarantee and a low score is not a failure, some great prompts break these rules on purpose. Use it to catch the unintentional ones. Nothing you paste leaves your browser.
No. Every check is a local rule (pattern matching and simple heuristics), so it is instant, free, and private. The deeper, model-based version is a separate idea for later.
It starts at 100 and subtracts points per finding, more for issues than for tips. It is a rough signal to draw your eye, not a precise measurement.
No. Linting runs entirely in your browser, so the prompt stays on your device and it works offline.
Pin down output with structured outputs and the Structured Output Builder, or browse AI Explained.