Prompt Doctor
Paste a prompt you already wrote. The Prompt Doctor scores it against the prompting practices that matter for Claude, flags exactly what's weak, links each finding to the page that fixes it, and hands you an improved scaffold to copy.
It runs 100% in your browser — your prompt is never sent anywhere. The checks are plain heuristics, not an AI call: fast, free, and transparent.
Heuristics, not an oracle — they catch the common failure modes, not every nuance. Use the linked pages to go deeper, and always test with a real eval.
What it checks
Each rule maps to a deeper page on this site:
| Check | Why it matters |
|---|---|
| Enough to act on | One-line prompts force Claude to guess your intent. |
| Sets a role | "You are a senior X" sharply focuses tone and judgment — see Roles. |
| States one clear task | A leading imperative verb ("Summarize…", "Extract…") removes ambiguity. |
| Specifies output shape | Without a format you get whatever length and structure Claude picks — see Structured Output. |
| Shows an example | For pattern tasks, one input→output pair beats paragraphs of description — see Few-Shot. |
| Structured when long | Long prompts need XML tags so Claude can tell instructions from content. |
| Guards against made-up answers | When you supply source text, tell Claude to use only it — the core defense against hallucination. |
| Avoids vague qualifiers | "good", "engaging", "professional" aren't measurable. Replace them with testable criteria. |
:::tip From score to ship A high score isn't the goal — a reliable result is. Treat the scaffold as a starting point, then prove it with a real eval before you depend on the output. :::
Next
- Prompting Basics — the five moves that fix most prompts
- Prompting Claude — what's specific to Claude
- The Pattern Library — copy-paste prompt patterns
- Prompt Builder — build a structured prompt from scratch