About this place

Most AI advice tells you what a good prompt looks like in general. Nobody tells you what's wrong with yours. That's the entire product: paste a prompt, get a graded report — a score, the specific defects, and a rewrite that fixes them.

The name is a provocation, but the grading isn't. Behind the red pen is a written rubric — seven dimensions, anchored scores, weights that shift by task type — and the total is computed in code, not improvised by the model. Good prompts score well. A genuinely excellent one gets told so, without a rewrite, because it doesn't need one. You can read the entire rubric — it's rendered from the same file the grader uses.

The analysis runs on Google's Gemini 3.5 Flash. The roast voice is deliberate: blunt about the work, never about you, and always carrying the fix. If it's ever mean without being useful, that's a bug.

Use it from your editor

There's an MCP server, so agents like Claude Code and Cursor can grade prompts without leaving the terminal — useful when you're iterating on a system prompt and want a second opinion on every draft.

  1. Create an account and sign in.
  2. Grab your session token from your browser's cookies for this site (the better-auth.session_token cookie, the part before the dot).
  3. Add the server:
    claude mcp add --transport http yourpromptsucks \
  https://yourpromptsucks.com/api/mcp/mcp \
  --header "Authorization: Bearer <your-session-token>"
  4. Ask your agent to roast_prompt anything. Reports land in your history like any other roast.

Session tokens expire with your session, and editor use shares the same 25-roasts-a-day limit as the site.

The straight answers

Is the score objective?
No automated grade of writing is. It's consistent — anchored rubric, computed total, same standards for everyone — which is the honest ceiling for this kind of tool. Treat the dimension notes and the rewrite as the product; the number is the scoreboard.
Who runs this?
A small team that got tired of watching good ideas die in bad prompts. No testimonials wall, no "trusted by 10,000 teams" — the site is new and we'd rather earn that than fake it.
What does it cost?
Nothing. Five roasts a day anonymous, twenty-five signed in. If that changes, grading a prompt stays free.