LUCID pressure-tests your prompt before it goes to the AI and reviews the answer after it comes back, against the professional Standard you define. It runs inside Claude, ChatGPT, Gemini and Grok, as a Chrome extension.
A model takes a prompt at face value. It fills the gaps silently, answers with confidence it has not earned, and pads the reply until the judgment is buried. In casual use that costs nothing. In a clinic, a courtroom, a boardroom or a lab, it is how a wrong premise becomes a decision.
LUCID stands between you and the send button. It reads what you stated, what you implied and what you left unstated, and it puts one hard question in front of you before the prompt leaves. When the answer returns, it reads that too, and flags hedging, flattery and padding, and any line that breaks the Standard you set.
LUCID does not write for you and does not answer for you. It makes you the better thinker in the exchange, and it keeps the AI honest to your profession.
You write your prompt as usual. Before it goes, LUCID maps what you stated, what you implied and what you assumed, scores how much the AI would have to guess, and asks the one question that challenges all three. You send as is, revise, or skip.
When the answer arrives, LUCID reads it against your Standard. Hedging that hides judgment, padding that displaces the answer, false confidence, an assumption inserted without disclosure: each is named inline, under the reply, with the line of your Standard it broke.
Five questions form your professional Standard once: what LUCID protects, which AI mistakes are dangerous in your work, what makes an answer usable, what must never cross into your professional reality, and what you use AI for. Every test and every review is calibrated to it.
How much the AI must assume to answer your prompt. CLEAR assumes little or nothing. PARTIAL fills some gaps. VAGUE means the AI invents the missing frame.
Maps the structure of your prompt: what you said, what you meant, what you assumed, and the question that challenges all three.
Scans the AI's reply for hedging, sycophancy and Standard violations. Flags appear inline below the response, each tied to the rule it broke.
The synthesized voice LUCID uses to calibrate every Pressure Test. View it, edit it line by line, or form it again.
When on, LUCID covers a failed answer with the Guardian's verdict and the fixes it needs, and offers to ask the AI to revise it.
Rules LUCID adds to the front of every prompt. Write them once, and the AI honors them every turn.
Keeps the Skip Next chip armed across sends until you click it off, for the stretches when you want the AI unguarded.
Off means off: LUCID stops capturing prompts and the Pressure Test bar does not appear. The sidebar stays open for reading earlier analyses.
LUCID asks five questions and writes your Standard from the answers. It is the document every Pressure Test and every Guardian review is measured against, and it is yours to edit. Pro users see which line of their Standard produced each Pressure Test.
