Prompt Inspector
See what your prompt is made of. A grade that points at the next edit. Analyzed in memory. Never stored.
- Persona4/5
- Task4/5
- Context4/5
- Format0/5
- Examples0/5
- Constraints1/5
Add examples and this becomes B.
Suggestions 3 available
- Examples+5 pts
Add one sample agenda block
A concrete sample shows the level of detail you want in the agenda.
Example: Day 1, 9:00 to 10:00, team introductions and goals.
- Format+5 pts
Specify the agenda structure
A clear structure makes the output easier to use and review.
Format: organize the agenda by day and include time blocks, activity names, and a short purpose for each item.
- Constraints+4 pts
Set practical limits
Limits help the agenda fit your offsite goals and schedule.
Constraints: keep each day to [X] hours, avoid [topics], and include at least [N] bonding activities.
What the Inspector looks for
A complete prompt has six parts. The Inspector highlights every one it finds, and the ones it cannot find are your to-do list.
Persona
Who the AI should be, or who you are: the role, expertise, and point of view the answer should come from.
Role framing changes vocabulary, depth, and judgment. 'You are a security reviewer' reads code differently than 'you are a beginner tutor'.
Task
The action you want taken: the verb and the deliverable.
A clear task verb is the single highest-value sentence in a prompt. Without it the model guesses your intent.
Context
Background facts the model cannot know: the situation, the audience, the real-world details.
Context is what separates a generic answer from your answer. Models fill missing context with averages.
Format
The shape of the output: structure, length, medium, and tone of the deliverable.
Format instructions are nearly free accuracy. A table request returns a table, not an essay you reformat by hand.
Examples
Sample inputs and outputs that show, rather than tell, what good looks like.
Few-shot examples are the strongest steering signal short of fine-tuning. One good example beats three paragraphs of description.
Constraints
Hard rules the output must respect: what to avoid, limits, and must-includes.
Stated limits prevent the most common failure: an answer that is good in general but wrong for you.
How grading works
The Inspector scores the same six pillars it highlights. Each pillar earns 0 to 5 points, so the total is out of 30. The total maps to a letter band: an A+ starts at 28, and the bands step down from there to an F.
The grade is a direction, not a verdict. Every result names the cheapest points you left on the table and ships with a note like “Add a persona and this becomes an A-”. The point is never to scold a C prompt. It is to show you the one edit that makes it a B.
Privacy
Prompts are processed in memory and discarded as soon as the analysis returns. We do not store them, we do not log them, and we never train anything on them. There is no account, so nothing you paste can be tied back to you.
FAQ
Is my prompt stored?
No. Your prompt is analyzed in memory and discarded as soon as the result returns. We never write it to a database or a log file, and there is no account to attach it to.
Which model analyzes my prompt?
A configured OpenAI model, called server-side. Your text goes to the model for one analysis and nowhere else. We never train on it.
Why did my grade change between runs?
Language models vary a little at the margins, so a borderline pillar can land a point apart on two runs. We call the model at temperature 0 to keep that variance small, and the rubric is versioned so scores stay comparable over time.
What is the quick check?
An offline rules pass that runs without calling the model. It spots the obvious gaps instantly, and it keeps working when you are rate-limited or the model is unavailable.
Is it free?
Yes. No signup, no paywall. We rate-limit per person so it can stay that way.
Behind the scores
The rubric comes from The Prompt Lab, where we take real prompts apart and rebuild them, one pillar at a time.
Learn the craft behind the grade →