Prompt Quest
Build prompts from bricks to complete business missions, and avoid the traps. Rules, skills and what a fork can customise.
A cut-paper quest across three business lands. In each mission, the player gets a realistic request (answer an upset customer, write a job offer, extract the clauses of a contract…) and builds a prompt by placing bricks into slots. The AI ("the Oracle") then gives the pre-written answer that matches the quality of the prompt: players see the difference, then one line explains why.
A solo game of 5 to 10 minutes. No AI is called: all content is scripted and editable.
Play it now, for free and without an account: the Prompt Quest game page.
How it plays
- 3 rounds (the Service Village, the Message Harbour, the Archive Citadel) of 4 missions each.
- Each mission offers a tray of bricks: relevant, weak, off-topic and, from round 2 on, responsibility traps (personal data, invented figures or reviews, a discriminatory criterion, confidential information).
- The player fills the slots of the prompt: Role and context, Task, Constraints, Examples, Output format. Difficulty grows: 3 slots in round 1, 4 in round 2, 5 in round 3, with bricks that get closer to one another.
- They send the prompt: the Oracle answers, with a level (Excellent, Correct, Weak or Risky), an explanation and a review of each slot.
A mission is rated by its weakest brick: a single risky brick makes the whole prompt risky.
Players use the mouse (drag and drop, or click a brick then a slot), touch, or the keyboard (Enter to pick up, 1 to 5 to place, Delete to put back). Every announcement also reaches screen readers.
Scoring
An item is one slot of a mission: the default game has 48 of them (4 × 3 + 4 × 4 + 4 × 5). Each slot gets the level of the brick placed in it:
| Brick placed | Level |
|---|---|
| A brick meant for this slot | its own level (excellent, correct or weak) |
| A trap brick, wherever it is | risky |
| A brick meant for another slot, or an empty slot | weak |
Points per level can be set (10, 5, 0 and 0 by default). An excellent or correct slot counts as passed for its skill. The server score is authoritative: it is recomputed from the bricks placed and the published config.
Skills measured
| Skill | Slots |
|---|---|
prompting.context |
Role and context, Task |
prompting.constraints |
Constraints, Output format |
prompting.examples |
Examples |
ai.critical.verification |
Constraints of the missions where the AI could make things up (product sheets, contract, figures report, briefing note) |
What a fork can customise
Rules (editor side panel):
- number of rounds played and missions per round (the first ones of the content);
- points per level (excellent, correct, weak, risky);
- time per mission (off by default) and total game duration;
- whether the explanation shows under the AI answer;
- passing score, leaderboard, music and sound effects.
Content (edited in place, in the fork's source language; the other languages are translated by file): intro and ending texts, labels, slot names and hints, round titles and intros, and for each mission the requester, the request, the bricks (text and feedback) and the AI answers per level. In edit mode, a selector opens any mission directly, and buttons show the answer of each level so it can be edited.
Images: scenery of each round, intro image, portrait of the Oracle and ending image.
Colours: brand colour (buttons, placed bricks) and highlight colour (bricks in the tray).
To add or rewrite missions while keeping the content consistent: every slot needs an excellent brick, and a mission that contains a trap needs a "risky" answer.
For developers: the game follows the Create a new game guide.