Reverse-engineer the prompt
Look at a real screen and reconstruct the spec that produced it — graded instantly by AI.
- Train the core product-engineering skill: seeing a UI and articulating the requirements behind it
- Instant AI grading across all six spec dimensions, with a written rationale
- Every submission is saved to your activity record
From screen to spec
Every screen you use was once a spec. Reverse-engineer flips the usual direction: study a screen in one of the real apps under test, then reconstruct the spec that would produce it.
The six dimensions
Your submission has six required sections:
1. One-line prompt — the single "Build a…" sentence behind the feature
2. Personas & jobs-to-be-done — who uses it, and what job it does for them
3. Requirements — functional and non-functional, as implied by the UI
4. Data model — the entities, fields and relationships the screens imply
5. Edge cases — the boundary and error situations the builder had to handle
6. Acceptance criteria — testable statements that verify it works
You can attach a screenshot of the screen you studied (optional, but useful evidence).
Instant AI grading
On submit, AI grades all six dimensions against a 100-point rubric — completeness, plausibility, insight — and returns your score and a rationale immediately. The submission is saved to your activity record; then pick another screen and go again.
How to score well
- Look past the pixels: every dropdown implies a data model, every error message implies an edge case.
- Make requirements testable — "loads fast" is weaker than "list renders under 200ms for 1,000 rows".
- For the data model, name entities and relationships explicitly ("an Order has many LineItems").
- Acceptance criteria should read like a checklist a QA engineer could execute.
The journey, step by step
| Step | What you do | What you get |
|---|---|---|
| Pick a screen | Choose one of the apps under test and study a screen in it | A real, live feature to deconstruct — optionally attach a screenshot |
| Write the spec | Fill all six sections, from one-line prompt to acceptance criteria | A complete reconstruction of the spec that would produce this feature |
| Submit | Submit for AI grading | A /100 score and rationale, instantly |
| Iterate | Read the rationale, pick another screen, go again | Each rep sharpens how you see requirements hiding in interfaces |
Frequently asked
What exactly am I reconstructing?
The spec behind a screen: the one-line prompt, who it's for and what job it does, the requirements the UI implies, the data model behind it, the edge cases handled, and acceptance criteria to verify it.
How is it graded?
AI grades all six dimensions against a 100-point rubric — completeness, plausibility and insight — and returns a score plus a written rationale immediately.
Do I need a screenshot?
No, it's optional — but attaching the screen you studied helps ground your submission and any later review.
Why does this skill matter?
Articulating precise requirements from a working product is the heart of product engineering — and of directing AI tools well. If you can spec what you see, you can spec what you want built.
Related guides
QA real, live products — then study their architecture and code once you unlock the advanced tabs.
ReadStructured defect reports on the apps under test: reviewed by mentors, valid bugs earn XP.
ReadHow your prompts are scored and improved.
ReadInvestigate a simulated production incident, resolve it, and get graded like an SRE.
Read