Reference
What it is great at, and where it stops
Here is the balanced view, including the places it falls short. Knowing the edges is what lets you lean on a tool with confidence.
It is great at
- •Turning vague requests into specific, well aimed prompts.
- •Turning a vague request into a specific one, without dropping anything you said or adding anything you did not.
- •Front loading the deliverable so you get usable output instead of caveats.
- •Doing all of it invisibly, in plain prose, in a couple of seconds, and teaching you good prompting over time.
Where it stops
- •It writes a better prompt. It does not run the task for you. The model you send it to still does the work, and the quality of that final answer depends on that model too.
- •On a bare lookup it sharpens the wording and stays out of the way, and when your request already looks complete it says so and offers to send it as it is. That is the design, not a shortcoming.
- •The library, while broad, is finite. If your domain is not represented yet, the generic path still gives a cleaner prompt. This is exactly where the community comes in.
- •It is opinionated about invisibility. It will not name a technique or explain the theory, by design. If you want the theory, these docs and the evidence page are where to find it.