Most "bad AI answers" aren't a model problem. They're a prompt problem.
You type something like:
fix the auth bug
…and the model fires back three clarifying questions, or confidently fixes the wrong thing. So you rephrase, retry, and lose ten minutes per task. The gap between what you mean and what the model can act on is structure.
Anthropic's own prompt-engineering docs are clear about what good structure looks like: give the model a role, context, the task, and an explicit output format — ideally with XML-style tags. The problem is doing that by hand, every time, in every app. Nobody does.
So I built a hotkey for it
It's a small desktop app called Prompt Enhancer. The flow is:
- Select any rough draft in any app (editor, chat box, browser field)
- Press a hotkey — Ctrl+Alt+P (or Control-Option-Cmd-P on macOS)
- Your text is replaced in place with a structured prompt
No copy-paste, no second tab, a few seconds.
Before / after
Before:
help me with a python parser for nested json
After:
<role>experienced Python developer</role>
<task>parse a nested JSON structure</task>
<instructions>
- use json + jsonpath-ng
- handle missing keys gracefully
- include example usage
</instructions>
<output_format>working code + 3 examples</output_format>
Same intent in, far more useful answer out.
How it works under the hood
- Calls Claude Haiku to do the restructuring (fast + cheap)
- Bring your own API key — your text goes straight to Anthropic, not through anyone's server. The key lives in your OS keychain.
- macOS build is native Swift; Windows is Tauri/Rust
- Works in any text field (Accessibility API on macOS, UI Automation on Windows, clipboard fallback for Electron/web apps)
- Presets for different styles: Default, Concise, Verbose, and Code
It's free to try (a handful of demo calls on a shared key), then free forever with your own key.
Try it / tell me it's wrong
Link: https://promptenhancer.online/
I'd genuinely like feedback on two things: (1) whether the restructured output actually helps in your workflow or gets in the way, and (2) which presets you'd want that don't exist yet. If you try it, drop a comment — happy to iterate.
Top comments (1)
A hotkey that restructures a lazy prompt into a structured one is a clever bit of UX, because the real problem isn't that people can't write good prompts, it's that good structure is tedious and they won't do it every time when a one-liner is right there. Removing the friction (auto-add the role, constraints, output format, the "think before answering" scaffold) is the right intervention - it makes the good behavior the default instead of the disciplined exception. The deeper truth it's leaning on: structure is most of what separates a useful answer from a generic one, so a tool that injects structure for free captures most of the prompt-engineering value with none of the effort.
The thing I'd watch: a generic structure template helps, but the biggest wins come from project/context-specific structure (your conventions, your constraints), which a universal hotkey can't know. That's the gap between prompt-templating and real context engineering - and it's exactly where I focus in Moonshift, the thing I work on (a multi-agent pipeline that takes a prompt to a deployed SaaS), where the structure fed to the model is project-aware, not a generic wrapper. Your hotkey is a great low-friction on-ramp to that idea. Multi-model routing keeps a build ~$3 flat, first run free no card. Nice tool. Is the structure a fixed template, or does it adapt to the task type? Task-adaptive scaffolding is where this gets really powerful.