Show HN: A simple ChatGPT prompt builder

Original Article ↗ Hacker News Discussion ↗

A lightweight “prompt builder” for ChatGPT draws mixed reactions: some find it handy for structuring instructions and suggest richer templates, keybindings, and better handling of formats like JSON or XML, while others argue it adds little beyond what ChatGPT can already generate for itself. The conversation broadens into whether “prompt engineering” is a real discipline or just good communication, how much detailed prompting still matters as models improve, and the need for tools that can empirically compare prompt variants rather than just assemble text blocks.

Overall reception

  • Several commenters find the prompt builder useful and “cool,” getting satisfactory or amusing results.
  • Others see it as just a text-block form that isn’t specific to ChatGPT and question whether it adds value over asking an LLM to write prompts itself.
  • There is mild skepticism about attaching it specifically to “ChatGPT” given it works with any LLM.

Bugs, wording, and technical details

  • Early on, users report a broken “please” field and typos (“Propmpt,” “input the final result” instead of “output”), which are quickly fixed.
  • One commenter notes surprisingly large page size for such a simple app (≈1.4MB).

Use cases, audience, and usefulness

  • Some argue the target audience is unclear: beginners may not understand the meaning or effect of vague selectors like “you will Detect.”
  • Others think it’s a good framework for people who know a bit about prompting and want structured guidance.
  • A recurring criticism: for many users, it may be easier to ask ChatGPT to generate or refine prompts directly.

Prompt engineering, templates, and best practices

  • Multiple people request more templates, especially for:
    • System-style pre-instructions.
    • Post-processing constraints and output formats (JSON, CSV, lists).
    • Jailbreak-resistant prompts.
  • Discussion around JSON formatting:
    • Some use very strict, “micromanaging” wording to enforce JSON.
    • Others point to function calling and JSON-mode APIs as cleaner solutions.
  • Broader debate on “prompt engineering”:
    • Some say complex “act as…” patterns are less necessary with newer models.
    • Others claim well-crafted system prompts and identities still help with issues like laziness or partial answers.

Evaluation and experimentation

  • Multiple commenters want tools to:
    • A/B test prompt variations and models on fixed input sets.
    • Systematically evaluate responses (e.g., via LLM-as-judge or string matching).
  • It’s noted that measuring “better” prompts is hard due to open-ended inputs and subjective outputs; results are often highly stochastic.

Feature requests & updates

  • Requests include keyboard navigation (tab/ctrl-enter), more examples per template, and clearer role/options.
  • A later update (v0.2) adds user-defined templates, variable menus, example presets, localStorage persistence, export/import, and in-place editing.