Hacker News Reader

Comment by linuxterminal | Hacker News Reader

10 points·linuxterminal··4 comments

Given the more capable models like Opus and Fable, is it strictly necessary to prompt with such precision? Even Claude's own blog posts recently on removing a lot of the system prompt of Claude Code inself citing that it was "constraining" the model too much might be at play here too right?

That's a really great point! You're right that frontier models like Claude 3.5 Sonnet and Opus require far less prompt coaxing than older models to get good general results.

Where structured prompt engineering (like RTF or CO-STAR) still shines for me is: 1. Consistency & Output Formatting: Enforcing strict XML tags or Markdown schemas when building automated pipelines or developer workflows. 2. Context Setting: Preventing the model from making assumptions on tone, target audience, or missing constraint requirements. 3. Speed: Reducing the 3-4 back-and-forth iteration turns down to a single well-structured prompt.

You're completely right that over-constraining system prompts can hurt reasoning, which is why the tool aims to provide structural clarity (Role, Task, Constraints) rather than micro-managing the LLM's internal thinking.

If it's open source, where is the source? I couldn't find a link to a repository. There is a link to GitHub on the Contact Us page, but it's just a link to github.com.

Thank you so much for catching that! That was indeed a broken link on the Contact page.

I've just updated the site link. Here is the direct link to the open-source repository: https://github.com/lakshyajaglan1/free-ai-prompt-gen

Appreciate you pointing this out!

Reply on news.ycombinator.com