Launched a searchable library of prompts that survived review

Close

Only prompts that produced code someone actually merged.

I have collected prompts for two years and thrown most of them away. The ones left are tagged by task, model, and what they get wrong. Every entry names the failure mode, because a prompt without its failure mode is just a nice sentence.

Claude Code built the search and the tagging. The part I did not expect: half my saved prompts stopped working when models changed, so the library now records which model version a prompt was tested on.