Constraint prompt builder
Most prompt failures are really missing constraints: the model did something you never said it could not, or skipped something you assumed was obvious. This builder makes constraints explicit and organised. You separate hard rules the model must obey from soft preferences it should honour, and positive requirements from prohibitions — then the tool assembles them into a clean block your model will actually follow.
How it works
You start with a base instruction describing the task. Then you fill four buckets: must do, must not do, prefer, and avoid. Each line you add becomes a labelled item in the assembled prompt. Hard constraints (must do / must not do) are presented as non-negotiable rules; soft constraints (prefer / avoid) are framed as priorities to honour when they do not conflict with the task.
You can pick a target model, which adjusts only the wrapper formatting — XML-style tags for Claude, numbered sections for GPT-4 and Gemini — because each follows those conventions most reliably. The constraint content itself is identical. The final prompt assembles instantly in your browser and copies with one click.
Tips and notes
Be specific and testable with every constraint: “must not exceed 100 words” beats “keep it short,” and “must cite a source for each claim” beats “be accurate.” Reserve hard constraints for things that would make the output unusable if violated; overloading the hard list dilutes the ones that truly matter. Phrase prohibitions as concrete behaviours to avoid, not abstract qualities. Finally, test the assembled prompt and prune: if a constraint never fires or the model ignores it, either it is redundant or it needs to be promoted from soft to hard and made more explicit.