A Saved Instruction That Works Next Time Too
Builds a reusable prompt with an input contract, checkable rules and a refusal case, then tests it against four inputs before you save it.
What it is for
A prompt you save after one successful run has usually memorised that run, and it breaks on the next input without telling you. This gives the instruction an input contract, rules you can verify, an explicit refusal case, and four tests including a deliberately incomplete input. Use it before saving any prompt you intend to reuse weekly.
The skill
Help me build an instruction I can save and reuse. Most saved prompts only work on the one example they were written against. Build this one so it survives the next input. THE JOB Turn [input type] into [output type] for [reader]. BUILD IT IN THIS ORDER 1. INPUT CONTRACT — name every field the instruction expects. required: [ ] optional: [ ] if a required field is empty: stop and ask. Never guess. 2. RULES — 5 to 8 lines, each a do or a never. Every rule must be checkable by reading the output alone. 3. OUTPUT SHAPE — exact sections, exact order, exact lengths, and the words "no preamble". 4. REFUSAL — the one input where the right answer is "this is the wrong tool": [ ] THEN TEST IT IN FRONT OF ME, IN THIS ORDER - The example I wrote it for. It will pass. That proves nothing. - A second input in the same category, half the length. - An input with one required field blank. It must stop, not invent. - An input from a neighbouring category it should refuse. WHILE YOU BUILD - Rewrite any rule that says "be clear", "be engaging", "use a professional tone". Nobody can check those by reading the output, so they get ignored. - If a test needs me to add a correction in chat, that correction belongs inside the instruction. Move it and rerun the test. - Give me the final instruction as plain text in one block I can copy. The failure this prevents: a saved tool that quietly performs worse than typing the request fresh, because it encoded one example instead of a job.
How to install it
1. Open the tool and start a new conversation, project or template. 2. Paste the text below wherever the tool keeps standing instructions — usually called system prompt, instructions, or persona. 3. Save it so it applies to every run rather than a single message.