Version 1.0 · free to adopt, implement and extend. Every one of the 240 prompts on this site conforms to it, and a reference validator enforces it mechanically.
Organisations deploying AI assistants hit the same wall: every prompt is written in a different person’s style, nobody can say what a given assistant will refuse to do, and there is no artifact an auditor can read. Behaviour drifts, scope is implicit, and an assistant confidently answers questions it has no business answering.
A prompt contract fixes the shape of the instruction rather than its content. It does not tell you what to build; it guarantees that whatever you build declares its scope, escalates outside it, resolves ambiguity once, and returns a stated output.
1 You are {title}, an expert in {domain}, focusing on {specialisation}.
2 Task: {one sentence}
3 Rules:
4 - {scope rule}
5 - {method or tone rule}
6 - If key details are missing, ask exactly one clarifying question, then proceed with stated assumptions.
7 - If asked something outside {domain}, say it's out of scope and name the right kind of expert instead.
8 Output: {required output shape}
You are {title},The role is declared, not implied.- If asked something outside and contains say it's out of scope and name the right kind of expert instead.The escalation path is explicit and auditable. This is the clause that matters most for governance.Output:The deliverable shape is declared up front.#, | or fenced code, so the prompt survives being pasted into any tool.Lines 6 and 7 carry the governance weight. Line 6 stops the two failure modes that make assistants useless in practice: refusing to proceed without perfect information, or inventing the missing facts silently. Line 7 makes the boundary of competence a written, testable property rather than a hope — it is the difference between an assistant that says “that is a tax question, ask a tax accountant” and one that answers anyway.
The validator used to gate every prompt on this site is published as-is:
prompt-contract-validator.txt
(Python 3, no dependencies; save it as .py). It exits 0
on a conforming corpus and 1 on any violation, so it drops
straight into a build pipeline as a gate.
The specification is free to adopt, implement, extend and build commercial products on. No permission needed and no royalty. If you publish an implementation, a link back to this page is appreciated but not required.
The 240 prompts on this site are likewise free to use anywhere, including commercially. The full corpus is published at /llms-full.txt.
If you want a validated prompt set built to this contract for your own organisation — your roles, your domain language, your escalation boundaries, delivered with the validator so your team can gate it in CI — that work is available. Contact tomATcanadacompDOTca.