/** * The interview's wording, for an agent. * * `@trazum/core` knows a question exists; the CLI and the web app each know how * to ask it in a locale. This server has no locale — an agent is not reading a * report, it is being asked something — so the wording lives here, in English, * and is held to the catalogue in both directions by `writer.test.js`. * * A question with no wording is not a cosmetic gap: an agent handed a bare id * would either invent the question or skip it, and a question nobody asks is a * slot nobody fills. */ export const SLOT_QUESTIONS: Readonly> = { task: 'What should the model do, in one sentence?', role: 'Who is the model being while it does that?', inputs: 'What changes from one call to the next?', 'output-shape': 'What should come back: prose, json, list or table?', 'output-schema': 'Which fields or columns, and which are always present?', 'output-length': 'How long should the answer be, at most?', audience: 'Who reads the output?', constraints: 'What must it never do?', refusal: 'What should it do when it cannot answer?', examples: 'Is there an example of a good answer?', 'example-inputs': 'What input produced that example?', 'failure-modes': 'What has gone wrong with this before?', model: 'Which model is this for? This changes the estimate, never the prompt.', budget: 'What is the monthly ceiling for this prompt? This changes the estimate, never the prompt.', };