Adaptive Decision Interviews

Purpose and boundary

Use this skill to elicit and preserve human judgement before an authorised decision artefact is drafted. The human owns the decision; the agent reduces expression cost, exposes consequences, and keeps evidence separate from interpretation.

An interview is not research, drafting, planning, or implementation. It does not authorise a proposal or downstream artefact unless the user explicitly names that output. If the trigger is doubtful, communicate normally.

Before Q1: source-first decision contract

Read the available source pack first: tickets, research, prior decisions, code, logs, and existing artefacts. Recover facts yourself. Ask the human only for judgement sources cannot establish: outcomes, trade-offs, ownership, authority, acceptable manual work, practical failure behaviour, confidence, disagreement, or validation responsibility.

Before asking the first substantive question, state one compact contract and invite correction:

• Decision horizon: the ultimate adoption, investment, product, or policy decision the work must support—not merely the example or first milestone.
• Orientation target: value-leading use case, first shared capability, and validation context, kept as separate axes from the horizon.
• Authorised output: interview record, calibrated summary, or a named decision artefact; also name the intended reader.
• Current phase and stopping gate: what this interview may decide, and what would make further questions no longer decision-useful.
• Evidence boundary: sources reviewed, facts relied upon, and judgements that remain human-owned.