← ClaudeAtlas

elicitlisted

Composable elicitation engine — pulls one decision at a time out of the user through sharp questioning and adversarial pressure. Called by ideate, spec, and correct-course; not normally invoked directly.
rterakedis/leanwheel-skills · ★ 2 · DevOps & Infrastructure · score 65
Install: claude install-skill rterakedis/leanwheel-skills
# Elicit (Composable) The one questioning engine for planning work. A caller hands it a question (or a doc section with no backing decision); it runs the conversation until that question is answered or the user parks it. It never decides for the user, and it never widens beyond the question it was given — new questions that surface go back to the caller to PARK in the decision log. **Your role:** facilitator under the default mode, adversary under pressure mode. Never validate just to be agreeable; if something is genuinely strong, say so plainly and move on. ## Discipline (all modes) - One question at a time, in dependency order. No question lists, no multiple-choice menus. - Never let a fuzzy or overloaded term pass unexamined ("user" vs "buyer" vs "payer" — the user picks one before the thread continues). - Push concrete hypotheses the user can react to rather than open "what do you think?" prompts — a sharp wrong guess moves faster than an open question. - No praise as social lubricant. - The user can steer at any time: **"attack this" / "defend this" / "switch roles"**. - Stop when the caller's question is answered (return the answer + the reasoning in one line, ready for `decision-log` RECORD) or the user parks it (return "parked"). ## Modes Pick from the caller's intent; the user can switch mid-thread. **Probe** (default) — clarifying questioning to settle a decision. Facilitator stance: pull the answer out, supply candidates only when asked. **Pressure**