pre-goallisted

# Pre-Goal: Turn a Vague Task into a Long-Running Goal Prompt ## Overview A goal run succeeds or fails on how precisely "done" (the Definition of Done) is defined. A vague definition forces the AI to guess what "good" means → it wraps up in three minutes with output the user never wanted. This skill guides the user to nail down five elements before launching `/goal`. ## The Iron Law **Never skip the interview and generate the goal prompt for the user.** The task the user brings is necessarily vague ("make it better", "make it prettier") — that is exactly why they came to you. Filling in the details yourself = you defining their Definition of Done = output they will not be satisfied with. **No exceptions:** - Do not skip because "the task looks simple" - Do not skip because "there is enough information in context" - Do not write a full prompt first and then ask "want any changes?" — the order is confirm first, assemble second ## Process 1. **Gather clues**: Read the user's task description. Mine the codebase and conversation history for usable facts (existing test commands, performance tools, directory layout) to use as proposal material. 2. **Propose + confirm, element by element**: For each of the five elements below, draft 1–3 **concrete proposals** based on the clues, then use AskUserQuestion to let the user pick or revise. Never throw open-ended questions at a blank page — you propose, the user corrects. 3. **Branch for subjective tasks**: If the Outcome is a matt