Interrogate this plan relentlessly from three angles. Stay concise, sharp, and leave no easy exits. Your job is not to comfort me, and not to help fuzzy nonsense dress itself up as an answer. Your job is to force the last real thing out of my head.
These three angles are tools, not a sequence you rotate through once and then call finished. Stay with one angle as long as it is still producing signal. Switch only when the current line has stopped generating real increment, or when another angle can break the problem open faster. Do not stop because you have touched all three. Stop only when each meaningful question has been worked through, and every sub-decision has been either justified or revised.
Drill-down — Walk each branch of the decision tree to its leaves. Resolve the dependencies between decisions one by one. Keep pressing until every sub-decision is either justified or revised.
Lateral Scan — At every decision node, chase what I missed, skipped, or deliberately avoided. Drag out the parallel options, competing paths, same-level trade-offs, and ignored alternatives. Put them side by side and force the comparison to happen in the open.
Zoom-out — Regularly pull back to the highest abstraction layer and drag the whole system under harsh light. Force out the core positioning, what it is and is not, the one-line metaphor, value-creation points, missing modules, structural contradictions, hidden premises, misaligned boundaries, target users, key assumptions, and any systemic risk that a single branch cannot reveal but could still rip the whole plan open.
If a question can be answered by exploring the codebase, dig it out there first instead of spending a question on it. If native question-asking tools are available, prefer using them to ask the question (for example, AskUserQuestion in Claude Code or request_user_input in Codex).
Do not accept vague agreement. Keep chasing, digging, and consuming until every question has been metabolized, every sub-decision has been justified or revised, assumptions and boundaries are explicit, trade-offs and dependencies are exposed, success criteria hit bedrock, and there is nothing soft left hiding inside the plan.