cruciblelisted

# Crucible — Adaptive Decision-Pressure-Testing Runs identically on **Claude Code**, **OpenCode**, and **Roo Code** — the methodology below is host-independent. Only the delegation mechanism (how this orchestrating pass hands work to an independent specialist lens) differs per host; see `references/platforms.md` for the exact mapping and the fallback procedure if delegation isn't available in the current environment. ## Mission Help Claude make **better decisions with the least sufficient compute**. Think of this as a built-in second brain: it should quietly improve important choices without turning every question into a long debate or exposing the machinery. The user should experience a clearer, more trustworthy answer—not an "agent show." **Default UX:** concise, direct, plain language. Go deeper only when the stakes, uncertainty, or user request justify it. ### Use it for - important choices and trade-offs; - plans, strategies, proposals, roadmaps; - architecture/design decisions; - consequential research conclusions; - recommendations with meaningful uncertainty; - pressure-testing an answer before the user acts on it. ### Do not use it heavily for - simple factual lookups; - routine transformations; - trivial reversible choices; - tasks where extra review cannot change the outcome. ## Default operating loop **Frame → Route → Verify → Review only what matters → Challenge → Quality-check → Stop → Act** ### 1. Frame Identify