prototypelisted

# Prototype Build the first working version of something real — fast, but as a base worth keeping. Speed comes from judicious choices (libraries, patterns, architecture that fit and last), not from cutting corners. The thin slice you ship is a sound foundation to extend, not a hack to tear down. **Task:** "$ARGUMENTS" ## Working principle Deliberate about what lasts, ruthless about scope. Spend judgment on the choices that are expensive to reverse — stack, data model, the seams between parts — and defer everything that isn't the core: features, edge cases, polish. Minimal tech debt by construction, not by later cleanup. Lean on parallel subagents for exploration and for separable building, so the main thread stays on decisions and its context stays clean. Scale the deliberateness to the task. A one-file script with an obvious stack goes near-straight to building; a feature inside an unfamiliar codebase earns the full first two phases. Don't ceremony-up the trivial; don't wing the consequential. ## Phase 1: Understand Gather the real information the Phase 2 decisions will rest on. Exploration and research run as parallel subagents — concurrent, and their findings return synthesized so the main context isn't flooded with raw output. ### 1.1 Frame - The **real outcome** wanted, in a sentence — the result, not the literal feature. Open-ended task ("here's our situation, build something useful")? Then *what to build* is the first and highest-leverage decision; spend real