abundant-ai

Drive a build-ready, vertically-sliced implementation spec all the way to shipped, tested code through an adversarial test-first loop — codex fresh-eyes simplification of the plan, a codex bug bake-off, then per-slice: subagent-per-task implementation (simplest possible code, long descriptive names, zero comments until green), codex simplify/refactor, a codex+claude bug-hunt competition, and a concise flow doc that flags and fixes suspicious parts. Invoke as `/vertical-slice-build <path-to-spec>` when the user has an already-designed spec split into ordered vertical slices (S1, S2, …) and wants it BUILT slice by slice with maximum review pressure. Not for greenfield design (there is no spec yet), one-off edits, or reviewing an existing diff.