What problem does it solve? Engineering tasks handled ad hoc by an AI agent often skip reproduction, verification, design exploration, and review, producing unverified code and sloppy prose. This Skill imposes a disciplined operating mode that matches each task to a playbook, applies named engineering principles, and requires runtime evidence before declaring work done. ## Core Features & Use Cases - Playbook routing: Matches tasks to over twenty playbooks covering bug fixes, features, refactoring, perf, investigations, PR babysitting, shipping, autonomous runs, and multi-phase plans, copying steps into a todo list. - Principle-driven decisions: Applies a catalog of core, architecture, verification, delegation, and meta principles (laziness protocol, model the domain, prove it works) and cites each one that shaped a decision. - Subagent orchestration: Delegates code to typed subagents with per-role model assignments from references/models.md, plus swarm verification, adversarial interrogate reviews, and arena bakeoffs. - Use Case: Ask the agent to fix a reported defect. It reproduces the bug on the real surface, binary-searches the root cause, delegates the fix, verifies the repro passes, stages the failing test before the fix, and opens a PR. ## Quick Start Ask the agent to enter poteto mode and fix the login timeout bug, verifying the reproduction passes before opening a PR.
