What problem does it solve? Multi-step engineering work with AI agents often drifts: unverified claims of done, bloated diffs, blocking questions to the human, and prose full of filler. This Skill enforces a disciplined operating style where every task routes to a named playbook, every change is verified against the real artifact, and every reply follows strict writing rules. ## Core Features & Use Cases - Playbook routing: Matches incoming tasks (bug fix, feature, investigation, perf issue, PR babysitting, shipping, autonomous runs, orchestration) to step-by-step playbooks under playbooks/ and executes their steps verbatim. - Principles system: Applies named engineering principles (model the domain, prove it works, fix root causes, sequence verifiable units, guard the context window) with mandatory citation of which principle drove each decision. - Subagent delegation with verification: Spawns background subagents with per-role model assignments, reviews their diffs directly, and requires runtime evidence before declaring work done. - Use Case: Ask the agent to fix a reported bug. It reproduces the bug on the real surface, binary-searches the root cause with runtime evidence, delegates the fix to a subagent, verifies the original repro now passes, and opens a PR with the failing repro committed before the fix. ## Quick Start Invoke poteto-mode on your next task by asking the agent to work in poteto mode, for example to investigate and fix a bug with full runtime verification.
