What problem does it solve? When an AI system maintains a registry of its own defects (cantiere-difetti.json), fixing them one by one produces unreadable pull requests, regressions, and fixes that pass shallow checks while the underlying cause survives. This Skill provides the operating manual for repairing those defects in batches grouped by root cause, with mandatory behavioral tests, mutation checks, and an exit gate that blocks delivery until everything is green. ## Core Features & Use Cases - Root-cause batching: Groups open defects by how they broke (not by count), prioritizing diseases that block revenue or corrupt the control panel, and sizes each batch to the shared fix module. - Behavioral verification: Requires executable tests on real data, a non-vacuity mutation check (the test must turn red when the fix is deliberately broken), and a sibling sweep that caps how widely a defect class may spread. - Gated delivery with independent review: Runs a lot gate (cancello-lotto.mjs) with hard blockers, mandates an independent acceptance pass by an agent that did not build the fix, and supports parallel lanes with disjoint file territories. - Use Case: Nicola says "risolvi i difetti principali" — the Skill loads, measures open defects by disease, picks the costliest cluster, extracts the decision logic into a pure module, writes per-defect tests with mutations, runs the gate, and delivers a reviewable PR without merging. ## Quick Start Ask the assistant to resolve the open defects in the machine's self-awareness registry, starting from the most blocking ones.
