breaking-driven

# Breaking-Driven Development Patching an accreted subsystem preserves its shape. Every fix threads through the structure that caused the bug, and the structure survives the fix. This skill stops patching. Tear the implementation out, state what it owes, and build the replacement from that statement — not from the wreckage. Bloat is the trigger, and derivation is the method. You do not know the right shape when you start; you derive it. What the old code did that the derived contract never names is **residue**, and residue gets cut. One veto point: a **boundary** surface — something outside this repo depends on — stops for an explicit yes before it is cut. Everything **interior** gets demolished without asking. ## Scope Repo-wide is opt-in, never inferred. - A named path, or a target identified in the request, is the whole job. Start at step 2 on it. - A bare invocation, or wording that asks for a sweep, surveys the repo and works the ranked list. "This module is bloated" demolishes that module. Escalating a named grievance into a whole-tree campaign is this skill's worst failure, and trigger phrasing alone cannot tell the two apart — so any identifiable target scopes to itself. A repo-wide run is a campaign, not an edit: one target at a time, each landing as its own atomic commit. Half-demolished is the forbidden state. Finish a target or revert it; never ship the middle. ## Workflow 1. **Pick the target.** A named path or identified target goes straight to step 2....