What problem does it solve? Code review and bug hunting often fail on subtle issues like name shadowing, scope ambiguity, and fixes that address symptoms instead of root causes. This Skill enforces mandatory verification checkpoints before any conclusion, catching the failure modes that even strong reasoning misses. ## Core Features & Use Cases - Patch Verification: Resolve every function call through a 5-step scope sequence, trace execution before and after the patch, check downstream callers for regressions, and emit a CORRECT/LIKELY_CORRECT/CONCERNS/BUGGY verdict. - Fault Localization: Trace the buggy code path line by line and apply a sufficiency test ("fix ONLY this line") to distinguish root causes from mere contributors. - Patch Equivalence: Compare two candidate fixes test by test to determine whether they produce identical outcomes, with counterexamples when they diverge. - Model-Adaptive Templates: Full procedural templates with worked examples for Haiku-class models, compact verification checkpoints for Sonnet/Opus-class models. - Use Case: When reviewing a diff that changes a function call, the Skill detects that a module-level function shadows a builtin, traces both execution paths, and confirms the patch correctly bypasses the shadowing. ## Quick Start Ask the AI to review this patch using semi-formal reasoning and give a verdict on whether the fix is correct.
