What problem does it solve? When an engineer disagrees with an AI-generated bug diagnosis, there is no consistent way to feed that feedback back into the investigation without losing prior reasoning or starting over blindly. This Skill defines a structured protocol for challenging and revising a bug diagnosis while preserving the full reasoning chain. ## Core Features & Use Cases - Mode Detection: Scans engineer feedback for reject signals ("wrong", "start over") versus layer signals ("also consider", "what about") to pick the right revision mode, defaulting to the safer layer mode when ambiguous. - Layer Mode: Merges the engineer's new hypothesis with the existing root cause into a single unified analysis, updating the spec in place. - Reject Mode: Archives the current Root Cause and Fix Plan into Investigation History, clears them, and re-runs the investigation starting from the engineer's hypothesis. - Append-Only History: Maintains a structured Investigation History with dated rounds, confidence levels, and verbatim feedback so no analysis is ever deleted. - Use Case: An engineer reviews a debug spec, believes the diagnosed root cause missed a concurrency issue, and writes "also consider a race condition in the worker pool" — the skill routes this as a layer challenge and produces a merged diagnosis. ## Quick Start Challenge the current bug diagnosis in the spec by providing your feedback and let the skill decide whether to layer it onto or reject the existing root cause.
