What problem does it solve? When an AI system produces a wrong or fabricated answer, teams often guess at fixes or rewrite prompts blindly. This Skill reconstructs exactly what evidence the system had when it failed, pinpoints where the failure originated, and recommends the smallest fix that addresses the actual cause. ## Core Features & Use Cases - Failure Reconstruction: Rebuilds the exact input, available context or retrieved documents, and the bad output to establish an evidence trail. - Root Cause Classification: Distinguishes no-evidence failures from evidence-ignored failures, and rules out retrieval misses, stale data, tool errors, and prompt framing before blaming model overreach. - Targeted Remediation: Recommends the highest-leverage fix for the identified origin point plus a verification case that would catch a recurrence. - Use Case: A RAG chatbot cites a policy that does not exist. Use this Skill to determine whether the retrieval pipeline missed the right chunk, the prompt rewarded confident completion, or the model inferred beyond the retrieved evidence. ## Quick Start Ask the agent to investigate why the system produced a specific wrong answer, providing the failing input, the output, and any available context or logs.
