utility-drift-detector

What problem does it solve?

Drift in AI system values and optimization targets after deployment can go unnoticed because outputs may still appear correct even when underlying objectives have shifted. This skill surfaces, quantifies, and helps remediate post-deployment drift by reconstructing the baseline, analyzing current outputs, classifying drift signals, and guiding responses.

Core Features & Use Cases

• Baseline reconstruction to capture deployment intent and acceptable behavior.
• Current output pattern analysis to detect shifts in language, tone, or decision patterns.
• Drift signal classification across value weighting, proxy substitution, scope creep, constraint erosion, and feedback loop drift.
• Magnitude assessment and root-cause analysis to enable actionable remediation and ongoing monitoring.

Quick Start

Install the skill and provide deployment baseline and current outputs to generate a drift report.