What problem does it solve? Changes introduced in later research phases can silently invalidate results that were verified in earlier phases through notation drift, convention changes, or modified shared artifacts. This Skill re-checks every claim marked VERIFIED in phase VERIFICATION.md files against the current project state so silent breakages are caught before the research narrative becomes internally inconsistent. ## Core Features & Use Cases - Verified Target Extraction: Discovers all VERIFICATION.md files under .gpd/phases, extracts VERIFIED claims, and categorizes them as dimensional, limiting_case, symmetry, conservation, numerical, artifact, analytical, or comparison targets. - Cross-Phase Consistency Checks: Detects notation drift, approximation regime violations, and modified upstream derivations by comparing conventions and git history across phases. - Severity-Classified Reporting: Generates .gpd/REGRESSION-REPORT.md with regressions classified as CRITICAL, MAJOR, or MINOR, a phase impact map, and a recommended fix order. - Use Case: After completing phase 5 of a physics derivation project, run the regression check to confirm that the dispersion relation verified in phase 2 still holds given convention changes made in phase 4. ## Quick Start Ask the AI to run the gpd regression check across all completed phases, or pass a phase number to limit the scope, and review the generated REGRESSION-REPORT.md.