What problem does it solve? When research findings are updated after artifacts (rules, skills, templates, agents) were extracted from them, or after schematics were curated against them, the artifacts silently go stale. This Skill scans those artifacts, compares each source finding's last_updated date against the artifact's extraction or curation date, and produces a dated drift report so a human can decide what needs re-extraction or re-evaluation. ## Core Features & Use Cases - Dual-root scanning: Walks extracts/{rules,skills,templates,agents}/ using scalar source_finding pointers and knowledge/schematics/ using array grounded_in pointers, flagging drift when a finding's last_updated strictly post-dates the artifact's baseline date. - Closed-enum recommendations: Each drift hit gets one of three extract recommendations (re-run /extract-artifacts, dismiss as cosmetic, reclassify) or one of two schematic recommendations, keeping the report machine-checkable. - Read-only reporting: Writes only a per-run report to operations/drift-reports/ and never modifies artifacts or invokes other pipeline skills. - Use Case: After a large research-finding intake, run the scan to learn which existing rules, skills, templates, agents, and schematics are now backed by newer source material before planning a re-extraction pass. ## Quick Start Run the detect-drift skill to scan all extracts and schematics for source drift and write a dated report to operations/drift-reports/.
