What problem does it solve?
When ingesting data bundles into a semantic layer, duplicate names, contradictory metric definitions, and re-ingested artifacts can silently corrupt the knowledge base. This Skill gives the agent a deterministic decision tree for classifying each collision and applying the correct resolution before any write lands.
Core Features & Use Cases
- Conflict classification: Distinguishes same-content duplicates, expression-only re-ingest changes, semantic breaks, structural duplicates, near-duplicate clusters, and definitional contradictions.
- Resolution rules: Applies canonical election by inbound-ref count, rename-and-capture disambiguation with domain suffixes, silent replacement for expression-only changes, and eviction handling for Stage 4 sweeps.
- Canonical pins and wiki capture: Honors prior user decisions from
<canonical_pins> blocks and writes unified <concept>-definitions.md wiki pages documenting every variant with provenance.
- Use Case: During a Stage 4 reconciliation sweep, two sources define
churn_risk with different formulas; the agent renames them to churn_risk_engagement_based and churn_risk_billing_based, writes a unified wiki page, and flags the contradiction for human review.
Quick Start
Ask the agent to triage the conflicts detected in the current ingest bundle and resolve each one according to the ingest triage decision tree.