causal-knowledge-graph

name	causal-knowledge-graph
description	Persist and query biomedical causal triples from BioCSSwitch conversations, literature checks, and debate outputs.

Causal Knowledge Graph

Use this skill when a biomedical answer produces causal claims that should be remembered, checked for conflicts, or converted into new hypotheses.

Workflow

1. Extract candidate triples with kg_extract_triples.
2. Curate entity names, context, evidence IDs, experiment type, model system, and confidence before persistence.
3. Before adding anything new, run kg_query for the subject/object/context to see what the local graph already contains.
4. Persist curated triples with kg_add_triples.
5. Before asserting a causal direction, run kg_conflict_scan.
6. For mechanistic explanations, run kg_causal_paths and treat paths as hypotheses unless every edge is strongly supported.
7. For next-step planning, run kg_gap_analysis and hand high-priority gaps to bio-experiment.agentic_experiment_plan.

Output Standard

Each causal edge should include subject, relation, object, evidence, direction, context, experiment type, model system, confidence, timestamp, and source. Do not store PHI. If a claim has no PMID, DOI, NCT, or local evidence record, mark it as a gap rather than established knowledge.