Drug-information question answering is a high-stakes setting where hallucinated facts can mislead clinical decision-making and the provenance of each cited fact matters as much as the fact itself. We present DrugClaw, a multi-agent retrieval-augmented system that queries a registry of drug and pharmacovigilance skills via a reflection-driven state-machine workflow and returns answers grounded in primary regulatory or peer-reviewed records. We also contribute DrugAudit, a 3,772-item authority-aware benchmark with an evaluation panel that scores upstream-of-gold source match, token-level semantic snippet overlap, and citation faithfulness under a dual-judge LLM-as-judge protocol with inter-judge κ = 0.88 (almost-perfect). Across DrugAudit plus drug-related subsets of MedQA (751) and PubMedQA (512), DrugClaw is top-1 on every column of the headline table: composite Evidence Index under both judges, judge-mediated answer correctness, primary-source rate (0.918, +10.1 pp over next-best), faithfulness (0.887, +5.9 pp), MedQA (0.920), and PubMedQA (0.693).

Drug–disease knowledge underpins drug discovery and precision medicine, yet most of it is locked in unstructured PubMed articles and only partially captured by curated databases. We release UniD³-KG, a suite of six knowledge graphs spanning three tasks (Drug–Disease Matching, DDM; Drug Effectiveness Assessment, DEA; and Drug–Target Analysis, DTA), and UniD³-data, three open QA datasets (28,915 DDM, 15,042 DEA and >4,000 DTA pairs), distilled from 157,849 PubMed articles with Llama-3.3-70B by the UniD³ KG-RAG pipeline, which couples task-specific prompting with a dual-stage entity extraction that re-extracts over the consolidated graph to recover cross-chunk and cross-document evidence. External validation on CDR, DrugReviews, P3ps and MedicationQA and a benchmark across seven LLM baselines and three retrieval variants show that UniD³-data is faithful to curated sources and that retrieval over UniD³-KG consistently improves biomedical QA; a clinician verification of 469 FDA-approved, US-marketed drugs further confirms strong agreement with the UniD³-data DDM split (accuracy 0.90, F1 0.90, AUROC 0.90). All artefacts—UniD³-KG, UniD³-data, benchmark scripts and the UniD³-QA explainable chatbot—are released as shared infrastructure for biomedical NLP and drug repurposing.