What problem does it solve? Turning a pile of verified papers into a coherent, judgment-driven literature review is hard: naive approaches produce per-paper summaries with decorative citations, missed contradictions, and fabricated references. This Skill enforces an evidence-first pipeline that clusters papers by theme, builds a MECE taxonomy, cross-compares findings, surfaces contradictions, and runs self-adversarial review before producing a structured synthesis. ## Core Features & Use Cases - Evidence-first synthesis: Assigns each paper to themes via a Source × Theme matrix before writing, so every claim traces back to allocated literature. - Quality and gate enforcement: Applies source-quality tiers (A/B core evidence), citation-cluster limits, and six internal quality gates (Angle, Coverage, Citation, Taxonomy, Calibration, Weaving) with explicit BLOCKED output codes when inputs are insufficient. - Contradiction and gap analysis: Presents conflicting findings with condition analysis, converts taxonomy gaps into open questions, and translates them into actionable research directions with topic, rationale, and method sketch. - Logic graph generation: Produces a Mermaid whiteboard logic graph (nodes = themes/schools with evidence IDs, directed edges = relations like extend/refute) via scripts/research_visuals.py. - Use Case: Given a verified list of 30 papers from a literature scout stage, produce a structured synthesis with core conclusions, thematic chapters, a theme-level summary table, controversies, research directions, and a handoff package for downstream review writing. ## Quick Start Run the research-synthesis stage on the verified literature list from the scout handoff to produce a structured thematic synthesis with claim, citation, tension, and gap pools.
