Multi-agent systems built on large language models increasingly exhibit complex social behaviors, yet the field lacks systematic frameworks for studying emergent discourse topologies. We introduce a blueprint framework that treats multi-agent deliberation as a Markov Decision Process and provides a reproducible testbed for measuring how communication modalities—prompt-level interventions including identity assignment, affective priming, and epistemic framing—shape network structure, lexical diversity, and affective dynamics. Our initial benchmark spans 12 experimental conditions across five conceptual dimensions (Identity, Affect, Epistemics, Architecture, Language), comprising 97 sessions and 3,869 agent-generated messages. Using reply-network Gini coefficients as our primary centralization metric, we identify three emergent topology clusters—egalitarian, moderate, and hierarchical—and report a counter-intuitive echo paradox: homophilic echo chambers produce flat, egalitarian networks (Gini = 0.026), while a single negative-valence agent induces extreme star topologies (Gini = 0.385, Cohen’s d = 7.50). We further show that epistemic framing doubles lexical diversity growth (moral: 21.5% vs. factual: 10.3%, p = .006), whereas sampling temperature yields no significant structural effect (p = .17). We release the complete framework—conditions, metrics pipeline, and dataset—as an open toolkit for researchers studying emergent social phenomena in artificial multi-agent populations. Keywords: Generative Multi-Agent Systems, Social Topology, AI Sociology, Sentiment Contagion, Network Centralization, Echo Chambers, AI Alignment.