ODD+LLM: Specifying LLM-Agent Behavior\\for Reproducible Social Simulation

The ODD protocol is the de facto standard for describing agent-based models, but it lacks vocabulary for the components that determine LLM-agent behavior: prompt assembly, memory serialization, parsing, and retry logic. We propose \textbf{ODD+LLM}, a backward-compatible extension adding five documentation blocks grounded in the CoALA cognitive-architecture framework. To demonstrate that the gap is consequential, we conduct a \emph{multiverse analysis} \citep{steegen2016} of a published naming-game simulation: we identify six specification choices (four determinable from supplementary materials, one specified in the main text but non-obvious, one genuinely open), implement eight defensible readings, and measure the spread in the paper's headline metric. A single choice (whether the agent is framed as a third-person observer or first-person participant, specified in the main text but easily missed without a standard home) shifts the probability of converging on M (the microscopically ``strong'' convention) from 0.35 to 1.00 in our reimplementation and accelerates convergence threefold. We complement the demonstration with a reconstruction-sufficiency review of 28 published LLM-agent simulations, finding that Block~D binding and resolution items are undetermined in 77\% of cases and that no paper's primary call site is fully determined. ODD+LLM gives the field a standard home for information it already produces but currently scatters across code repositories, supplementary files, and undocumented defaults.