FactorSmith

FactorSmith unifies two complementary approaches: the factored POMDP decomposition from FactorSim for principled context reduction, and the planner–designer–critic agentic pattern from SceneSmith for iterative quality refinement. By embedding SceneSmith's three-agent workflow within each step of FactorSim's factored generation pipeline, FactorSmith synthesizes playable game simulations from natural language descriptions with improved prompt alignment, fewer runtime errors, and higher code quality compared to either approach alone.

Architecture

The pipeline operates in three phases:

1. High-Level Decomposition

The game specification is decomposed into a sequence of modular steps via Chain-of-Thought prompting, where each step describes a self-contained module (e.g., "introduce a ball that falls under gravity").

2. Factored Step Execution

Each step is processed by a series of agentic trios (planner/designer/critic). A State Agent identifies relevant and new state variables, selecting only the contextual subset from the session database. A Decompose Agent splits the step into three MVC components — input logic, state transition, and UI rendering — each handled by its own agentic trio. Within each trio, the designer proposes code, the critic evaluates it via structured scoring rubrics, and the planner orchestrates iterative refinement with checkpoint rollback.

3. Assembly and Validation

All generated functions and state variables are assembled into a complete executable simulation and validated through sanity checks.

All state variables and functions are persisted in a SQLite-backed session database, enabling factored context selection for subsequent steps.

Available Games

pong, snake, pixelcopter, puckworld, waterworld, flappy_bird, breakout, catcher, space_invaders, monster_kong, raycast_maze

Getting Started

Paper

FactorSmith: Agentic Simulation Generation via Markov Decision Process Decomposition with Planner-Designer-Critic Refinement

Authors: Ali Shamsaddinlou, Morteza NourelahiAlamdari

Abstract: Generating executable simulations from natural language specifications remains a challenging problem due to the limited reasoning capacity of large language models (LLMs) when confronted with large, interconnected codebases. This paper presents FactorSmith, a framework that synthesizes playable game simulations in code from textual descriptions by combining two complementary ideas: factored POMDP decomposition for principled context reduction and a hierarchical planner-designer-critic agentic workflow for iterative quality refinement at every generation step. Drawing on the factored partially observable Markov decision process (POMDP) representation introduced by FactorSim [Sun et al., 2024], the proposed method decomposes a simulation specification into modular steps where each step operates only on a minimal subset of relevant state variables, limiting the context window that any single LLM call must process. Inspired by the agentic trio architecture of SceneSmith [Pfaff et al., 2025], FactorSmith embeds within every factored step a three-agent interaction: a planner that orchestrates workflow, a designer that proposes code artifacts, and a critic that evaluates quality through structured scoring, enabling iterative refinement with checkpoint rollback. This paper formalizes the combined approach, presents the mathematical framework underpinning context selection and agentic refinement, and describes the open-source implementation. Experiments on the PyGame Learning Environment benchmark demonstrate that FactorSmith generates simulations with improved prompt alignment, fewer runtime errors, and higher code quality compared to non-agentic factored baselines.

Paper Link: https://arxiv.org/pdf/2603.20270

Visit the GitHub repository to get started with FactorSmith.