Universal mathematical modeling agent for all competitions. Activates when the user mentions "math modeling", "mathematical modeling", "美赛", "国赛", "建模", "MCM", "ICM", "MathorCup", or asks to solve a modeling competition problem. Also use when the user wants to analyze a problem, build a mathematical model, decompose it into subtasks, and solve each subtask with code execution. Covers the complete workflow from problem understanding to LaTeX paper generation.
Stage 3 of the mathematical modeling pipeline. Solves each subtask by dispatching independent subagents that retrieve methods from HMML, generate formulas, write and execute Python code, and interpret results. Each task runs in a subagent to minimize main session context usage. Invoked by the math-modeling skill during Stage 3. Do not invoke directly.
Stage 1 of the mathematical modeling pipeline. Performs deep problem analysis using Actor-Critic self-improvement with percent-based scoring. Invoked by the math-modeling skill during Stage 1. Do not invoke directly — use /math-model instead.
Stage 2 of the mathematical modeling pipeline. Performs high-level modeling, problem decomposition into subtasks, and DAG dependency analysis. Uses Actor-Critic with percent-based scoring. Invoked by the math-modeling skill during Stage 2. Do not invoke directly.
Stage 3.5 of the mathematical modeling pipeline. Performs global quality review of all task outputs using an independent subagent with percent-based scoring. Checks cross-task consistency, numerical credibility, requirement coverage, model coherence, evidence sufficiency, and chart/figure quality. Invoked by the math-modeling skill after Stage 3. Do not invoke directly.
Stage 4 of the mathematical modeling pipeline. Generates LaTeX paper source from all workspace outputs, compiles to PDF. Supports multiple competition templates. Invoked by the math-modeling skill during Stage 4. Do not invoke directly.
Start the mathematical modeling pipeline. Use when the user types /math-model.