An Agentic Bug-Hunt in SymPy

The open-source computer algebra system SymPy is a core piece of mathematical infrastructure, widely trusted as an oracle for symbolic computation. In this work, we present an agentic harness, built on current coding agents, that audits such software: it searches for candidate correctness bugs, independently verifies them, checks for public duplicates, diagnoses likely source-level mechanisms, and emits reproducible artifact bundles. Applied to SymPy, the harness finds many silent correctness failures in routine algebra and calculus. We present a curated catalog of representative bugs, each accompanied by code that reproduces the failure and a likely explanation of the root cause.