Mathbox

Mathbox is a plugin for Claude Code and Codex providing a set of Agent Skills for rigorous AI-assisted mathematical research. Each skill handles one job — pursuing a research program or a single bounded proof attempt, auditing an existing proof, designing or auditing a supporting computation, verifying a literature citation or bounded novelty claim, refereeing a whole manuscript, integrating validated results into a LaTeX manuscript, or proofreading — with its own evidence standard and stopping condition. Optional standard-library Python tools track claims, evidence, and dependencies in a local ledger.

The skills are research workflows and safeguards, not a computer algebra system or a replacement for mathematical review: a bounded computation is treated as evidence only for its stated range, proofreading never changes an argument, and a failed literature search never supports a claim of global novelty. Every skill can also be installed on its own as a standalone Agent Skill.