Everythingmath — Development skill for Claude Code

everythingmath

**One command for any math problem, start to finish.** A Claude Code skill that acts as problem-setter, solver and the hardest referee you will meet. It works out what math job you have, reads the problem exactly, picks the best skills for it out of a catalog of about 120, solves with a plan, then tries to break its own answer three ways, fixes what it finds, and proves the result does what you asked before handing it over, with an honest confidence label.

/everythingmath prove that (a+b+c)^2 >= 3(ab+bc+ca) for positive reals
/everythingmath compute this integral and double check it
/everythingmath is my proof right? (paste or attach it)
/everythingmath solve this Putnam problem
/everythingmath explain Stokes' theorem from scratch
/everythingmath formalize this lemma in Lean
/everythingmath here is my problem set, help me with it

You do not have to type the command. Describing the math is enough ("solve this", "prove it", "check my answer").

Why it is different

Most "solve my math problem" prompts produce a fluent answer that nobody checked. everythingmath is built around the ways math answers go wrong.

• It tests claims, not formatting. Every identity, derivative, integral value, closed form and limit goes through a bundled checker. A false statement is refuted by a counterexample, not argued around.
• Two routes to every answer. One derivation is a guess that looked consistent. The final answer must be confirmed by a second, independent method.
• A blind re-solve. A fresh agent sees only the problem, never your work, and solves it. The answers must match. A mismatch is the strongest error signal there is.
• The referee reads every step. step_review.py numbers each step; each is read against a ten-point test (theorem hypotheses, quantifier order, division by zero, case coverage, induction base, interchange of limits, and so on). A gate refuses to pass while a step is unread or a serious issue is open.
• **An adver