Guide · Agent skill
How to use the Math Olympiad Solver agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The math-olympiad skill is a set of instructions and helper files that tells an AI agent how to handle olympiad-style problems from the IMO, Putnam, USAMO and AIME. Instead of giving a quick answer, the agent checks how the problem could be read, generates several candidate solutions, and has a separate verifier try to break each proof. This guide explains what the skill does, how to install it, how to trigger it, and where it falls short, so you can decide whether it fits your work.
Math Olympiad Solveranthropics/claude-plugins-official · Apache-2.0 Download the skill Quick start
- 1Download the ZIP. Get the math-olympiad skill ZIP from this page.
- 2Unzip into the skills folder. Unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder.
- 3Or upload in the Claude apps. In the Claude apps, open Settings → Capabilities → Skills and upload the ZIP instead.
- 4Phrase your request. Paste the full problem statement and say whether you want a solution, a proof check, or a counterexample.
- 5Review the result. Read the proof and confidence note. If the agent says it has no confident solution, treat that as a real answer.
What is an agent skill, and what does this one make the agent do?
An agent skill is a folder with a SKILL.md file and supporting reference files. When you install it, the agent reads the instructions when your request matches the skill's description, and follows its workflow for that task.
This skill targets competition math. Its SKILL.md describes an interpretation check, parallel solver attempts that each refine their own work, a cleaning step that removes the solver's reasoning trace, and an adversarial verifier with no tools that looks for specific failure patterns. If no candidate survives verification, the skill says it has no confident solution rather than guessing. If LaTeX is installed, it can compile the finished proof into a PDF.
- Interpretation check: list two or three readings of the problem and pick the intended one.
- Parallel candidate solutions from several solver agents, each with a different starting angle.
- Cleaning: the verifier sees only the problem statement and the cleaned final argument.
- Adversarial verification against known failure patterns.
- Abstain when nothing survives, and optional PDF output.
How do you install the math-olympiad skill?
Download the skill ZIP from this page. Then install it using one of the two methods below. The skill folder must contain SKILL.md at its top level after unzipping.
For a personal install, unzip the file into ~/.claude/skills/. To limit it to one project, unzip it into that project's .claude/skills/ folder instead. In the Claude apps, you can also upload the ZIP under Settings → Capabilities → Skills.
- Personal: unzip into ~/.claude/skills/
- Single project: unzip into .claude/skills/ inside the project folder
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
What do you need before you start?
The skill itself needs no network access. Its scripts, scripts/check_latex.sh and scripts/compile_pdf.sh, are only for producing a PDF from a LaTeX proof. Solver and verifier agents run as pure reasoning and do not call tools, so the PDF step is the only place scripts come in.
For parallel runs, the source expects an agent harness that can launch subagents. For PDFs, pdflatex or xelatex must be installed. Both are optional. The skill still works for answers and proof text without LaTeX.
- Optional: pdflatex or xelatex for PDF compilation
- Optional: a subagent-capable agent harness for parallel solver and verifier runs
- Network: not used by the skill
- Scripts: scripts/check_latex.sh and scripts/compile_pdf.sh are included; read them before running
How do you trigger the skill with a natural request?
Describe the task in plain words that match the skill's description. The skill is designed to activate for requests such as 'solve this IMO problem', 'prove this olympiad inequality', 'verify this competition proof', 'find a counterexample', or 'is this proof correct'. Problems that mention IMO, Putnam, USAMO, olympiad or competition math also match.
Paste the full problem statement, and say whether you want a solution, a check of an existing proof, or a counterexample search. Asking for a check of an existing proof uses the verification path directly and skips solution generation.
- Solve: 'Solve this IMO problem: [statement]'
- Check: 'Is this proof correct? [proof text]'
- Counterexample: 'Find a counterexample to this claim: [claim]'
- Full set: 'Work through all six problems from this competition and compile a PDF.'
What should you expect in the output, and what are the limits?
The output is the agent's final answer, its proof, and its confidence. A proof that passes verification may still be wrong. The source says verification can miss errors, and cites a study where self-verified IMO results dropped sharply under human grading. Treat the output as a strong draft that needs review, especially when a proof matters.
The skill is slow and uses many tokens, because it runs several solver and verifier agents per problem. It is also not a numerical tool. Its design avoids computation, so arithmetic checks or large numeric searches are outside its purpose.
- Expect a 'no confident solution' result for hard problems instead of a guess.
- Expect long runs: many agents per problem.
- Do not rely on it for routine homework or simple calculations.
- Check the final proof yourself before relying on it.
What people use it for
- Checking a proof a coach or student has written for an olympiad inequality.
- Getting a second opinion on whether a competition proof has a gap.
- Searching for a counterexample to a conjecture before trying to prove it.
- Running a multi-agent pipeline over a full problem set and collecting per-problem results, with an optional PDF.
Frequently asked questions
How do I install the math-olympiad skill?
Download the skill ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.
Is the math-olympiad skill free?
This guide does not state pricing or licence terms beyond the listed licence, Apache-2.0. Check the licence file in the source repository for the terms that apply to you.
Does the skill need internet access?
No. The listing says it does not use the network. It includes scripts for LaTeX PDF compilation, so read those scripts before running them.
Do I need LaTeX to use it?
No. LaTeX is optional. pdflatex or xelatex is only needed if you want the skill to compile the finished proof into a PDF.
Can the skill prove a result that is actually wrong?
Yes. The skill says it can still be wrong after verification, so a confident answer should be checked before you rely on it. It is designed to say 'no confident solution' rather than guess when nothing survives verification.
How is this different from asking a chatbot to solve an olympiad problem?
A plain chatbot answer is one pass. This skill generates several candidate solutions, removes the reasoning trace, and has a fresh verifier attack the proof using known failure patterns. It is slower and uses more tokens, so it suits proofs that need checking more than quick answers.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you are a student or coach who needs rigorously checked olympiad proofs, want a second opinion on a competition proof, or are building a pipeline that solves whole problem sets. Pick something lighter if you only need a quick answer to homework or a numerical calculation, since the skill is slow, token-heavy and not designed for computation.
Download the skill