Skip to content

Proof Checker

Lets the agent rigorously verify and fix a LaTeX math proof using a second AI reviewer.

Stars
18,482
Licence
MIT
Files
1 file

What it does

When you ask the agent to check a mathematical proof, this skill makes it read a LaTeX proof, look for logical gaps and unjustified steps, and send the proof to a separate AI model acting as an adversarial reviewer. The agent then fixes each flagged gap with fuller derivations, has the reviewer re-check the fixes, and produces an audit report that tracks each issue. The file describes a fixed issue taxonomy and an objective pass/fail acceptance gate.

Great for

  • Researchers preparing theory papers who want a systematic second check of their proofs
  • Graduate students reviewing a lemma or theorem before submission
  • Authors who want an audit trail of which gaps were found and how they were closed

Skip it if

  • Quick informal sanity checks where a single human read is enough (the workflow is multi-round and slow)
  • Proofs that need formal machine verification in a proof assistant, since this relies on LLM review rather than a formal checker
  • Situations where the proof text is confidential and cannot be sent to an external reviewer model (the reviewer backend receives the proof content)

Try asking

  • “Check this proof in paper/main.tex and fix any gaps”
  • “Verify the proof of Theorem 3 with a deep fix pass”
  • “Run a proof check on my convergence lemma and generate an audit report”
Step-by-step guide: how to use Proof CheckerInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 1 original files, the MIT licence and an attribution note.

    proof-checker.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip proof-checker.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip proof-checker.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Point it at a specific .tex file or theorem so the reviewer has clear scope
  • Keep the reviewer backend choice deliberate: the default sends content to an external model, and manual mode requires a non-Claude reviewer model
  • Read PROOF_AUDIT.md after each run, since the reviewer can still be wrong and human checking remains necessary

What's inside

  • SKILL.md

Redistributed unmodified from wanshuiyin/Auto-claude-code-research-in-sleep under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.