
Proof Writer
Drafts rigorous math proofs for ML/AI theory claims, flagging when a claim is not actually provable.
- Stars
- 18,482
- Licence
- MIT
- Files
- 1 file
What it does
Once installed, the agent writes structured mathematical proofs for theorems, lemmas, propositions and corollaries, such as those in ML/AI theory papers. Before proving anything, it checks whether the conclusion really follows from the stated assumptions, and it may report that a claim is only provable after weakening it or cannot currently be justified, instead of producing a fake proof. Results go into a proof document (by default PROOF_PACKAGE.md in the project root), and it can optionally call a separate Lean formalization skill for machine-checked verification.
Great for
- ML or AI theory researchers drafting appendix proofs for papers
- Students checking whether a proof sketch actually works under its assumptions
- Reviewers who want to test whether a claimed theorem is correctly scoped
Skip it if
- Users who need a finished, peer-reviewed proof without expert review: the output is a draft and can contain errors or gaps the agent does not catch
- Non-mathematical writing or general coding tasks, which this skill does not address
Try asking
- “Prove this lemma from my assumptions in the attached notes and list any hidden assumptions”
- “Fill in the missing steps in my proof sketch for the convergence bound”
- “Determine whether this theorem can be proved as stated, or what extra assumption it needs”
- “Formalize this proof sketch and write the result to PROOF_PACKAGE.md”
Install in 3 steps
- 1
Download the skill
The ZIP contains the 1 original files, the MIT licence and an attribution note.
proof-writer.zip - 2
Add it to your agent
Claude Code (all projects)
mkdir -p ~/.claude/skills && unzip proof-writer.zip -d ~/.claude/skills/Claude Code / Agent SDK (this project only)
mkdir -p .claude/skills && unzip proof-writer.zip -d .claude/skills/Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.
- 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
- Give the exact theorem statement and all assumptions up front, since the skill refuses to strengthen assumptions silently.
- Point it to existing notes or draft files so it can reuse nearby lemmas.
- Treat the output as a draft to be checked by a domain expert before it goes into a paper.
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.