proof-writer

Writes rigorous mathematical proofs for ML and AI theory claims.

52|Updated Apr 24, 2026
One-click install
npx skills add https://github.com/Immortalqx/my_codex_skills --skill proof-writer-immortalqx
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: proof-writer
Source: https://github.com/Immortalqx/my_codex_skills/tree/main/proof-writer
Command: npx skills add https://github.com/Immortalqx/my_codex_skills --skill proof-writer-immortalqx

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Drafting proofs for theorems, lemmas, and propositions in ML/AI research often produces polished but logically flawed arguments with hidden gaps, unstated assumptions, or overclaimed conclusions. This Skill enforces mathematical honesty by triaging feasibility, mapping dependencies, and explicitly flagging claims that cannot be proven as stated. ## Core Features & Use Cases - Feasibility Triage: Checks whether the conclusion actually follows from the stated assumptions before writing, and refuses to fabricate proofs for unjustified claims. - Structured Proof Package: Produces a standardized document with the exact claim, assumptions, notation, proof strategy, dependency map, numbered steps, and open risks. - Honest Status Reporting: Labels every result as PROVABLE AS STATED, PROVABLE AFTER WEAKENING, or NOT CURRENTLY JUSTIFIED, with counterexamples or repaired statements when the original claim fails. - Use Case: You have a convergence lemma for a new optimizer with a rough proof sketch. Provide the statement and sketch, and receive either a complete rigorous proof, a weakened provable variant, or a blockage report identifying the missing assumption. ## Quick Start Use the proof-writer skill to write a rigorous proof of the lemma in my notes file, or tell me if it cannot be proven under the stated assumptions.

Frequently Asked Questions about proof-writer

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I write a rigorous proof for a machine learning theorem?

Provide the exact theorem statement, assumptions, notation, and any proof sketch you have. The skill normalizes the claim, builds a dependency map of lemmas and cited theorems, then writes a numbered proof with justification for every nontrivial step in a structured proof package file.

How to check if a proof can be completed under given assumptions?

The skill performs feasibility triage before writing: it verifies the conclusion follows from the listed assumptions, checks cited theorems are used within their conditions, and looks for counterexamples or quantifier failures. If unjustified, it reports the blockage instead of fabricating a proof.

What happens when my theorem statement is too strong to prove?

The skill never silently strengthens assumptions or narrows scope. It either proposes an explicitly weakened claim with a full proof, or issues a blockage report with a counterexample or repaired statement, preserving your original theorem statement throughout.

Can I use this to fill in missing steps of an existing proof draft?

Yes. Point to your existing proof file, theorem notes, or appendix drafts, and the skill reads them first, updates the relevant claim section, and fills gaps without duplicating prior content. It defaults to writing in PROOF_PACKAGE.md unless you specify another target.

What are the limitations of automated proof writing for research papers?

The skill produces rigorous human-readable proofs, not machine-checked formalizations like Lean or Coq. Genuinely open problems will receive a blockage report rather than a proof, and any remaining fragile points are listed explicitly under Open Risks.