What problem does it solve? Developing a rigorous mathematical proof requires decomposing a high-level proof sketch into individually verified steps, each with self-contained local lemmas, discharged citation assumptions, and explicit notation provenance. This Skill automates the disciplined construction of a single proof step so reviewers can audit every local unit without hidden gaps. ## Core Features & Use Cases - Local Proof Unit Decomposition: Splits each nontrivial subclaim into named lemmas, propositions, claims, hypotheses, or invariants with stable IDs and paper-ready LaTeX labels. - Citation Assumption Checking: Restates every cited theorem in current notation and verifies where each required assumption is discharged before use. - Assumption and Constant Provenance Tracking: Classifies every new symbol, constant, rate, and invariant condition as setting-derived, dependency-supplied, or locally proved. - Use Case: Given an accepted proof sketch for a learning-theory theorem, invoke this Skill on Step 3 to produce a self-contained proof artifact with local derivations, checked cited results, and an explicit target-step assembly ready for independent review. ## Quick Start Ask the agent to prove Step 3 from the accepted proof sketch in perspective_1/idea_2 using the proof-step workflow and write the result to proof_steps/step_003/proof.md.
