What problem does it solve? When developing machine-learning theory proofs, jumping straight from a high-level proof sketch to detailed step-level proofs often hides missing mechanisms, broken dependency flows, and theorem-contract defects until late in the process. This Skill produces a review-ready, theorem-level diagnostic draft that exposes hard steps, closure mechanisms, and obstructions before any step-level proof work begins. ## Core Features & Use Cases - Theorem-Level Diagnostic Drafting: Expands an accepted proof sketch into a detailed whole-proof draft with named theorem-level blocks, dependency flow, and assembly logic. - Mechanism and Closure Audits: Applies shared gates (Assumption Provenance, Generated Output Flow, Theorem-Critical Mechanism Witness, Residual-To-Target Adequacy) to certify that every theorem-critical condition has a concrete closure mechanism and source. - Failure Classification and Routing: Classifies hard obligations as step-local, sketch/interface defects, or idea/theorem-contract defects, and routes blockers back to the proof-sketch stage with COMPLETE, PARTIAL_BLOCKED, or SKETCH_BLOCKED status. - Use Case: After a proof sketch for a convergence theorem is accepted, run this Skill to generate global_proof.md, which maps every sketch step to theorem-level blocks, certifies recurrence and invariant closures, and flags any missing mechanism before step workers start. ## Quick Start Ask the agent to run the global-proof stage on the accepted proof sketch in the current idea branch to produce the theorem-level diagnostic draft global_proof.md.
