What problem does it solve? Developing a formal proof for a machine-learning theory result requires a structured roadmap before any step-level proof writing begins. This Skill converts a formalized setting and goal into an explicit proof sketch with lemma-sized steps, dependency graphs, assumption provenance, and mechanism witnesses, so downstream proof stages have a verifiable plan instead of broad prose. ## Core Features & Use Cases - Proof Roadmap Selection: Selects or repairs a proof roadmap targeting the exact formalized goal, using prior branch failures as negative signals and technical surveys for candidate proof frameworks. - Lemma-Sized Step Decomposition: Splits the roadmap into stable step IDs with exact intended claims, acyclic dependencies, assumption IDs, proof tools, rate objectives, and pending review status. - Contract-Gated Viability Checks: Applies shared gates for assumption provenance, noncircular closure, scope accumulation, generated output flow, and baseline invariance, recording blockers instead of deferring hard obligations. - Use Case: Given a formalized theorem setting and a proof tracker from a research branch, produce a proof_sketch.md that decomposes the argument into reviewable steps with explicit mechanism sources before any detailed proof writing. ## Quick Start Ask the agent to run the proof-sketch stage on the current idea branch to generate or repair proof_sketch.md from the existing setting.md and proof_tracker.md.
