Savant OpenAI Research

Plugin savant-openai in Data-Wise/savant-openai. Commands read from the repo's manifest — 18 days ago.

Skills (5)

• savant-learning-review

  Review bounded-learning candidate lessons for evidence, scope, safety, and regression risk, then report an explicit recommendation. Use when a user asks to inspect, approve, reject, supersede, or revert a Savant learning candidate; never infer promotion from model confidence.

• savant-repo-scaffold

  Create or finish a Codex repository with a GitHub remote, safe branch layout, Codex-aware project detection, preflight checks, and explicit approval gates. Use when initializing a Codex/OpenAI plugin repository, connecting a local repository to GitHub, preparing a feature branch for a PR, or repairi

• savant-research-verify

  Audit mathematical proofs, statistical claims, simulations, or reproducibility with explicit evidence and a scoped verification verdict. Use when the user asks whether a derivation, proof, analysis, or simulation is correct.

• savant-teach-mathematical-proofs

  Convert a stated mathematical proof into a fully taught proof without changing its mathematics. Use when asked to teach, expand, or annotate proofs in Markdown, LaTeX, Quarto, or pasted text: restore skipped algebra, justify each step, define notation at first use, audit definition order, and option

• sequential-wt

  Coordinate bounded implementation tasks from an integration branch by creating, delegating, reviewing, and cleaning up one feature worktree at a time. Use when work should remain isolated from dev while the coordinator retains repository context.