by Anna Rosen
Research-coding workflow for computational science, built with a JAX/Python astrophysics group and domain-agnostic in its contracts. The researcher owns the science: scientific assumptions are proposed and approved before they become load-bearing, experiments carry a derived prediction, and validation is staged (physics and analytic checks in development, published-result or reference-code comparison before release). 43 skills: model development and ideation, literature, scoping, building correctly, equation-critical sources, verification and numerical validation, Bayesian inference, code and figure review, performance, records and provenance, MyST documentation, reproducibility and release. Plus an equation-verifier agent, six slash commands, twelve self-limiting hooks (Stop gates warn by default; RWF_STRICTNESS=standard blocks), and per-domain lenses for reference-parity work. The same skill files load in Codex via a symlink.
Claude Code30 Skills