repro-auditlisted

# Repro Audit (Opus) **Purpose**: Walk a project repository against the four-layer reproducibility checklist (environment pinning, data versioning, workflow orchestration, deterministic randomness) and produce a specific, actionable remediation list. Designed to be run before manuscript submission, before sharing code with collaborators, or after the first "I can't reproduce your result" email. **Model**: Opus 4.7 — broad knowledge of ecosystems (Python, R, Julia, conda, container) and ability to read code at scale. **When to invoke autonomously**: - The user mentions "I'm about to submit", "share with reviewers", "reproducibility", or "rerun". - The user reports non-deterministic behaviour: "I get different numbers each time". - The user has a Jupyter notebook with no environment file. - The user is preparing a Zenodo deposit or DOI for code. **Do NOT invoke** for: - New projects with no code yet (use the [[Reproducible Research Workflows]] template instead). - Pure-statistics issues (use `/stats-consult`). - Code review for correctness (use `@code-review-expert`). ## Usage ``` /repro-audit Audit this repo before I submit to bioRxiv /repro-audit Why do I get different UMAP coordinates every time I rerun? /repro-audit Check if a reviewer could rerun my analysis from this folder /repro-audit Is my notebook pipeline reproducible? ``` ## What This Skill Does ### 1. Inventory Pass First, characterise the project. Run (or ask the user to run): ```bash ls -la