What problem does it solve? Analytical code often runs cleanly while producing silently wrong results — wrong statistical tests, data leakage, or fabricated outputs — and these errors slip past generic code review. This Skill audits PRs, commits, diffs, and working trees for the failure modes that matter in scientific data work. ## Core Features & Use Cases - Six specialized review passes: statistics/causal inference, data pipeline leakage, bioinformatics, LLM failure modes, documentation fidelity, and code quality, each with its own checklist. - Prioritized synthesis: findings are deduplicated, recalibrated to Major/Minor severity, grouped by category, and written to a dated report under .living/outputs/reviews/. - Grill mode and tripwires: conversationally interrogate consequential analytical decisions one question at a time, or design perturbation tests that verify the pipeline fails at named scientific boundaries. - Use Case: Before merging a PR that adds a differential expression analysis, run the review to catch pseudoreplication, uncorrected multiple comparisons, and report numbers that no longer match the source CSV. ## Quick Start Ask the assistant to review the current working tree or a specific PR for statistical, leakage, and code-quality issues using the mycelium review skill.
