Critique ML analyses for methodological flaws and actionable improvements.
npx skills add https://github.com/tim-krausz/mlstack --skill model-critique
Please help me install this Agent Skill. Skill: model-critique Source: https://github.com/tim-krausz/mlstack/tree/main/model-critique Command: npx skills add https://github.com/tim-krausz/mlstack --skill model-critique
Adversarial model evaluation and critique to reveal hidden biases, methodological gaps, and overconfident conclusions. It helps ensure the analyst's choices are robust, transparent, and publication-ready.
Provide a structured adversarial critique of the given analysis, identifying framing flaws, data leakage, modeling decisions, and evaluation weaknesses, then propose concrete improvements.
High-intent search queries and answers about installing and using this skill.
An adversarial review evaluates ML analyses by scrutinizing problem framing, data handling, modeling choices, and evaluation strategies to surface methodological flaws. It outputs a structured critique with severity levels and concrete fixes for publication readiness.
Data leakage detection traces preprocessing steps to identify where training information improperly influences validation sets. Auditing data handling reveals leakage risks, quality issues, and preparation biases that compromise model validity.
Yes, evaluation and inference scrutiny assesses validation strategy, metrics, statistical significance, and claim validity. It examines whether your evaluation framework supports robust, transparent conclusions without overfitting or inappropriate measurements.
Auditing a notebook involves evaluating whether the defined problem aligns with business or research objectives, flagging misframing, and tracing data quality issues. It checks training dynamics and hyperparameters for interpretability and validity.
Critiquing modeling decisions examines model choice, hyperparameters, and training dynamics to ensure validity and interpretability. The critique identifies overconfident conclusions and proposes actionable improvements to make training robust.
Adversarial evaluation reveals hidden biases by rigorously challenging the analyst's choices throughout the pipeline. It is limited to analyzing the provided notebooks, pipelines, or markdown plans and does not execute code to validate runtime behavior.
debug-zhuweijianAnalyze ML experiment results into reports, statistics, and figure catalogs.
CHENyiru3Critique bioinformatics manuscripts for methodological flaws and data leakage.
KentoShimizuAutomate ML data preprocessing planning with leakage-free transformation specifications.
cosmixEvaluate machine learning models with metrics, validation, and monitoring.
