gradient-accumulation-deterministic
Implement gradient accumulation that produces bit-identical results to standard batching. Use when: comparing GA vs non-GA runs, debugging training reproducibility.
memory-write
Persist a decision or learning to Mother-Harness long-term memory
speckit-specify
Create or update the feature specification from a natural language feature description.
experiment-logger
Log ML experiments with hyperparameters, metrics, and plots; human interprets results and plans next experiments
categorical-encoder
Эксперт categorical encoding. Используй для ML feature engineering, one-hot, target encoding и embeddings.
machine-learning-engineer
Use when user needs ML model deployment, production serving infrastructure, optimization strategies, and real-time inference systems. Designs and implements scalable ML systems with focus on reliability and performance.
paper-replication
深度学习论文复现的skill。可以读取pdf并解析其中的图片、公式、表格等内容，然后参考下面的prompts。触发词包括"帮我复现这篇论文"、"论文复现"、"实现这个模型"，或当用户提供深度学习论文需要转化为PyTorch代码时。
calibration
Applies decision thresholds for high-confidence inputs or enforces conservative safety margins for low-confidence cases
moai-lang-python
Python best practices with modern frameworks, AI/ML integration, and performance optimization for 2025
ground-truth-management
Comprehensive guide to creating, managing, and maintaining ground truth datasets for AI evaluation including annotation, quality control, and versioning
scaling-analysis
Run scaling experiments to understand model/data/compute relationships. Use when investigating scaling laws, compute-optimal training, or model size decisions.
quant-resource-patterns
Follow these patterns when implementing quant domain resources like Dataset, Signal, Alpha, Portfolio, Strategy, Universe, Backtest, or MonitoringRun in OptAIC. Use for creating DB models, DTOs, services, and tests for trading-specific entities.
time-series-models
Bayesian time series models including AR, MA, ARMA, state-space models, and dynamic linear models in Stan and JAGS.
production-api-tester
Live testing and validation of production research API for strategy optimization loops
feature-extraction
Extracts vector embeddings for fairness analysis using MobileNetV3 with FairDisCo disentanglement
markov-regime-features
Debugging constant Markov regime features in RL observations - when HMM probabilities show uniform values instead of dynamic regime estimates
mlops-patterns
Follow these patterns when implementing MLOps features in OptAIC. Use for ML model definitions (5-component structure), model instances, training/inference pipelines, model registry, and monitoring. Covers signal models, macro regime models, relevance models, and signal combining/filtering models.
exps-logistic
Documentation for the logistic regression MI estimation experiment (exps_logistic)
network-architecture-sizing
PPO network architecture sizing for trading models. Trigger: (1) model files are unexpectedly small/large, (2) choosing hidden_dims for training, (3) balancing model capacity vs inference speed.
refactorscikit-learn
Refactor Scikit-learn and machine learning code to improve maintainability, reproducibility, and adherence to best practices. This skill transforms working ML code into production-ready pipelines that prevent data leakage and ensure reproducible results. It addresses preprocessing outside pipelines, missing random_state parameters, improper cross-validation, and custom transformers not following sklearn API conventions. Implements proper Pipeline and ColumnTransformer patterns, systematic hyperparameter tuning, and appropriate evaluation metrics.
training-archive-gating
Mandatory training archive with model gating (APPROVED/REVIEW/DROP). Trigger when: (1) training run completes, (2) need to decide which models to deploy, (3) want historical training reference, (4) need checkpoint recommendations for overfitting.