Skill UI

A comprehensive guide for defining, tracking, and interpreting product metrics across all stages—from discovery to maturity. This skill covers selecting metric frameworks (AARRR, HEART), defining stage-appropriate KPIs, designing multi-layered dashboards, and executing deep analyses like cohort and retention modeling. It helps teams transform raw data into actionable product strategies, moving beyond vanity metrics.

This skill provides a systematic framework for analyzing cyber campaign attribution. It involves collecting, weighting, and correlating diverse indicators—such as infrastructure overlaps, TTP consistency, malware code similarities, and operational patterns—to determine the responsible threat actor. Utilizing structured methodologies like Analysis of Competing Hypotheses (ACH) and the Diamond Model, users can build confidence-weighted assessments to enhance threat intelligence accuracy.

This skill provides a comprehensive guide to using the Malpedia API for advanced malware intelligence analysis. Users can track malware family relationships, analyze variant evolution, extract associated YARA rules for detection, and map connections to specific threat actors. It is essential for security researchers and SOC analysts building robust threat models.

This skill implements the Diamond Model framework, providing a structured approach to analyzing complex cyber intrusions. It allows users to programmatically classify and correlate events based on four core elements: Adversary, Capability, Infrastructure, and Victim. Key functionalities include building activity threads, identifying pivot points, and generating comprehensive, pivot-ready threat intelligence reports.

This advanced skill systematically verifies the novelty of a proposed academic method or research idea against the latest scientific literature. It performs multi-source searches across databases like arXiv, Google Scholar, and major conference proceedings (ICML, NeurIPS). Using a multi-step process, it extracts core claims, searches multiple sources, performs cross-model verification via LLM, and outputs a structured Novelty Report, detailing overlap, prior work, and overall recommendation.

This skill implements a multi-factor scoring model to assess the criticality of IT assets. By weighting factors such as business function impact, data sensitivity, regulatory scope, and network exposure, it assigns a risk tier. This tier is then used to dynamically adjust the Service Level Agreements (SLAs) for vulnerability remediation, ensuring that organizations prioritize patching efforts on their most critical 'crown jewel' systems first.

bdistill transforms AI interactions into a structured, queryable knowledge base. It extracts, structures, and quality-scores domain knowledge from both closed-source AI models (in-session) and local open-source models (via Ollama). Users can build comprehensive reference datasets, Q&A pairs, or generate structured training data for traditional machine learning, ensuring validated and exportable domain expertise.

This workflow bridges the gap between an abstract experiment plan and concrete, executable code. It reads structured experiment roadmaps (e.g., EXPERIMENT_PLAN.md), automatically generates or modifies necessary code scripts, handles prerequisite setup (like cloning base repositories), deploys the experiments to compute resources (GPU), and meticulously collects initial results. It includes optional steps for cross-model code review, ensuring robustness and correctness before execution. Ideal for moving from research hypotheses to measurable, running results.

This skill provides interactive, periodic monitoring of deep learning training runs. It continuously checks critical metrics—such as loss trends, evaluation scores, and gradient norms—against defined baselines to detect signs of divergence, NaN values, plateaus, or general instability. It delivers a structured report and recommends whether to continue, wait, or stop the training process, ensuring computational resources are not wasted on failing models.

Collection of battle-tested PyTorch training recipes covering architecture selection, optimizers, LR scheduling, precision tips, and debugging to train or fine-tune LLMs, vision, diffusion, and biomedical models with high GPU throughput.

This expert tool provides deep, systematic auditing of datasets to assess completeness, consistency, accuracy, and structural validity. It performs full data profiling, detects anomalies (outliers), analyzes missing data mechanisms (MCAR/MAR/MNAR), and calculates a Data Quality Score (DQS). Use it when validating data for model training, debugging pipelines, or establishing ongoing monitoring checks.

This skill provides comprehensive management for annotation schemas (configs) and human review workflows (queues) within the Arize platform. It supports defining categorical, continuous, and freeform label types. Users can programmatically apply these annotations to project spans and datasets using the Python SDK, facilitating structured human feedback loops and improving model monitoring.