Skill UI

Combines a curated Japanese NLP resource dataset with live web research to produce an up-to-date trend report on models, tooling, benchmarks, embeddings, speech, and related subfields.

A specialized skill for optimizing content to be cited as authoritative sources by major AI language models (ChatGPT, Perplexity, Claude, Gemini). Unlike traditional SEO, AEO focuses on E-E-A-T signals, factual density, and structured data to ensure LLM integration. Features include content auditing, E-E-A-T scoring, structural rewriting, and citation tracking.

This skill implements Autonomous Optimization (AO) using Hypothesis Tree Refinement (HTR). It iteratively improves concrete artifacts (like code, models, or data pipelines) by running structured, long-term experiments against a measurable objective. It maintains a persistent hypothesis tree, coordinating subagents to test hypotheses in isolation while using a held-out test evaluator to prevent overfitting and ensure reliable, cumulative research progress. Ideal for complex benchmark tuning, agent refinement, or optimizing entire workflows.

This skill provides a comprehensive workflow for generating, editing, and manipulating images using Google's Gemini Nano Banana Pro model (gemini-3-pro-image-preview). It is designed for tasks such as creating visuals, generating logos, modifying photos, or creating product mockups, offering a robust, file-based approach to handle complex image data and API interactions.

This skill provides comprehensive tools to detect data poisoning and model backdooring attacks throughout the machine learning lifecycle. It checks data provenance, detects label-flip anomalies (using Cleanlab), and analyzes model activations and spectra (using ART) to identify malicious triggers or compromised weights, ensuring ML supply chain integrity.

This skill provides methods and frameworks to detect sophisticated AI attacks, including model stealing, membership inference, and model inversion. It focuses on monitoring query patterns, analyzing confidence exposure, and implementing defensive controls on inference APIs to protect intellectual property and data privacy.

This skill utilizes NVIDIA's open-source garak framework to conduct comprehensive red-teaming assessments on Large Language Models (LLMs). It tests for critical vulnerabilities such as prompt injection, jailbreaks, data leakage, and toxic content generation by sending thousands of adversarial probes. Ideal for pre-deployment security validation, API guardrail testing, and generating defensible evidence for AI risk assessments.

This tool automatically reads usage data from active AI coding sessions (JSONL format). It tallies token usage per model and tier, calculates the total cost in USD using specified pricing models, and persists a structured record into the cost-tracking namespace. Use this after completing a chunk of work or at the end of a session to provide accurate data for cost reporting and optimization.

This tool executes the advanced `metaharness/darwin` security benchmark (ADR-155). It rigorously evaluates the security detection capabilities of AI systems against a controlled ground-truth corpus, providing critical metrics like TPR, FPR, and patch-pass rates. It compares the champion model against multiple baselines, offering empirical data crucial for iterative security improvement and drift detection.

This skill connects to the public, read-only AI HOT API to retrieve the latest Chinese AI news, curated selections, hot topics, and daily reports. It is essential for users needing up-to-date information on large language models, industry breakthroughs, or major AI events, ensuring results are based on real-time data rather than pre-trained knowledge.