3. Data, Science & AI

Migrates AI image generation from Google Gemini 2.5 Flash to BytePlus SeeDream v4.5, optimizing cost and image quality for developers.

Automates end-to-end dataset updates including PR creation, snapshots, and processing through meadow, garden, and grapher stages.

Enables AI assistants to access data and files through MCP resources using URIs, caching, and streaming protocols.

Generates images via ModelScope's Z-Image-Turbo AI model for user requests of pictures, illustrations, or visual content.

Extracts structured website data via Firecrawl API and converts pages to LLM-ready markdown for AI applications.

Generates optimized prompts for AI interactions using templates, argument handling, and context injection to improve response quality and task alignment.

Enables CLI-based delegation workflows for Google's Gemini AI models, implementing core task distribution logic for model interaction.

Provides a CLI workflow for delegating tasks using Alibaba's Qwen models via delegation-core implementation.

Delegates tasks to external LLM services (Gemini, Qwen) with built-in quota management, logging, and error handling for seamless integration.

Provides a multi-LLM framework for strategic decision-making through simulated expert consultation under pressure.

Creates or updates skills to capture new knowledge, enabling Claude to remember patterns, workflows, and project-specific guidance.

Enforces rigorous validation steps including baseline comparisons, cross-validation, and leakage prevention in machine learning pipelines.

Provides standardized patterns for data loading, exploration, and statistical analysis to enhance data science workflows.

Enforces rigorous ML pipeline validation through baseline comparisons, cross-validation, model interpretation, and data leakage prevention.

Provides best practices for creating scientifically rigorous and reproducible experimental designs in data science and research contexts.

Provides standardized patterns for loading, exploring, and performing statistical analysis on datasets.

Provides best practices for designing reproducible experiments in data science and AI research contexts.

Creates optimized prompts for LLMs to enhance output quality and task performance in system prompts, user prompts, and few-shot examples.

Guides creation, configuration, and utilization of Claude Code subagents and Task tool for specialized AI agent workflows.

Extracts and deduplicates kernel schemas from SGLang model implementations for AI model development and flashinfer_trace integration.

Comprehensive data visualization skill for building charts, graphs, and dashboards with guidance on chart selection, layout algorithms, and library implementation.

Ensures questions are answered literally, preventing misinterpretation as instructions or criticism in conversational AI.

Provides tools for processing student databases including grade calculation, duplicate name detection, and filtering by TOEFL scores and recommendation letters.

Orchestrates between multiple LLMs (Haiku, Sonnet, Opus, Codex) to select the optimal model based on task complexity.