Agent Skills Hub

vibe-agents

Generate AGENTS.md and AI configuration files (Cursor, Claude, Gemini, Copilot) for your project to streamline your vibe-coding workflow and maintain context across sessions.

training-data-curation

Guidelines for curating high-quality datasets for LLM post-training (SFT/DPO/RLHF), covering data formats, quality filtering, and collection strategies.

update-llm-model-list

Audit and synchronize the supported LLM model list in assets.py against the authoritative litellm registry.

xhs-writing-coach

AI-powered coach for Xiaohongshu (XHS) note writing. Generate viral, platform-optimized content with storytelling templates, engagement hacks, and automated compliance tagging.

langgraph

Expert LangGraph architect skill for designing stateful, multi-actor AI agent workflows with robust persistence, conditional branching, and ReAct patterns.

gemini-api

Google Gemini Image Generation API interface for text-to-image, editing, style templates, and automated retry workflows.

bdg

A terminal-based Chrome DevTools Protocol client designed for AI agents. Provides direct, session-persistent control over browser navigation, DOM manipulation, scraping, and network inspection.

claude-opus-4-5-migration

Migrate your codebase, prompts, and API calls from Claude Sonnet 4.0/4.5 or Opus 4.1 to the advanced Opus 4.5 model with automated configuration adjustments.

context-driven-testing

Apply context-driven testing principles to adapt testing strategies based on project goals, risks, and constraints rather than relying on universal best practices.

excel-auditor

Analyze and audit Excel spreadsheets to understand logic, identify formula errors, detect risks, and generate documentation for legacy or unknown files.

math-extractor

Extracts mathematical content like definitions, theorems, and proofs from documents (PDF, MD, TEX, TXT) using AI-based cleaning and conversion.

massive-context-mcp

Process massive files and large codebases (10M+ tokens) by recursively chunking, sub-querying, and aggregating results to overcome LLM context limits.