Agent Skills Hub

sparc-methodology

SPARC methodology for multi-agent development: systematic Specification, Pseudocode, Architecture, Refinement, and Completion workflows via Claude Flow orchestration.

claude

Anthropic Claude AI models for high-performance coding, large-context analysis, and GUI interaction.

xp-practices

Implement Extreme Programming (XP) practices including TDD, pair programming, and continuous integration to enhance team collaboration and technical excellence in software engineering.

skill-creator

A framework for building modular, reusable agent skills. Provides guidelines for structuring SKILL.md, bundled scripts, references, and assets to extend Claude's capabilities.

debug-distributed

Debugging guide for AReaL distributed training issues, including hangs, NCCL errors, OOM, and numerical consistency in FSDP2/TP/CP/EP.

backend-rag-implementation

Build RAG systems to ground LLMs in proprietary data. Includes vector database integration, embedding strategies, hybrid search, and advanced retrieval patterns for FastAPI backends.

context-degradation

Diagnose, isolate, and mitigate LLM context failures like lost-in-middle, poisoning, distraction, and context clash to improve agent reliability.

skill-memory

A Git-backed memory store for agent skills. Download, version, edit, and share custom agent behaviors and procedural knowledge using a CLI.

ultraqa

Autonomous QA cycling workflow that runs test-verify-fix loops until your quality goals are met.

flow-nexus-neural

Train and manage neural networks in distributed E2B sandboxes using the Flow Nexus platform, supporting custom architectures like Transformers, LSTMs, and GANs.

vvm

VVM (Vibe Virtual Machine) is a language for agentic programs where the LLM acts as the runtime. Orchestrate multi-agent workflows, manage state, and build resilient AI pipelines.

log-debug-issue

Maintain a structured DEBUG_LOG.md for recording bugs, debugging processes, and solutions to ensure project stability and knowledge retention.