Глубокая интеграция agentic-flow в Claude Code

integration

v3-integration-deep is a Claude Code skill that transforms claude-flow from a parallel implementation into a specialized extension of agentic-flow@alpha following ADR-001, eliminating over 10,000 duplicate lines of code. The skill structures migration across three phases: building an adapter layer with backward compatibility, replacing SwarmCoordinator (800+ lines), AgentManager (1,736+ lines), and TaskScheduler (500+ lines) with native agentic-flow equivalents, then cleaning the codebase down to under 5,000 lines. It integrates five SONA learning modes (real-time, balanced, research, edge, batch), Flash Attention delivering 2.49x–7.47x speedup with 50–75% memory reduction, and AgentDB using HNSW indexing for 150x–12,500x search improvement. The skill also wires in 213 pre-built MCP tools, 19 hook types, and eight RL algorithms including PPO, DQN, MCTS, and Decision-Transformer. It targets engineering teams consolidating multi-agent systems onto a unified architecture while preserving compatibility with existing claude-flow APIs.