V3 Deep Integration

Migrates claude-flow onto agentic-flow@alpha adapters to eliminate duplicate orchestration code.

1|1|Updated Nov 28, 2025
One-click install
npx skills add https://github.com/33may/robotics --skill v3-deep-integration-33may
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: V3 Deep Integration
Source: https://github.com/33may/robotics/tree/main/humanoid/.claude/skills/v3-integration-deep
Command: npx skills add https://github.com/33may/robotics --skill v3-deep-integration-33may

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires agentic-flow.

What problem does it solve? claude-flow duplicates thousands of lines of orchestration logic (SwarmCoordinator, AgentManager, TaskScheduler) that already exist in agentic-flow@alpha, creating maintenance burden and missing performance gains from SONA, Flash Attention, and AgentDB. ## Core Features & Use Cases - Adapter Layer Migration: Extends agentic-flow@alpha Agent classes with claude-flow-specific behavior while preserving backward compatibility through a legacy compatibility layer. - Feature Integration: Integrates 5 SONA learning modes, Flash Attention (2.49x-7.47x speedup), AgentDB HNSW search (150x-12,500x), 213 MCP tools, and 19 hook types. - Phased Code Deduplication: Replaces SwarmCoordinator, AgentManager, and TaskScheduler with agentic-flow equivalents, reducing orchestration code from 15,000+ to under 5,000 lines. - Use Case: A platform engineer migrating claude-flow v2 to v3 uses this Skill to plan the adapter layer, migrate swarm and agent management systems feature-by-feature, validate parity, and remove deprecated implementations. ## Quick Start Ask the agent to design the agentic-flow@alpha adapter layer and plan the phased migration of claude-flow's swarm, agent, and task systems.

Frequently Asked Questions about V3 Deep Integration

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I migrate claude-flow to agentic-flow@alpha?

Migrate in three phases: build an adapter layer extending agentic-flow Agent classes, migrate systems (swarm, agents, tasks) feature-by-feature with parity validation, then remove deprecated implementations. Dual operation keeps the old system running during transition.

How much duplicate code does the agentic-flow integration eliminate?

The integration targets reducing orchestration code from over 15,000 lines to under 5,000. SwarmCoordinator (800+ lines), AgentManager (1,736+ lines), and TaskScheduler (500+ lines) are replaced by agentic-flow equivalents with 50-80% overlap.

What performance gains does Flash Attention integration provide?

Flash Attention integration targets 2.49x-7.47x speedup with 50-75% memory reduction. It supports multi-head, linear, local, and global attention mechanisms through the agentic-flow@alpha attention API.

Does the migration maintain backward compatibility with claude-flow v2?

Yes, backward compatibility is maintained through a legacy compatibility layer and gradual migration. The process uses dual operation, feature-by-feature migration with parity validation, and only deprecates the old system after full parity is confirmed.

What SONA learning modes are available after integration?

Five SONA modes are integrated: real-time (~0.05ms adaptation), balanced, research, edge, and batch. Each mode is activated through the agentic-flow SONA API depending on latency, resource, or throughput requirements.