{
  "id": "page:library-ruflo",
  "_kind": "Page",
  "_file": "wiki/library/ruflo.md",
  "_cluster": "wiki",
  "attributes": {
    "nodeKind": "Page",
    "title": "Ruflo Methodology (Library)",
    "displayName": "Ruflo Methodology (Library)",
    "slug": "library/ruflo",
    "articlePath": "wiki/library/ruflo.md",
    "article": "\n# Ruflo Methodology\n\n**Source**: [ruvnet/ruflo](https://github.com/ruvnet/ruflo) by ruvnet\n**Category**: Multi-Agent Swarm Orchestration / Self-Learning Architecture\n**License**: See upstream repository\n\n## Overview\n\nRuflo v3 is a multi-agent orchestration platform deploying 60+ specialized agents in coordinated swarms with self-learning and self-optimizing architecture. It features Q-Learning-based smart routing, hierarchical agent topologies (Queen/Worker), multiple consensus protocols, and the RuVector intelligence layer for continuous self-improvement.\n\n## Core Principles\n\n- **Smart Routing** - Q-Learning router selects optimal execution path (Booster/Medium/Complex)\n- **Swarm Coordination** - Topology-aware agent deployment with anti-drift enforcement\n- **Weighted Consensus** - Queen agents have 3x voting weight; supports Raft/Byzantine/Gossip/CRDT\n- **Self-Optimization** - SONA adaptation with EWC++ anti-forgetting and ReasoningBank learning\n- **Security First** - AIDefence layer with prompt injection blocking and sandboxed execution\n- **Agent Booster** - WASM fast-path for simple transforms (352x faster, $0 cost)\n\n## Process Files\n\n| Process | File | Description | Task Count |\n|---------|------|-------------|------------|\n| Swarm Orchestrator | `ruflo-orchestrator.js` | Main pipeline: routing -> swarm -> execution -> consensus -> verify | 8 |\n| Swarm Coordination | `ruflo-swarm-coordination.js` | Topology, consensus, anti-drift, agent lifecycle | 8 |\n| RuVector Intelligence | `ruflo-intelligence.js` | Pattern extraction, ReasoningBank, SONA, knowledge graph | 8 |\n| Smart Task Routing | `ruflo-task-routing.js` | Complexity assessment, Agent Booster, Q-Learning, MoE | 7 |\n| Security Audit | `ruflo-security-audit.js` | AIDefence: injection, validation, SAST, sandbox, compliance | 8 |\n\n## Skills Catalog\n\n| Skill | Directory | Description |\n|-------|-----------|-------------|\n| swarm-orchestration | `skills/swarm-orchestration/` | Multi-agent swarm formation and coordination |\n| smart-routing | `skills/smart-routing/` | Complexity-based task routing and model selection |\n| consensus-mechanisms | `skills/consensus-mechanisms/` | Raft/Byzantine/Gossip/CRDT consensus |\n| self-optimization | `skills/self-optimization/` | SONA adaptation and ReasoningBank learning |\n| vector-memory | `skills/vector-memory/` | HNSW vector search and knowledge graph |\n| agent-booster | `skills/agent-booster/` | WASM-based instant code transforms |\n| anti-drift | `skills/anti-drift/` | Hierarchical coordination and drift detection |\n| security-hardening | `skills/security-hardening/` | AIDefence layer and sandboxed execution |\n\n## Agents Catalog\n\n| Agent | Directory | Role |\n|-------|-----------|------|\n| strategic-queen | `agents/strategic-queen/` | Long-term planning and goal setting |\n| tactical-queen | `agents/tactical-queen/` | Execution coordination and resource allocation |\n| adaptive-queen | `agents/adaptive-queen/` | Real-time optimization and adaptation |\n| swarm-coordinator | `agents/swarm-coordinator/` | Topology management and consensus |\n| coder | `agents/coder/` | Code implementation and modification |\n| tester | `agents/tester/` | Test creation and execution |\n| reviewer | `agents/reviewer/` | Code quality analysis |\n| architect | `agents/architect/` | System design and architecture |\n| security-auditor | `agents/security-auditor/` | Vulnerability detection and hardening |\n| optimizer | `agents/optimizer/` | Performance tuning and token optimization |\n\n## Architecture Layers\n\n```\nUser Layer (CLI/MCP) -> Routing Layer (Q-Learning + 8 MoE) -> Swarm Coordination (Topology + Consensus) -> Agent Execution (Queen + Workers) -> Intelligence Layer (RuVector/SONA) -> Resource Layer (Memory/Workers)\n```\n\n## Task-to-Agent Mapping\n\n| Task Type | Agents |\n|-----------|--------|\n| Bug Fix | swarm-coordinator + coder + tester |\n| Feature | swarm-coordinator + architect + coder + tester + reviewer |\n| Refactor | swarm-coordinator + architect + coder + reviewer |\n| Performance | swarm-coordinator + optimizer + coder |\n| Security | swarm-coordinator + security-auditor + coder |\n\n## Swarm Topologies\n\n| Topology | Best For | Communication |\n|----------|----------|---------------|\n| Mesh | Small swarms (<8), high collaboration | All-to-all |\n| Hierarchical | Large swarms, clear delegation | Queen-to-Workers |\n| Ring | Pipeline/sequential tasks | Neighbor-to-neighbor |\n| Star | Fan-out/fan-in patterns | Central coordinator |\n\n## Consensus Protocols\n\n| Protocol | Fault Model | Quorum |\n|----------|-------------|--------|\n| Raft | Crash faults | n/2 + 1 |\n| Byzantine | Byzantine faults | 2n/3 + 1 |\n| Gossip | Partition tolerant | Eventual |\n| CRDT | Always convergent | None needed |\n\n## Anti-Drift Mechanisms\n\n- Hierarchical coordinator checkpoints every 2 subtasks\n- Shared memory coherence validation\n- Role specialization enforcement\n- Short task cycles with bounded execution windows\n- Drift scoring with configurable threshold (default: 0.3)\n\n## Quality Gates\n\n1. **Anti-Drift Checkpoint** - Drift score < threshold (blocking)\n2. **Consensus Voting** - Weighted majority with Queen 3x (blocking)\n3. **Output Verification** - Quality score >= threshold (blocking, max 3 attempts)\n4. **Security Audit** - No critical vulnerabilities (blocking)\n",
    "documents": [
      "specialization:ruflo"
    ]
  },
  "outgoingEdges": [
    {
      "from": "page:library-ruflo",
      "to": "specialization:ruflo",
      "kind": "documents"
    }
  ],
  "incomingEdges": [
    {
      "from": "page:index",
      "to": "page:library-ruflo",
      "kind": "contains_page"
    }
  ]
}