Neural Training
Train and route neural patterns in claude-flow with SONA, MoE, and EWC++ when you are hardening a multi-agent system beyond one-off prompts.
Overview
Neural Training is an agent skill for the Build phase that trains and optimizes neural routing patterns with SONA, MoE, and EWC++ via claude-flow.
Install
npx skills add https://github.com/ruvnet/ruflo --skill neural-trainingWhat is this skill?
- Intelligence pipeline: RETRIEVE (HNSW), JUDGE, DISTILL (LoRA), CONSOLIDATE (EWC++)
- SONA self-optimizing adaptation, MoE with 8 experts, and Flash Attention speed claims in docs
- CLI entry points: neural train, status, patterns, predict, and optimize via npx claude-flow
- Explicit skip guidance when learning, routing optimization, or consolidation is not required
- MoE routing with 8 experts per skill table
- Documented pipeline steps: RETRIEVE, JUDGE, DISTILL, CONSOLIDATE
Adoption & trust: 650 installs on skills.sh; 58.5k GitHub stars; 3/3 security scanners passed (skills.sh audits).
What problem does it solve?
Your multi-agent setup repeats mistakes and cannot adapt routing or consolidate learnings without catastrophic forgetting.
Who is it for?
Advanced indie agent builders already on claude-flow who need documented train/status/predict flows for MoE and continual learning.
Skip if: Simple tasks, one-off chat jobs, or builders who do not want Node-based neural tooling or experimental agent stacks.
When should I use this skill?
Pattern learning, model optimization, knowledge transfer, or adaptive routing; skip when simple tasks, no learning required, or one-off operations.
What do I get? / Deliverables
After training and optimization commands, claude-flow exposes updated patterns and routing you can query with neural status, patterns, and predict.
- Trained or updated neural patterns via claude-flow neural train
- Status, pattern inventory, and predict outputs for routing decisions
Recommended Skills
Journey fit
Pattern training and expert routing are build-time agent infrastructure, not launch or growth work. Neural train, predict, and status commands extend the agent-tooling shelf for advanced Ruflo/claude-flow stacks.
How it compares
Specialized claude-flow neural ops—not a general fine-tuning notebook or vanilla prompt library.
Common Questions / FAQ
Who is neural-training for?
Solo builders shipping agent-heavy products who use Ruflo/claude-flow and want SONA, MoE, and EWC++ pattern training wired into their workflow.
When should I use neural-training?
During Build agent-tooling when training patterns, optimizing routing, consolidating knowledge, or running predict against new task descriptions—and skip when no learning loop is needed.
Is neural-training safe to install?
It drives local CLI neural training; review the Security Audits panel on this page and treat training data and model artifacts as sensitive.
SKILL.md
READMESKILL.md - Neural Training
# Neural Training Skill ## Purpose Train and optimize neural patterns using SONA, MoE, and EWC++ systems. ## When to Trigger - Training new patterns - Optimizing agent routing - Knowledge consolidation - Pattern recognition tasks ## Intelligence Pipeline 1. **RETRIEVE** — Fetch relevant patterns via HNSW (150x-12,500x faster) 2. **JUDGE** — Evaluate with verdicts (success$failure) 3. **DISTILL** — Extract key learnings via LoRA 4. **CONSOLIDATE** — Prevent catastrophic forgetting via EWC++ ## Components | Component | Purpose | Performance | |-----------|---------|-------------| | SONA | Self-optimizing adaptation | <0.05ms | | MoE | Expert routing | 8 experts | | HNSW | Pattern search | 150x-12,500x | | EWC++ | Prevent forgetting | Continuous | | Flash Attention | Speed | 2.49x-7.47x | ## Commands ### Train Patterns ```bash npx claude-flow neural train --model-type moe --epochs 10 ``` ### Check Status ```bash npx claude-flow neural status ``` ### View Patterns ```bash npx claude-flow neural patterns --type all ``` ### Predict ```bash npx claude-flow neural predict --input "task description" ``` ### Optimize ```bash npx claude-flow neural optimize --target latency ``` ## Best Practices 1. Use pretrain hook for batch learning 2. Store successful patterns after completion 3. Consolidate regularly to prevent forgetting 4. Route based on task complexity