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
- RETRIEVE — Fetch relevant patterns via HNSW (150x-12,500x faster)
- JUDGE — Evaluate with verdicts (success$failure)
- DISTILL — Extract key learnings via LoRA
- CONSOLIDATE — Prevent catastrophic forgetting via EWC++
Components
Commands
Train Patterns
Check Status
View Patterns
Predict
Optimize
Best Practices
- Use pretrain hook for batch learning
- Store successful patterns after completion
- Consolidate regularly to prevent forgetting
- Route based on task complexity
