What problem does it solve?
AI agents typically execute tasks without learning from past outcomes, repeating mistakes and missing opportunities to improve quality over time. This Skill adds a self-optimizing learning layer that records every task execution and continuously improves agent performance without catastrophic forgetting.
Core Features & Use Cases
- Adaptive Learning: Learns from every task execution using SONA (Self-Optimizing Neural Architecture), delivering up to +55% quality improvement with sub-millisecond learning overhead.
- LoRA Fine-Tuning & EWC++: Achieves 99% parameter reduction and 10-100x faster training while EWC++ continual learning prevents catastrophic forgetting of earlier knowledge.
- Pattern Discovery & LLM Routing: Retrieves k=3 similar patterns at 761 decisions/sec to apply learned strategies, and automatically routes to cost-effective models for up to 60% cost savings.
- Use Case: An engineering team running repeated code-generation tasks uses the pre-task and post-task hooks so the agent records each outcome, discovers recurring patterns, and steadily improves code quality across sessions.
Quick Start
Invoke the agent-sona-learning-optimizer skill and run the pre-task hook with npx claude-flow@alpha hooks pre-task before your task, then record the outcome with the post-task hook.
