v1.9.0 — AI Phase 3: model-routing, cost-optimization, skill updates

AI Phase 3 — Complete

New Skills

• model-routing — task-based routing table, cascade escalation (haiku→sonnet→opus), content-based routing, fallback chains, routing metrics. 87% cost reduction potential.
• llm-cost-optimization — prompt caching (~42% reduction), output token reduction, batch API (~50% discount), per-workflow budget tracking, application-level caching

Updated Skills

• ai-harness — statistical rigor (t-test, Cohen's d, minimum sample sizes), dataset maintenance (stale detection, production failure ingestion)
• ai-product-design — agent UX patterns (progress display, transparent tool use), citation/attribution for RAG content, accessibility (ARIA live regions, mobile, RTL/i18n)

Agent updates

• @ai-engineer now includes llm-cost-optimization + model-routing
• @ai-product now includes model-routing

---

AI Skills Complete Summary (v1.7.0 → v1.9.0)

Skill	Type	Description
agentic-eval	NEW	Trajectory eval, pass@k, tool hallucination detection
agentic-security	NEW	Prompt injection, sandboxing, audit logging
agentic-memory	NEW	4-tier memory architecture, context management
agentic-reliability	NEW	Retry/circuit breaker/fallback for agents
model-routing	NEW	Cost-aware model selection and cascade
llm-cost-optimization	NEW	Caching, token reduction, batch API
agentic-ai-patterns	UPDATED	+LangGraph, memory arch, human-in-loop
ai-safety-guardrails	UPDATED	+Jailbreak, bias/fairness, EU AI Act
llm-evaluation	UPDATED	+Ragas, latency/cost, stratified eval
llm-observability	UPDATED	+Drift detection, agent metrics, OTel
prompt-engineering	UPDATED	+Tool use, extended thinking, multi-turn
rag-architecture	UPDATED	+Hybrid search, HyDE, agentic RAG
ai-harness	UPDATED	+Statistical rigor, dataset maintenance
ai-product-design	UPDATED	+Agent UX, citations, accessibility