Phase 7 — Observability & Tier 3
April 2026
Agent-level LLM cost + token observability, OWASP MCP Top 10 security scanning, Tier 2 API-benchmark runner, Tier 3 OSS-agent sandbox foundations, public API v1 scaffolding.
- OWASP scan pipeline + supply-chain CVE tracking via OSV.dev
- MetrxBot-powered internal LLM cost observability (every BenchLytix LLM call emits a structured event)
- Tier 2 API-benchmark runner with AES-256-GCM credential encryption + SSRF guard
- Tier 3 OSS sandbox: Railway runtime + per-run scoped LLM keys (flag-off, soaking)
- Community review UI with helpfulness voting + tier badges
- NL buyer search scaffolding (Haiku + Sonnet, flag-off)
- Agent claiming flow + admin moderation queue