What's New
Closes the analytics gaps exposed by Claudoscope while protecting Token Optimizer's moat (quality scoring, smart compaction, drift detection, coaching).
Per-Turn Token Breakdown
Click any session in the dashboard to see input/output/cache per API call. Spike detection highlights context jumps. Cache values are dimmed when repeated (prompt caching working correctly) and highlighted when they change.
Cost Per Session
Every session now shows estimated API cost. Daily totals in the trends view. Supports four pricing tiers: Anthropic API, Vertex AI Global, Vertex AI Regional (+10%), AWS Bedrock.
Session Quality Overlay
Color-coded quality scores on every session row. 5-signal scorer (context fill, message count, cache hit rate, output/input ratio, API efficiency). Green = healthy, yellow = degrading, red = trouble.
Cache Visualization
Stacked bars showing input vs output vs cache-read vs cache-write split. Global cache hit rate stat card.
Kill Stale Sessions
Terminate zombie headless Claude Code processes from the dashboard or CLI with explanation text.
New CLI Commands
measure.py conversation [session-id]- Per-turn token breakdownmeasure.py pricing-tier [tier]- View/set pricing tiermeasure.py kill-stale [--dry-run]- Kill sessions running >12h
Review Squad Verified
Architecture, security, and code simplicity reviews passed. Dead TypeScript code removed. Opus pricing corrected. PID sanitization added.
Full Changelog: v2.5.0...v2.6.0