[akhilsinghcodes]

Summary

• New Analytics tab: per-session practice scorecards across 4 categories (Prompt Quality, Session Hygiene, Code Review, Tool Mastery), backed by all 45 detection rules ported from microsoft/AI-Engineering-Coach
• Parses the agent's own ~/.claude/projects / ~/.codex logs, scoped to the exact AgentFleet session time window (not the repo's full history), and runs them through the rule engine to surface anti-patterns (severity, occurrences, description, suggestion) plus per-category and overall practice scores
• New packages/ai-coach-analytics workspace package (parsers, rule engine, 45 rule + 10 metric definitions) and session_analytics table (additive only — no changes to existing tables/routes)
• Backfill script (apps/server/scripts/backfill-analytics.ts) to analyze historical stopped sessions that predate this feature
• Fixed a session-delete regression (FOREIGN KEY constraint failed) introduced by the new session_analytics table
• Docs: AI_COACH.md explains the scoring formula and all 45 rules by category; NOTICE updated with attribution

Test plan

• pnpm -r typecheck passes
• pnpm -r lint passes
• pnpm --filter @agents_fleet/server test passes (15/15)
• Backfilled 13 historical sessions; confirmed distinct, non-uniform scores/anti-patterns per session via direct SQLite query (e.g. one session scored 92 with a real "Auto Model Avoidance" anti-pattern, 36 occurrences)
• Verified session delete no longer throws FK errors
• Manually click through Analytics tab in the browser for a few real sessions