Add RevenueAlertEscalationSkill - revenue anomaly detection with incident escalation

[lbartoszcze]

Summary

• RevenueAlertEscalationSkill - Revenue-specific anomaly detection and auto-incident escalation ([AGENT] Add CONTRIBUTING.md #1 priority from session 200 MEMORY: "Revenue Alert Escalation")
• Bridges RevenueObservabilityBridgeSkill → IncidentResponseSkill for autonomous revenue monitoring
• 5 default alert rules: revenue_drop (>30% baseline drop), zero_revenue (critical), low_success_rate (<80%), no_customers, revenue_spike (>200% increase)
• 6 actions: check (evaluate metrics against rules), rules (CRUD for alert rules), status (active alerts), history (audit trail), configure (escalation behavior), health (revenue health dashboard with score and recommendations)
• Rule engine: configurable conditions (below, equals, above, drop_percent, spike_percent) with severity levels and cooldown periods
• Baseline calculation: moving average from last N snapshots for trend-based anomaly detection
• Auto-escalation: creates incidents via IncidentResponseSkill when alerts fire, auto-resolves when metrics return to normal
• Health dashboard: 0-100 health score with severity-weighted penalties, trend analysis, and actionable recommendations
• Persistent JSON state: alert rules, active alerts, metric snapshots, incident links, operation stats
• Registered in autonomous_agent.py DEFAULT_SKILL_CLASSES
• 16 new tests, all passing. 17 smoke tests passing.

Pillar: Revenue (primary), Self-Improvement (supporting)

This closes the critical gap between revenue monitoring and incident response. Previously, revenue drops went unnoticed until manual inspection. Now the agent autonomously monitors its own revenue health, detects anomalies via configurable rules, auto-creates incidents for the response team, and provides a health dashboard with actionable recommendations. This is the "immune system" for the agent's revenue pipeline.

Test plan

• 16 unit tests covering all 6 actions
• Tests for: empty metrics, healthy metrics, zero revenue alerts, low success rate, alert resolution, rule CRUD, history filtering, configuration, health scoring
• 17 smoke tests still passing

🤖 Generated with Claude Code