Continuum exposes 4 tools with minimal schema rigor. Tool definitions lack parameter type information, descriptions are generic or missing, and output schemas are undocumented. The health and open_incidents tools have empty input schemas ({}), which is acceptable for read-only discovery, but post_alert and select_query lack proper parameter validation guidance. Descriptions are present but fall short of LLM-optimized standards (10 - 1024 chars, with context on WHEN and WHY to use each tool). No error handling guidance, no idempotent hints, no permission declarations. The server appears to be a thin HTTP wrapper around incident orchestration logic without agent-centric tool design.

• Add enum constraints to post_alert.severity (e.g., 'critical', 'high', 'medium', 'low') and document valid AWS regions for the region parameter.
• Document output schemas for all tools. Example: post_alert should return {incident_id, status, created_at, correlation_id}. open_incidents should return {incidents: [{id, service, severity, status, created_at}], total_count, next_cursor}.
• Expand select_query description to include: max result limit (e.g., 1000 rows), pagination guidance, and when to use open_incidents instead (for pre-built queries vs ad-hoc SQL).
• Add error handling guidance: 'If correlation_id is not unique, post_alert returns 409 Conflict. Retry with a new correlation_id or call open_incidents to check for duplicates.'
• Declare tool permissions and idempotency hints. Mark post_alert as destructive/write; mark health and open_incidents as read-only and idempotent.

Inferred effective spec: 2025-06-18+.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry