Omni-NLI provides 2 tools with clear verb-based naming and documented schemas. Both tools have functional descriptions and complete input parameter schemas with types and descriptions. However, output schemas are not explicitly documented, parameter descriptions could be more detailed with constraints, and error handling patterns are absent. The server follows basic tool design patterns but lacks production-grade robustness. Tool names are appropriately specific (evaluate_nli, list_providers) and start with action verbs. Input schemas are well-formed JSON Schema with proper type declarations and field descriptions. The main gaps are: (1) no documented output schemas visible in the source, (2) limited error recovery guidance, (3) missing enum constraints for some parameters that could validate early, and (4) no indication of result pagination or scaling behavior for the analyze tool.

• Document output schemas for both tools. For evaluate_nli: specify that it returns {relationship: string, reasoning: string, confidence: float}. For list_providers: specify the shape of the provider list (e.g. [{name: string, status: string, config: object}]).
• Add error recovery guidance: 'If backend connection fails, try list_providers() to verify configuration status.' This helps LLMs self-correct.
• Enhance parameter descriptions with explicit constraints: 'premise (1-4096 chars): The base factual statement to analyze. Plain text or natural language preferred.' Repeat the minLength/maxLength from schema in human-readable form.
• Add a 'Retry Policy' note to evaluate_nli description: 'This tool is idempotent, calling it multiple times with the same inputs produces identical results. Safe to retry on transient failures.'
• Consider adding a dry_run parameter to evaluate_nli: 'If true, validates inputs and returns analysis structure without making external API calls. Useful for planning complex analyses.'
• Provide model discovery: Either add an enum of valid models per backend (if known), or document a pattern: 'To see available models, call list_providers() which returns supported models per backend.'
• Add composition guidance in evaluate_nli description: 'Use this after using list_providers() to select a backend/model. Returns relationship (entailment|contradiction|neutral) suitable for decision logic in downstream tools.'

Inferred effective spec: <=2025-11-25.

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