AIForge exhibits significant definition quality gaps. Tool descriptions are present but often generic (e.g., 'List all available tools' for tools/list). Input schemas are visible for 11/20 tools but many lack parameter descriptions or type constraints. Critical issues: (1) tool names mix verb_noun (recommend, ingest) with slash-separated paths (tools/list, tools/call) creating inconsistency; (2) 9 tools have no visible input schema (initialize, ping, health, get_artifact, list_artifacts, get_environment, list_pending_discoveries lack explicit schema definitions in source); (3) descriptions are too brief (10-50 chars) for LLM disambiguation, 'Ping the MCP gateway server' and 'Get server health status' do not explain WHEN to use each; (4) no output schemas documented anywhere; (5) error handling guidance absent, tools return errors but do not guide recovery. The tools/call tool accepts 'arguments' as free-form object with no validation hints. Composition is reasonable (single responsibility per tool) but parameter naming inconsistencies (artifact_id vs discovery_id) and missing enum constraints on filter parameters (type, tag) invite hallucinated values.

• Add explicit input schemas to all 9 tools currently lacking them. Use JSON Schema with type, description, and enum/pattern constraints for every parameter.
• Expand descriptions to 50-200 chars. For each tool, answer: What does it do? When should the LLM call it instead of a similar tool? Example: 'Get server health status and active skill count. Call this to verify the gateway is responsive before routing tool calls.'
• Standardize tool naming to verb_noun format. Rename tools/list → list_tools, tools/call → call_tool. This matches production baselines and helps LLMs parse intent from names alone.
• Document output schemas for all tools. Specify which fields are returned, their types, and which are required for chaining (e.g., recommend returns skill_id for use in ingest).
• Add error handling guidance. When a tool fails, return actionable recovery hints. Example: 'Artifact not found. Call list_artifacts() to see available IDs, or check the artifact_id format (must be UUID).' This enables LLM self-correction.

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

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