k8s-ariadne-rs presents a focused, domain-specific tool set for Kubernetes graph querying. All three tools have clear, substantive descriptions (ranging from 160-190 characters) and well-structured input schemas with proper JSON Schema typing. Tool naming follows verb_noun patterns (graph_query, graph_schema, graph_health). However, some parameter descriptions lack depth, output schemas are not explicitly documented in the code samples provided, and there is no evidence of error recovery guidance or actionable error messages. The server is semantically clean and well-scoped but lacks some patterns from the 54-pattern baseline around error handling, composition guidance, and output documentation.

• Document the output schema for each tool explicitly. For graph_query, specify: {rows: array of objects, columns: array of strings, row_count: integer, truncated: boolean, execution_time_ms: number}. For graph_schema, specify whether 'compact' returns a text string or structured JSON. For graph_health, define the schema for each detail level (compact, full, debug).
• Expand parameter descriptions to explain purpose and trade-offs. Example for graph_schema: 'format="compact" returns a text summary optimized for token efficiency; format="structured" returns the full machine-readable graph schema with all node labels, property definitions, and relationship types, use structured only when you need to understand traversal directions or validate Cypher syntax.'
• Add error classification documentation to graph_query. Document error kinds (e.g., 'syntax_error', 'timeout', 'resource_not_found', 'permission_denied') and provide recovery guidance: 'If kind=syntax_error, check Cypher syntax and call graph_schema to verify label and property names. If kind=timeout, narrow your query with LIMIT or add WHERE filters.'
• Clarify the semantics of the 'limit' parameter and 'truncation status' field. State explicitly: 'limit caps the response after query execution for transport safety. If the Cypher query returns 500 rows and limit=100, only 100 are returned; truncated will be true. To avoid truncation, add LIMIT to your Cypher query or increase the limit parameter (max 1000).'
• Add composition hints. E.g., 'Call graph_schema first if you are uncertain about labels, properties, or relationships. Then use graph_query with a narrowly scoped Cypher query to explore specific data.' This guides multi-step agent planning.

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

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