AxonFlow SDK exposes 17 tools, but most lack visible input schemas in the source code provided. Only 2 tools (mcpCheckInput, mcpCheckOutput) show explicit schema definitions in governed-tool.ts. The remaining 15 tools are referenced in example files without schema visibility, making them inferred rather than directly verifiable. Descriptions are present but generic (avg ~80 chars, below the 194-char baseline). No output schemas documented. Error handling is minimal, PolicyViolationError is raised but lacks recovery guidance. Tool naming follows verb_noun convention (good), but many tools conflate multiple concerns (e.g., proxyLLMCall handles both policy enforcement AND optional LLM routing). Security concern: userToken and credentials passed as parameters rather than server-side injection.

• Add explicit JSON Schema definitions for all 15 tools currently lacking visible schemas. Include type, description, and constraints for every parameter (e.g., enum for operation: 'execute'|'query').
• Move userToken, credentials, and tenant_id from tool parameters to server-side configuration via environment variables or vault. Implement server-side authentication and pass only opaque session/request IDs to tools.
• Document output schemas for all tools. Specify field names, types, and whether results are paginated. Include example responses showing the structure LLMs should expect.
• Split proxyLLMCall into separate tools: checkPolicyCompliance (input validation only) and routeLLMCall (routing logic). This enables LLMs to compose them independently.
• Enhance error responses with recovery guidance. Replace generic PolicyViolationError with structured errors: {allowed: false, block_reason: '...', suggested_action: 'retry_with_different_params|contact_admin|use_alternative_tool'}.