HEngram defines 5 tools with complete JSON schemas and descriptions. All tools have proper input schemas with typed properties and required fields. Descriptions are present (avg ~100 chars) but lack depth on WHEN to use each tool and error recovery guidance. Parameter descriptions are minimal (e.g., 'Domain category for the information' for 'domain' param). No output schemas documented. No error handling guidance. Tools follow single-responsibility principle (memorize, search, link, recall, patch) but lack composition hints (e.g., which tools chain together). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear read/write semantics.

• Add output schemas to all tools. E.g., memorize should document: {node_id: string, created_at: ISO8601, embedding_dim: number}. This enables agents to extract node_id for downstream link/patch calls.
• Convert node_type parameter to an enum constraint in the schema: enum: ['fact', 'preference', 'project', 'person', 'decision', 'constraint', 'event', 'concept', 'procedure', 'summary']. Update description to reference the enum.
• Add tool annotations: mark search/recall with readOnlyHint=true; mark memorize/link/patch with destructiveHint=true. This signals to agents which operations are safe to retry.
• Expand tool descriptions to include error recovery guidance. E.g., 'If node not found, try search() first to locate the node_id. If link fails due to duplicate relationship, use patch() to update the existing link.'
• Document composition chains in descriptions. E.g., memorize description: 'Returns node_id. Use this ID in link(), recall(), or patch() for follow-up operations.'