Server has 7 tools with Japanese descriptions and structured schemas. All tools have input schemas with typed parameters and descriptions. However, descriptions are terse (avg ~60 chars, well below the 194-char baseline), lack WHEN/WHY context, and omit recovery guidance. Tool names follow verb_noun pattern (guide_*) but are domain-specific and lack clarity for LLMs unfamiliar with decision-tree terminology. Output schemas are partially documented in code comments but not formally exposed. Error handling exists but lacks actionable recovery steps. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels in metadata.

• Expand all tool descriptions to 100-200 chars. Include WHEN to call (e.g., 'Call after guide_start to check if the run is still active'), WHAT it returns (e.g., 'Returns current node, collected values, and path taken'), and any prerequisites.
• Add tool annotations: set readOnlyHint=true for guide_flows and guide_state; set destructiveHint=true for guide_answer, guide_revise, guide_close, guide_draft.
• Enhance error messages with recovery guidance. E.g., 'Run R-8K2M4 is closed (outcome=resolved). To continue, call guide_start() with a new flow_id and subject.'
• Document output schemas formally in tool registration. Include field names, types, and descriptions for every response object (e.g., 'Returns {run_id: string, current_node: Node, steps: Step[], values: object, next: string}').
• Add parameter descriptions for 'run_id' format flexibility: 'Run ID in format R-XXXXX (case-insensitive, hyphens optional). E.g., R-8K2M4, r-8k2m4, R8K2M4 all accepted.'