Server has 7 tools with basic schemas and descriptions, but significant quality gaps. All tools have descriptions (good), but most are generic and lack actionable detail. Schemas are present but minimal, parameters lack type constraints, enums, and validation guidance. No output schemas documented. Error handling is absent; tools fail silently or with raw Taskwarrior errors. Naming is inconsistent: task_natural and task_raw are vague; task_smart_add mixes concerns (parsing + adding). No tool annotations (readOnlyHint, destructiveHint). Security: raw command execution (task_raw) is dangerous without input sanitization. Composition: task_natural duplicates task_raw functionality, forcing LLM to choose between overlapping tools.

• Remove task_raw or gate it behind explicit confirmation. Replace with task_advanced_query accepting a structured query object (filters, sort, format) instead of raw commands.
• Document output schemas for all tools. E.g., task_natural returns {tasks: [{id, description, priority, due, project, tags}], summary: string}. task_eisenhower returns {urgent_important: [], urgent_not_important: [], not_urgent_important: [], not_urgent_not_important: []}.
• Add parameter constraints: task_context_set.context should accept enum of known projects or a regex pattern. task_smart_add.priority should be enum [H, M, L]. task_smart_add.due should document ISO 8601 or natural language format.
• Implement error handling with recovery hints. E.g., 'Task ID 999 not found. Call task_natural with "list all" to see available tasks.' Catch Taskwarrior stderr and translate to actionable messages.
• Add tool annotations: mark task_raw, task_smart_add, task_context_set, task_ticket_sync as destructiveHint=true. Mark task_natural, task_eisenhower, task_where_am_i as readOnlyHint=true. Mark task_smart_add as idempotentHint=false (duplicate adds create duplicate tasks).