Strong tool definitions with comprehensive schemas and clear descriptions. All 10 tools have explicit input schemas with typed parameters and descriptions. Tool names follow verb_noun convention (dispatch, status, cancel, list, register, etc.). Most descriptions are 50-200 chars, meeting LLM-optimization baselines. However, some parameter descriptions lack constraint details (e.g., 'radius_km' max 20000 not stated in description, only in schema). Output schemas are not documented, LLMs cannot infer response structure. Error handling guidance is absent. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk classifications in metadata.

• Add output schemas to all tools. Document response structure: task_id (string), status (enum: pending|in_progress|completed|failed|cancelled), created_at (ISO 8601), updated_at (ISO 8601), proof_urls (array), result (object). This enables LLMs to chain tools and extract required fields.
• Add tool annotations: readOnlyHint=true for status/list/backends/get_provider tools; destructiveHint=true for remove_provider; idempotentHint=true for dispatch_task (via idempotency_key). Annotations guide agent planning and prevent accidental misuse.
• Move constraint details from schema into parameter descriptions. E.g., 'radius_km: Acceptable radius in kilometers from the specified point (1-20000, default 5)' instead of relying on schema maxItems.
• Add recovery guidance to error scenarios. E.g., 'If task_id not found, call human_dispatch_list with status=pending to discover active tasks.' This enables agents to self-correct.
• Document pagination in list tools. Add 'Returns up to {limit} results. If total > limit, use offset to fetch next page. Response includes total_count and has_more fields.' This clarifies how to iterate large result sets.