Viche provides 8 tools with complete JSON schemas and descriptions. Naming follows verb_noun convention (viche_discover, viche_send, viche_reply, viche_broadcast, viche_join_registry, viche_leave_registry, viche_list_my_registries, viche_whoami). All tools have descriptions (avg 120 chars, within 10-1024 range). All parameters have type definitions and descriptions. However, descriptions lack actionable context about when to use each tool vs alternatives, error recovery paths, and prerequisites. Output schemas are not documented, callers cannot predict response structure. Error handling guidance is absent. Parameter descriptions are minimal (e.g., 'Message content' for body, 'Agent ID (UUID format)' for to). No examples of chaining IDs across tools or pagination guidance for discovery results.

• Document output schemas for all tools. Example: 'viche_discover returns {agents: [{id: UUID, name: string, capabilities: string[], status: online|offline}], total: number, next_cursor?: string}'. This enables LLMs to extract agent IDs for downstream viche_send calls.
• Add error recovery hints to parameter descriptions. E.g., viche_send body: 'Message content (max 4096 chars). If agent is offline, message queues for delivery. Retry is safe, messages are idempotent by agent_id + timestamp.'
• Clarify tool selection in descriptions. viche_send: 'Send a message to a specific agent by ID. Use viche_discover() first to find agent IDs.' viche_reply: 'Reply to an agent that sent you a task. Only use when responding to an inbound message.'
• Add pagination to viche_discover. Include limit (default 20, max 100) and next_cursor parameters. Return total count so LLMs know if more results exist.
• Document idempotency and side effects. E.g., viche_broadcast: 'Broadcasts to all agents in registry. Idempotent, calling twice with same body sends two separate messages (not deduplicated). Use for announcements, not critical commands.'