Loading ToolBench
MCP server for routing and resolving customer conversations across multiple channels (Instagram DM, WhatsApp, etc.) with omnichannel commerce and helpdesk capabilities
Critical failures across all dimensions. The single tool 'execute_skill_action' has a vague, action-stacking name that violates verb_noun convention. Description is 156 chars but lacks WHEN/WHY guidance and does not explain the tool's actual behavior (routing, intent detection, checkout link generation are buried in example output, not the description). Input schema exists but parameters lack descriptions entirely, 'conversation_channel', 'customer_handle', 'message_text' have no guidance on format, constraints, or valid values. No output schema documented. No error handling, no recovery guidance. The tool conflates multiple concerns (routing, intent detection, commerce operations) into one monolithic action, violating single-responsibility principle. Code is a stub (prints JSON, no actual implementation), making it impossible to verify real behavior.
Route and resolve customer conversations across multiple channels with intent detection and omnichannel commerce capabilities
Tool name 'execute_skill_action' is generic and action-stacking. Does not start with a clear verb (get_, create_, send_, etc.). Violates verb_noun naming convention. LLM cannot infer intent from name alone.
Input parameters lack descriptions entirely. 'conversation_channel', 'customer_handle', 'message_text' have no guidance on format, constraints, enums, or valid values. LLM cannot determine what values to pass.
No output schema documented. Example output in client.py shows 7 fields (ticket_id, channel, intent_detected, auto_generated_checkout_link, stock_available_units, ai_assisted_first_response_time_seconds, omnichannel_customer_unified_profile) but no formal schema definition. LLM cannot plan downstream calls or extract data reliably.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 29 | 2026-07-28+ | v2 |
Tool description does not explain WHEN to use it, what it modifies, or prerequisites. 156 chars is adequate length but content is vague. Does not state that this tool performs writes (creates tickets, generates checkout links), which is critical for agent planning.
Tool conflates multiple responsibilities: routing, intent detection, inventory lookup, checkout link generation, and customer profile enrichment. Should be split into separate tools (route_conversation, detect_intent, get_inventory, generate_checkout_link, get_customer_profile) so agent can compose them independently.