Loading ToolBench
Analyzes sales call transcripts to detect objections, assess deal health, calculate win probability, and recommend next best actions for revenue intelligence
Single tool 'execute_skill_action' has a reasonable description (118 chars, within 10-1024 range) and two parameters with types and defaults. However, the input schema is inferred from example_usage.py rather than explicitly visible in mcp_server.py. The mcp_server.py file contains only a stub that prints JSON metadata, no actual tool registration, schema definition, or MCP protocol implementation. Parameter descriptions are present but minimal (e.g., 'The full transcript text' lacks guidance on format, length, or preprocessing). Output schema is completely undocumented, the client.py mock returns a dict with fields like 'detected_objections', 'deal_health_score', 'win_probability_pct', but no schema is declared to the MCP protocol. No error handling, validation, or recovery guidance. Tool does not follow verb_noun naming (should be 'analyze_sales_call' not 'execute_skill_action'). No tool annotations (readOnlyHint, destructiveHint, idempotentHint).
Analyzes sales call transcripts to detect objections, assess deal health, calculate win probability, and provide next best action recommendations
Tool definition not explicitly registered in MCP server code. mcp_server.py contains only a stub that prints metadata; no actual MCP tool registration, schema binding, or protocol handler visible. Tool existence and schema inferred from example_usage.py and client.py mock.
Output schema completely undocumented. client.py returns a dict with 'detected_objections' (array of objects with 'category', 'quote', 'risk_level'), 'deal_health_score' (number), 'win_probability_pct' (number), 'next_best_action' (string), but no schema is declared to MCP clients. LLMs cannot plan downstream operations or validate responses.
Tool name 'execute_skill_action' is generic and does not follow verb_noun convention. Should be 'analyze_sales_call' to clearly signal the action (analyze) and resource (sales call). Current name is vague and forces LLMs to read description to understand intent.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 25 | 2026-07-28+ | v2 |
Parameter descriptions lack actionable detail. 'The full transcript text of the sales call' does not specify format (plain text, JSON, structured), length limits, preprocessing requirements, or language. 'The deal size in USD' lacks range constraints (min/max).
No error handling or recovery guidance. If transcript is malformed, deal_size is negative, or analysis fails, no error schema or actionable message is defined. LLMs receive no guidance on retry, fallback, or user correction.