Loading ToolBench
An MCP server that implements an approval-first agent operating system for executing skill actions with human approval workflows
Critical failures across all dimensions. The single tool 'execute_skill_action' has a vague, generic description (56 chars) that does not explain WHAT the tool does, WHEN to use it, or what it returns. Input schema is present but severely underdocumented: 'action_payload' lacks type specificity (generic 'object'), and 'approval_required' has no description of its semantic meaning or side effects. No output schema is documented. The tool name uses a generic verb ('execute') rather than a domain-specific action. No error handling guidance, no recovery paths, no validation rules. The implementation in mcp_server.py is a stub that prints JSON and exits, it does not actually implement the MCP protocol (no request/response loop, no tool invocation handling). This is not a functional MCP server.
Execute a skill action with approval workflow
Tool name 'execute_skill_action' is generic and does not convey domain intent. Verb 'execute' is vague, does it deploy, approve, queue, or run? Should be domain-specific (e.g., 'approve_skill_action', 'queue_skill_execution', 'submit_skill_for_approval').
Tool description 'Execute a skill action with approval workflow' (56 chars) is too vague. Does not explain: What is a 'skill action'? When should the LLM call this vs. other tools? What does it return? What are the side effects? Minimum 100 chars with explicit WHAT/WHEN/RETURN guidance required.
Input parameter 'action_payload' has type 'object' with no schema details. What fields does it require? What are valid keys? LLM cannot construct valid payloads. Must define nested schema with required fields, types, and descriptions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 20 | 2026-07-28+ | v2 |
Parameter 'approval_required' has no description. Does setting it to false bypass human approval? What are the security implications? Does it affect audit logging? Description is mandatory for all parameters.
No output schema documented. What fields does the tool return? What is 'approval_token'? What does 'execution_status' contain? LLM cannot plan downstream calls or extract required data without documented return structure.