Loading ToolBench
Data Loss Prevention (DLP) governance gateway that inspects and masks personally identifiable information (PII) in prompts before external LLM dispatch, supporting compliance policies like HIPAA, GDPR, and PCI_DSS
Single tool 'execute_skill_action' has a reasonable name starting with verb, but the implementation is severely deficient. The tool definition exists in mcp_server.py but lacks a proper MCP server implementation, it only prints a JSON status message and does not actually handle tool calls. The input schema is visible with two parameters (user_prompt, compliance_policy), but the schema is incomplete: compliance_policy is an array of strings with no enum constraints, and user_prompt lacks format/length guidance. The description is 97 characters, adequate but generic. Critically, the mcp_server.py does not implement the MCP protocol, it does not listen for tool calls, parse requests, or return structured responses. The client.py contains mock logic that always returns hardcoded results regardless of input, making the tool non-functional. No error handling, no output schema documentation, no parameter validation.
Execute DLP governance skill action to inspect and mask PII in prompts
No functional MCP server implementation. mcp_server.py only prints a status message and does not implement the MCP protocol, it does not listen for requests, parse tool calls, or return results. The server is non-functional.
Input schema incomplete. compliance_policy parameter is an array of strings with no enum constraint, LLMs will hallucinate invalid policy names. Should define enum: ['HIPAA', 'GDPR', 'PCI_DSS', 'SOC2', 'CCPA'] or similar.
No output schema documented. Tool returns a complex object with fields like 'sanitized_prompt', 'pii_entities_redacted', 'dlp_verdict', but LLMs have no formal schema to understand the response structure. Must document return type.
Mock implementation returns hardcoded results. client.py always returns the same masked entities and verdict regardless of input. Tool does not actually inspect or mask PII, it is a stub.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 37 | 2026-07-28+ | v2 |
No error handling or validation. Tool does not validate user_prompt length, compliance_policy values, or handle edge cases. No error messages guide LLM recovery.