Loading ToolBench
SupraWall security gateway as an MCP server. Provides policy evaluation, approval workflows, and audit logging for AI agent tool calls.
Three tools with basic definitions but significant gaps. check_policy and request_approval have reasonable descriptions (80-120 chars), but log_action is underdeveloped (60 chars). All three tools have input schemas with type definitions, but parameter descriptions are minimal or missing context. No output schemas documented. Error handling guidance absent. Naming follows verb_noun pattern (check_, request_, log_) but lacks clarity on when to use each vs. alternatives. No evidence of idempotency, pagination, or composition patterns.
Use this tool BEFORE executing any sensitive or potentially dangerous action. It checks if the intended action complies with the organization's security policies. Input the action and context. It returns ALLOW, DENY, or REQUIRE_APPROVAL.
Use this tool to record significant actions to the secure audit trail. Call this AFTER successfully performing an action, or when an action fails due to policy violations.
Use this tool when check_policy returns REQUIRE_APPROVAL or when you suspect an action is highly sensitive. It pauses your workflow to ask a human operator for permission via Slack/Email.
Output schemas not documented. LLMs cannot plan downstream calls or extract return values. check_policy returns ALLOW/DENY/REQUIRE_APPROVAL but structure is not specified.
log_action description (60 chars) is below 10-1024 baseline and lacks context. Does not explain when to call it, what happens on failure, or whether it blocks execution.
Parameter 'context' in check_policy is typed as object but has no schema constraints or description of expected fields. LLMs cannot infer what context keys are valid.
No error handling guidance. Tools do not document what errors are retryable, user-fixable, or fatal. No recovery hints (e.g., 'If policy check fails, try request_approval').
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
Tool composition unclear. No documentation of expected call sequence (check_policy → request_approval → log_action). LLMs must infer workflow from names alone.