Loading ToolBench
Runtime MCP proxy, x402 receipt ingestion, and adaptive spend firewall for wallet-enabled AI agents.
mythos-sentinel defines 11 tools with complete input schemas and descriptions. Naming follows verb_noun convention (sentinel_scan_path, sentinel_check_*). Descriptions are present and substantive (avg ~120 chars), explaining WHAT each tool does. However, parameter descriptions are sparse or missing context about constraints, formats, and error recovery. Output schemas are not documented, LLMs cannot predict return structure. Error handling guidance is absent; tools do not indicate what to do on failure. Security-sensitive tools (scan_path, check_command, check_file) lack explicit permission gates or audit trail documentation. No tool annotations (readOnlyHint, destructiveHint) despite all tools being read-only operations.
Check whether a shell command is allowed, blocked, or requires approval by the local Sentinel policy.
Check whether an agent can read or write a file path under the local Sentinel policy.
Check whether network access to a domain is allowed by the local Sentinel policy.
Check whether an x402/Base payment is allowed by the local Sentinel policy.
List supported RouteScore service categories and aliases for routing paid agent APIs.
Normalize an x402 payment receipt or payment-response payload without storing prompts, responses, secrets, or wallet balances.
Output schemas not documented. LLMs cannot predict return structure (fields, types, pagination). Forces agents to guess what data is available after each call.
Parameter descriptions lack constraint details. 'domain' parameter has no format guidance (FQDN? IP?). 'amountUSDC' has no min/max bounds. 'operation' enum is present but not described in the parameter itself.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2025-06-18+ | v2 |
Recommend a seed x402/Base paid API by category, price, and RouteScore before an agent spends money.
Select the best known x402 service plus fallback services by category, price, query, and local RouteScore telemetry.
Scan an agent skill, MCP server, repository, or folder for risky commands, prompt injection, secrets, wallet access, and policy violations.
Return the seed catalog score for a payment domain if Sentinel knows it.
Create a hash snapshot of a workspace before an agent changes files.
No error handling guidance. Tools do not indicate what to do on failure (retryable? user-fixable? fatal?). No recovery suggestions (e.g., 'If policy not found, use default policy' or 'If domain not recognized, try sentinel_recommend_x402_service').
Tool annotations missing. All 11 tools are read-only (no side effects), but none declare readOnlyHint=true. This prevents LLMs from optimizing retry logic and understanding safety guarantees.
sentinel_list_service_categories has empty inputSchema properties. No parameters documented, making it unclear what filtering or discovery options exist.