/scan-tool-call — Check if a tool call is safe
Scan tool calls for safety before execution to prevent dangerous operations.
Scan a tool call for dangerous operations before execution.
Usage
/scan-tool-call <tool_name> <arguments as JSON>
Behavior
Use the Sentinel AI scan_tool_call MCP tool to:
- Parse the tool name and JSON arguments
- Check for dangerous shell commands (rm -rf, credential access, etc.)
- Check for data exfiltration patterns (curl to external servers, etc.)
- Check for privilege escalation attempts
- Report risk level and whether the call should be blocked
Present findings with:
- Risk level (NONE, LOW, MEDIUM, HIGH, CRITICAL)
- Threat type (dangerous_command, exfiltration, sensitive_file, privilege_escalation)
- Description of the threat
- Whether the tool call would be blocked
Input
$ARGUMENTS