Jolt transformation with 2 level nesting and flattening it

I am trying to convert the following input to the below output using java but not able to get the inner nested attributes at all. So from the below JSON I want to match the product class and subclass and use the correct output from the rules section. So, below, we have Home and Cutlery, I want that section to be matched from the rules and get the attributes of it.

Input

{
  "product_class": "Home",
  "product_subclass": "Cutlery",
  "rules": {
    "Home": {
      "Cutlery": {
        "is_sharp": true,
        "is_prohibited_ca": false,
        "requires_age_verification": false,
        "compliance_policy_id": "CP-1"
      },
      "Rug": {
        "is_sharp": false,
        "is_prohibited_ca": true,
        "requires_age_verification": true,
        "compliance_policy_id": "CP-4"
      }
    },
    "Lighter": {
      "Default": {
        "is_sharp": false,
        "is_hazmat_ground": true,
        "is_hazmat_air": true,
        "compliance_policy_id": "CP-210"
      }
    }
  }
}

And I have tried the bunch of JOLT specs but not getting success, I just don't get any value in attributes but empty. The latest being the following: Spec used:

[
  {
    "operation": "default",
    "spec": {
      "rules": "=@(rules)"
    }
  },
  {
    "operation": "shift",
    "spec": {
      "rules": {
        "@(1,product_class)": {
          "@(2,product_subclass)": {
            "*": "attributes"
          }
        }
      },
      "product_class": "product_class",
      "product_subclass": "product_subclass"
    }
  }
]

Expected Output

{
  "attributes" : {
    "is_sharp" : true,
    "is_prohibited_ca" : false,
    "requires_age_verification" : false,
    "compliance_policy_id" : "CP-1"
  },
  "product_class" : "Home",
  "product_subclass" : "Cutlery"
}

Actual Output:

{
  "attributes" : null,
  "product_class" : "Home",
  "product_subclass" : "Cutlery"
}

Could some one please guide on this and provide an answer. I even tried GPT and it started hallucinating without provide the correct answer! So looking for guidance some help here..

FOLLOW-UP Thank you @Barbaros Ozhen for providing the jolt spec. Now, I am using this in Java where passing the input to the method to scan the rule file accordingly.

Here is my method:

public String transform(String inputJson) throws IOException {
        // Parse the input JSON string into a Java Map
        Map<String, Object> inputMap = objectMapper.readValue(inputJson, new TypeReference<Map<String, Object>>() {});

        // The JOLT transform method can take a "context" map, which allows the spec
        // to reference external data. We'll provide our rules here.
        Map<String, Object> context = Map.of("rules", this.rules);

        // Perform the transformation
        Object transformedOutput = chainr.transform(inputMap, context);

        // Convert the final Java Object back to a pretty-printed JSON string
        return objectMapper.writerWithDefaultPrettyPrinter().writeValueAsString(transformedOutput);
    }

...
// Call to above method
String finalOutput = transformer.transform("{\"product_class\": \"Home\", \"product_subclass\": \"Cutlery\"}";);

How should I modify the JOLT spec here when input is separate from the rules file?

The trick is using "@<int>,&" model as the key element( "@2,&" in this case ) in order to provide sub-element-wise matching through use of ,&.

Therefore you can use the following spec :

[
  {
    "operation": "shift",
    "spec": {
      "rules": {
        "@1,product_class": {
          "*": { //the elements of the above node
            "@2,&": { //value matching with the aabove node 
              "@4,product_subclass": { //need to proceed 4 levels upper to arrive the level of this attribute
                "*": {
                  "@2,&": "attributes"
                }
              }
            }
          }
        }
      },
      "*": "&" //the elements other than "rules"
    }
  }
]

the demo on the site Jolt Transform Demo Using v0.1.1 is :