Hi,

I found out that the issue is with the JSON input format. The actual input looks like this:

        "values": [
            {
                "recordId": "1",
                "data": {
                    "file_data": {
                        "data": "<base64-encoded-pdf>"
                    }
                }
            }
        ]
    }

However, I had assumed it followed the format shown in the documentation for #Microsoft.Skills.Util.DocumentExtractionSkill:
https://learn.microsoft.com/en-us/azure/search/cognitive-search-skill-document-extraction

Also, I found a similar question here:
How do I read the original pdf file in set indexer datasource in a custom WebApiSkill after enabling "Allow Skillset to read file data"