The Agent Skills Directory

[SAFE]: The skill provides placeholders like {key}, {endpoint}, and {SAS_URL_of_training_data} instead of hardcoding any real credentials. Additionally, it explicitly demonstrates the use of DefaultAzureCredentialBuilder, which is a security best practice for avoiding long-lived API keys.
[DATA_EXPOSURE_SURFACE]: The skill documentation includes clear security warnings regarding document boundaries and data handling. It specifically instructs users to obtain permission before uploading sensitive records (identity/financial), use short-lived SAS URLs, and avoid logging extracted content.
[EXTERNAL_DOWNLOADS]: The skill references the official com.azure:azure-ai-formrecognizer Maven dependency. Per the analysis rules, dependencies from well-known technology companies (Microsoft/Azure) are documented neutrally and do not escalate the security verdict.
[INDIRECT_PROMPT_INJECTION]: The skill describes processing untrusted external data (PDFs/Images). It mitigates injection and reliability risks by advising users to check null types for nested fields, validate OCR results against source documents, and handle polling failures.

azure-ai-formrecognizer-java