The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted external data through indexing and search operations, which could contain malicious instructions if the results are later processed by an agent.
Ingestion points: Data is ingested through the uploadDocuments method and search queries are processed via the search method in SKILL.md.
Boundary markers: The provided code snippets do not implement specific delimiters or instructions to ignore embedded commands within the indexed content or search queries.
Capability inventory: The skill utilizes network capabilities to interact with the Azure AI Search API to manage indexes and retrieve data.
Sanitization: The implementation does not demonstrate explicit sanitization or filtering of the content retrieved from or sent to the search service.
[EXTERNAL_DOWNLOADS]: The skill installs official Azure SDK packages from a well-known service.
Evidence: Installs @azure/search-documents and @azure/identity via npm.
[CREDENTIALS_UNSAFE]: The skill references environment variables for authentication, supporting both admin keys and secure identity-based access.
Evidence: References AZURE_SEARCH_ADMIN_KEY and DefaultAzureCredential in SKILL.md. The use of DefaultAzureCredential is a recommended security practice for Azure services.

azure-search-documents-ts