The Agent Skills Directory

[Log Analysis Security]: The skill includes a Node.js-based analyzer (analyze-speech-sdk-log.mjs) designed to process Azure Speech SDK logs locally. The script implements an allowlist-only extraction pattern, validating extracted fields with regular expressions and withholding potentially sensitive values using placeholders (e.g., present_value_withheld). This approach effectively manages the risks associated with processing untrusted log data.
[Authentication Guidance]: The documentation (references/auth-best-practices.md) provides clear, security-focused guidance, recommending the use of Managed Identities and Azure RBAC in production over hardcoded credentials or broad fallback mechanisms like DefaultAzureCredential.
[Secure Data Ingestion]: The skill provides explicit instructions on how to handle untrusted diagnostic data, advising against pasting raw logs into chat and instead using the provided local analyzer to generate a safe summary. This demonstrates a proactive approach to mitigating indirect prompt injection and data exposure risks.
[Dependency Integrity]: References to external SDKs and packages across multiple languages (npm, pip, Maven, NuGet) target official Microsoft and Azure repositories. These resources are appropriate for the skill's intended purpose and align with established vendor infrastructure.
[Execution Safety]: The provided Bash and PowerShell wrappers for the log analyzer perform environment checks (such as verifying Node.js version) and execute the local analyzer without performing unauthorized installations or system modifications.

azure-ai