The Agent Skills Directory

Indirect Prompt Injection Mitigation: The skill ingests data from Entra ID (Microsoft Graph), such as application display names, which could potentially contain malicious instructions. The instructions explicitly address this by directing the agent to treat the discovery JSON as untrusted data and to disregard any instructions embedded in these fields.\n- Structured Configuration Management: The skill manages auth.config.json by updating environment and client identifiers. It incorporates security best practices by explicitly prohibiting the storage of secrets, tokens, or Dataverse identity fields within these configuration files.\n- Controlled Command Execution: The skill utilizes local scripts for specialized tasks like environment resolution and registration discovery. It uses standard bash utilities for file verification and basic JSON parsing, maintaining a clearly defined scope of operation.\n- Input Validation: User-provided inputs, such as pasted Client IDs, are subjected to GUID format validation before being written to configuration files, reducing the risk of malformed data entry.

set-app-registration-native