The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation provides several PowerShell and Bash code blocks designed to be executed to configure Entra ID policies and interact with Cloud Access Security Broker (CASB) APIs.
[EXTERNAL_DOWNLOADS]: The workflow includes curl commands targeting api.cloudappsecurity.com. This is an interaction with a well-known security service (Microsoft Defender for Cloud Apps) and is considered a neutral finding for this use case.
[INDIRECT_PROMPT_INJECTION]: The included scripts/agent.py script retrieves external data that could be manipulated by an attacker to influence the agent's behavior if the output is processed as natural language instructions.
Ingestion points: The script fetches application display names from the servicePrincipals endpoint and user names from the riskyUsers endpoint within the Microsoft Graph API.
Boundary markers: The script does not implement delimiters or specific instructions to the agent to treat these retrieved strings as untrusted data.
Capability inventory: The script has the ability to write JSON reports to the local file system and perform network requests via the requests library.
Sanitization: No validation or sanitization is performed on the strings retrieved from the API before they are printed or saved to the report.

implementing-zero-trust-for-saas-applications