The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The execution workflow in SKILL.md (Step 4) instructs administrators to generate Temporary Access Passes (TAPs) for users and export them to a local file named tap_assignments.csv. TAPs are highly sensitive, time-limited credentials capable of bypassing password authentication. Storing them in a plain-text CSV file creates a significant risk of credential exposure if the host machine's storage is accessed by unauthorized processes or users.
[INDIRECT_PROMPT_INJECTION]: The audit agent script possesses an attack surface where untrusted data from an external source is processed and presented in the agent's context.
Ingestion points: The scripts/agent.py file retrieves user metadata (e.g., displayName, userPrincipalName) and sign-in log details from the Microsoft Graph API.
Boundary markers: There are no specialized delimiters or instructions to ignore embedded content when presenting potentially attacker-controlled strings (like account names) in the audit outputs.
Capability inventory: The overall skill provides the capability to execute high-privilege PowerShell commands that modify global authentication and access policies.
Sanitization: The script does not sanitize or validate strings retrieved from the API before including them in console logs or JSON report files.

implementing-passwordless-auth-with-microsoft-entra