The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill processes untrusted external data (project files and metadata) to generate security-sensitive authentication and middleware code.
Ingestion points: Step 1 gathers Azure AD configuration and reads package.json; Step 2 loads project_overview.md and common_patterns.md from the project's memory.
Boundary markers: Absent. The instructions do not define delimiters to isolate external project content from the agent's logic.
Capability inventory: The skill generates executable authentication code (MSAL configuration, JWT validation logic) and has file-write permissions (Step 5 saves files to /claudedocs/).
Sanitization: Absent. There is no specified validation or escaping for the project-sourced data before it is interpolated into code templates.
[EXTERNAL_DOWNLOADS] (LOW): The skill references the jose and msal libraries and fetches public keys from Microsoft's JWKS endpoint.
Evidence: Step 4d specifies downloading keys from https://login.microsoftonline.com/{tenantId}/discovery/v2.0/keys.
Trust Scope: These are trusted sources (Microsoft and standard OIDC discovery), downgrading the download risk to LOW.

azure-auth