The Agent Skills Directory

External Downloads (SAFE): The skill configures a project-specific MCP server using @supabase/mcp-server-supabase via npx.\n
Evidence: Found in SETUP.md configuration examples.\n
Status: The download originates from a Trusted External Source (supabase organization), which downgrades the finding severity per the [TRUST-SCOPE-RULE].\n- Indirect Prompt Injection (LOW): The skill implements a CMS architecture where external data is ingested from the database and processed, creating an attack surface for indirect instructions.\n
Ingestion points: Data enters the context via the page_content table (DATABASE.md) and user_metadata fields such as full_name (AUTH.md).\n
Boundary markers: Absent; the code snippets do not show explicit delimiters or instructions for the agent to ignore embedded commands in the CMS strings.\n
Capability inventory: The agent has capabilities to perform database operations and dispatch emails via the Resend API.\n
Sanitization: No explicit sanitization or strict schema validation for content fields was observed in the provided logic.\n- Privilege Management (LOW): The application uses service_role clients to perform administrative operations that bypass standard RLS.\n
Evidence: AUTH.md utilizes createServiceRoleClient() for role queries and user invitation management.\n
Risk: While standard for these operations, it necessitates strict protection of the SUPABASE_SERVICE_ROLE_KEY, which the skill correctly identifies in the setup instructions.

supabase-vercel-shop