The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents patterns for workflows that process external data from Webhooks, REST APIs, and RAG document stores. This creates a surface where an attacker could provide malicious instructions via external data that are then processed by an AI Agent or downstream nodes.\n
Ingestion points: webhook_processing.md (form inputs), http_api_integration.md (external API responses), and ai_agent_workflow.md (document loaders/RAG).\n
Boundary markers: The documentation explicitly mentions the risk of indirect prompt injection in ai_agent_workflow.md and recommends mitigations like human review and read-only scopes.\n
Capability inventory: The patterns involve file system access (writing backups), network operations (HTTP requests), and database writes.\n
Sanitization: The skill provides guidance on input validation, signature verification for webhooks, and parameterized SQL queries to prevent injection.\n- [DYNAMIC_EXECUTION]: The documentation provides multiple JavaScript code snippets intended for use in n8n 'Code' nodes. These examples include the use of Node.js built-in modules to perform sensitive operations.\n
Evidence: database_operations.md and scheduled_tasks.md provide code for encryption using crypto and executing system commands like pg_dump via child_process.exec for database backup automation.

n8n-workflow-patterns