The Agent Skills Directory

[PROMPT_INJECTION]: The instructions include a guideline to 'Do not produce visible output until the final step, unless blocked.' This is interpreted as a standard user-experience optimization for complex multi-tool agents to reduce intermediate message noise, rather than a malicious attempt to conceal tool actions from the platform or user.
[INDIRECT_PROMPT_INJECTION]: As the skill ingests user requirements to build functional code, it possesses a data ingestion surface common to AI agents. However, the skill incorporates specific 'Knowledge Base Guardrails' and strict placeholder requirements for sensitive data, which are effective mitigation strategies for the intended use case.
[DYNAMIC_EXECUTION]: Generating executable code is the primary purpose of this skill. The security risk is managed by generating code for a restricted sandbox environment (n8n) where the SDK explicitly forbids native loops, globals, and network access for Code nodes, preventing typical runtime escape or exfiltration attacks.

workflow-builder