The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documentation describes the ingestion of untrusted data from an LLM via the query variable. It mitigates this risk by explicitly instructing developers to use structured JSON schemas for validation and providing example code that performs strict type checking and range validation on inputs.
Ingestion points: The query (JS) and _query (Python) variables in the @n8n/n8n-nodes-langchain.toolCode node.
Boundary markers: Encourages the use of specifyInputSchema to define a strict JSON Schema boundary.
Capability inventory: Execution of JavaScript or Python logic within a restricted n8n sandbox that lacks access to network helpers (this.helpers.httpRequest) and standard workflow execution data ($fromAI).
Sanitization: The provided examples demonstrate sanitizing inputs using JSON.parse and manual validation checks such as Number.isFinite() and typeof checks.
[COMMAND_EXECUTION]: While the skill explains how to write code for a tool that executes code, all examples provided are for pure mathematical computation (car loan calculations) and do not involve system commands or subprocess spawning.
[REMOTE_CODE_EXECUTION]: No remote code downloads or execution patterns were detected. The skill references official n8n documentation and source code on GitHub, which are well-known and trusted sources.

n8n-code-tool