Secure design patterns for your AI Code LLM applications: external content ingestion, endpoint hardening, output validation, system prompt design, and multi-agent action surface control.
Maps to OWASP Top 10 for LLM Applications 2025 (November 2025). Includes /llm-secure-patterns:report to generate audit-ready SECURITY_POSTURE.md and DEVELOPER_RECOMMENDATIONS.md files.
Skills 6 in the Claude listing
Agent action surface Use when designing or wiring up tool_use, function_calling, MCP servers, multi-agent chains, or calling multiple LLM providers (e.g., Gemini for vision plus Claude for reasoning) in the same pipeline Llm endpoint hardening Use when designing or building a web route (FastAPI, Flask, Express, or similar) that accepts user input and forwards it to any LLM API (Claude, OpenAI, Gemini, or similar) — fire this skill during ar Output validation Use when designing or implementing code where LLM responses are rendered as HTML or markup in a frontend, written to a database, passed to another service, or used to generate emails or documents — fi report Review or audit an existing codebase's security posture against the OWASP LLM Top 10. Secure external ingestion Use when designing or implementing code that fetches URLs, scrapes pages, calls external APIs, parses PDFs/HTML from untrusted sources, builds RAG retrieval pipelines, or processes any external conten System prompt design Use when designing, writing, editing, or reviewing system prompts for any LLM application — fire this skill during architecture/planning discussions (before prompts are drafted), not just during imple 