prompt-engineering

The prompt-engineering skill is designed to enhance the secure and efficient interaction between large language models (LLMs) and user inputs. It addresses key challenges such as prompt injection, multi-step task orchestration, and safe execution of LLM-driven workflows. By focusing on secure prompt construction and output validation, the skill ensures that LLM responses are controlled, reliable, and suitable for automated actions, mitigating potential security risks inherent in direct AI interactions.

This skill provides a comprehensive set of capabilities, including layered prompt architecture for separating system instructions, task-specific content, and user input. It emphasizes defense-in-depth strategies for injection prevention, context-aware task routing, multi-turn orchestration, and output sanitization. Key features include secure system prompt design, prompt injection detection, intent classification, task routing, and validation of LLM outputs before any action execution. Additionally, the skill follows principles like test-driven development for prompts, token optimization, instruction hierarchy, and fail-safe defaults to ensure robust performance and security.

Typical use cases include constructing safe prompts for JARVIS AI interactions, orchestrating complex multi-step workflows, classifying user intents, and securely calling tools or functions based on LLM outputs. Target users are developers and AI engineers working with LLM-based assistants who require secure prompt engineering, workflow orchestration, and reliable output validation. The skill is particularly relevant in environments where AI outputs can trigger automated actions, making security and reliability paramount.