What problem does it solve? LLM application development often relies on ad-hoc prompt tweaks and untested assumptions, leading to weak prompts, truncated context windows, insecure agents, and missing evaluation pipelines. This Skill replaces guesswork with six structured, repeatable AI engineering workflows that produce consistent, quantifiable results. ## Core Features & Use Cases - Prompt Evaluator: Scores prompts across 8 dimensions (Clarity, Specificity, Safety, etc.) on a 0-100 scale, identifies the weakest dimensions, and generates improved rewrites. - Context Budget Planner: Analyzes token allocation across 5 context zones and produces an optimized distribution plan with compression strategies. - RAG Pipeline Architect: Walks through architecture decisions from parsing and chunking to embedding models, retrieval methods, and evaluation metrics. - Agent Safety Guard: Runs a 65-point authorized red-team audit covering prompt injection, information extraction, tool abuse, and goal hijacking, with fix recommendations. - Eval Harness Builder & Product Sense Coach: Designs LLM-as-Judge evaluation pipelines with bias mitigation, and guides product strategy decisions before coding. - Use Case: Before launching a customer support agent, run the prompt evaluator to raise a 28/100 system prompt to 82/100, then run the security audit to catch injection bypasses and path traversal flaws before deployment. ## Quick Start Ask the assistant to evaluate your system prompt or run a security audit on your AI agent using the ai-engineering-toolkit workflows.
