A comprehensive AI agent skill for anyone working with AI models. Helps you choose the right model for any task, write effective prompts, evaluate model outp...
The same AI model, given two different inputs, can produce outputs so different in quality that they might as well have come from different systems entirely. One person asks a question and gets a shallow, generic response that tells them nothing they did not already know. Another person asks about the same topic and gets something that changes how they think about the problem.
The model did not change. The interaction did.
This gap — between what AI models are capable of and what most people actually extract from them — is the defining productivity divide of the next decade. The people who close it will have access to a cognitive multiplier that compounds across everything they do. The people who do not will use AI as a slightly faster search engine and wonder why the results feel disappointing.
This skill closes the gap.
The AI landscape in 2025 contains more capable models than at any previous point, from more providers, at more price points, optimized for more different tasks. This is good news for people who understand how to navigate it and confusing noise for everyone else.
The skill helps you choose the right model for any specific task. Not the most powerful model — more powerful is not always better, and the most capable models are often slower and more expensive than the task requires. The right model for writing a first draft of a long document is different from the right model for answering a specific factual question, which is different from the right model for writing and debugging code, which is different from the right model for analyzing a complex PDF.
It explains the meaningful differences between frontier models across the major providers — what each one genuinely does better than the alternatives, where each one has consistent weaknesses, what tasks each one is specifically optimized for. It helps you build a mental map of the landscape that lets you make these choices quickly rather than defaulting to one model for everything.
A prompt is not a question. It is an instruction set. The difference between a prompt that produces useful output and one that produces generic output is almost never about the underlying capability of the model. It is about the specificity, structure, and context of the instruction.
The skill teaches prompting as a craft rather than a trick. Not a collection of magic phrases that unlock hidden capabilities, but a set of principles that produce better outputs across any model and any task.
The most important of these principles: models perform better when they understand the purpose behind a request, not just the request itself. Telling a model what you want is less effective than telling it what you want, why you want it, who it is for, and what a good outcome looks like. The additional context costs you thirty seconds. It changes the quality of the output significantly.
The skill helps you apply this principle and others to your specific use cases — writing, analysis, research, coding, planning, communication — with examples calibrated to what you are actually trying to accomplish rather than abstract demonstrations.
The most dangerous way to use AI is uncritically. A model that sounds confident and produces fluent, well-structured text can be wrong in ways that are not obvious without domain knowledge — and the fluency of the output can create a false sense of reliability.
The skill builds a critical evaluation framework for AI outputs. What to check in factual claims and how. Where models systematically underperform and why — the tasks that look like they should be easy and are not, the failure modes that recur across models and tasks. How to use the model's own uncertainty as a signal rather than ignoring it. When to verify independently and when the cost of verification exceeds the risk of being wrong.
This is not a counsel of skepticism about AI. It is a counsel of appropriate trust — high where models are reliably strong, calibrated where they are inconsistently reliable, skeptical where they are systematically weak.
A single good prompt is a one-time result. A reliable workflow is a repeatable system that produces consistently useful outputs across different inputs and different days.
The skill helps you build workflows that work reliably. How to structure multi-step tasks so that each step produces output the next step can use effectively. How to handle the variability in model outputs — the fact that the same prompt does not always produce the same result — by building checks into the workflow rather than assuming the first output is always good enough. How to combine AI capabilities with human judgment at the points where human judgment is irreplaceable.
For the tasks you do repeatedly — the weekly report, the client communication, the research synthesis, the first draft — the skill helps you build a workflow that reduces the effort required while maintaining or improving the quality of the output.
Most people who use AI models regularly have only a vague understanding of what they are and how they work. This vagueness produces unrealistic expectations in both directions — tasks assumed to be easy that are actually hard, and tasks assumed to be hard that are actually trivial.
The skill explains what language models are in terms that are accurate without being technical. What training data means for what a model knows and does not know. Why models confabulate — produce confident-sounding false information — and under what conditions this is most likely. What context windows are and why they matter for how you structure long interactions. Why the same model can produce different outputs to identical inputs and what this means for how you should use it.
This understanding does not require a technical background. It requires thirty minutes of clear explanation, which the skill provides through your actual questions rather than a generic tutorial.
The AI model landscape changes faster than almost any other technology domain. Models that were state of the art six months ago have been superseded. Capabilities that did not exist a year ago are now standard. Pricing that made certain use cases impractical has dropped to make them routine.
The skill helps you stay oriented in this landscape without needing to follow every benchmark release and research paper. When something changes that is relevant to how you work — a new model that genuinely outperforms what you are currently using for your specific tasks, a capability that did not previously exist and now does, a pricing change that affects the economics of your workflow — it surfaces this as useful information rather than noise.
The goal is not to always be using the newest thing. It is to always be using the right thing.
ZIP package — ready to use