Beyond the Prompt: Building Reusable AI Skills and Agents

Session Submission Type: AI Peer Learning Lab

Abstract

Turning recurring tasks and research context into AI tools participants can use repeatedly

PRESENTATIONS

Building an AI Second Brain for Research Using Claude Skills
Facilitator: Sodanid Phuong - Arizona State University, PhD Student
Training Category: Build a Second Brain as Claude Skill
AI Tool : Claude Participant Type: Students Skill level recommended: Beginner Recommended preparation or pre-work before session: Bring a laptop and sign up for a free Claude account before attending the training. 

If you often find yourself using AI tools merely as chatbots in short, disconnected sessions or chats, then this training is for you. Most graduate researchers waste time in re-explaining the necessary contexts of their research every time they create a new chat. This practical training session will teach participants how to avoid this inefficiency by creating a reusable Claude Skill that is more widely known as a “Second Brain.” Once there, Claude will consult it automatically whenever necessary, without users having to re-explain or re-upload files in every chat session. In 20 minutes, participants will follow along with trainer’s demo of building a “second brain” as Claude Skill on the screen while they are building one for themselves. A tentative activity outline includes introduction to the training (overview and learning objectives), live demo and interactive build-along, closing, and announcement of voluntary training evaluation form fill-in request. After attending the training, participants will be able to (1) develop an understanding of a “second brain” and Claude Skill, (2) develop the ability to use Claude Skill, (3) build a “second brain” for their research (or other matters) that they can keep using and expanding.

********
Teach AI How You Work: Building Reusable Skills for Policy Research
Facilitator:  Albert Okwei - Virginia Commonwealth University, PhD Student, Public Policy & Administration | Graduate Research/Teaching Assistan
Training Category: Research Productivity & Workflow Automation
AI Tool: Claude
Participant Type: Researchers
Skill level recommended: Intermediate
Recommended preparation or pre-work before session: Bring a laptop and be signed in to Claude/ChatGPT. Before the session, think of one task you repeat in your research or professional work and the instructions you usually give yourself, a colleague, or an AI tool to complete it. No coding experience is required.

In my own research, I often find myself giving AI the same instructions again and again. I may want it to review a methods section using the same criteria, check code for the same problems, organize evidence consistently, or turn research findings into a policy-facing format without changing the meaning. Rather than rewriting those instructions every time, I have started thinking about these recurring tasks as reusable AI skills. This session will show participants how to take a task they regularly perform and turn their process into a reusable set of instructions that an AI system can apply again. I will demonstrate the process using a policy research example, starting with an ordinary request and then building a skill that specifies when to use it, what information it needs, the steps it should follow, the checks it should perform, and the format of the final output. Participants will then identify one recurring task from their own work and sketch a simple skill for it. Examples could include reviewing a research design, checking causal claims, screening literature, or auditing code. We will test the skill on a new example and discuss what happens when the instructions are incomplete, too broad, or poorly specified.  The goal is not to create a perfect prompt. It is to show researchers how to make useful parts of their own workflow reusable, consistent, and easier to improve over time. Participants will leave with a simple skill-building template they can adapt to their own research or professional work.

********
Optimizing Workflows and Processes: Building a Responsible AI Agent
Facilitator: DORCAS ATTAH - Georgia State University, Policy Specialist
Training Category: Building AI Agents
AI Tool : Copilot
Participant Type: Policy researchers, analysts, practitioners, program managers, nonprofit leaders, educators, and students
Skill level recommended: Open_to_all
Recommended preparation or pre-work before session: Participants should have prior experience using Microsoft Copilot for basic tasks in their work, research, studies, or daily lives. Before the session, participants who wish to complete the live build-along should confirm that they can sign in to Microsoft Copilot and bring a laptop with internet access. No coding experience is required. Participants should not upload confidential, proprietary, personally identifiable, or otherwise sensitive information during the session. A publicly available policy document will be used for the demonstration.

Researchers, practitioners, and students regularly translate lengthy reports, policy guidance, and program materials into concise information for decision-makers. This interactive tutorial will demonstrate how to build a no-code Microsoft Copilot agent that supports this process while keeping human judgment at the center. Using a publicly available policy document as the demonstration source, participants will learn how to define an agent’s purpose, identify its intended users, establish trusted knowledge sources, write clear instructions and prompts, and specify a consistent decision-brief format. I will then test the agent using realistic questions and demonstrate how to identify unsupported claims, missing context, weak source attribution, and other output limitations. Participants with Copilot Studio access will be invited to follow the steps on their own devices, while others can participate through the guided demonstration and agent-design template. The session will follow a concise build-test-refine structure. First, participants will identify the elements that distinguish a reusable agent from a one-time prompt. Next, they will observe the step-by-step construction of an agent in Microsoft Copilot Studio. The group will then evaluate a sample response, identify potential risks, and recommend improvements to the agent’s instructions, guardrails, and human-review checkpoints. By the end of the session, participants will be able to outline a simple agent for a recurring research, policy, or program-management task; distinguish among instructions, knowledge sources, prompts, and guardrails; and apply a basic testing framework before relying on AI-generated output. The demonstration will use only public information and will address privacy, bias, hallucinations, source verification, and appropriate human oversight. Although Microsoft Copilot Studio will be used for the demonstration, the underlying agent-design framework is platform-neutral and can be adapted to other AI environments. No coding experience is required.

Policy Area

• AI Peer Learning Lab

Facilitators

• Sodanid Phuong, Arizona State University

• Albert Nii Noi Okwei, Virginia Commonwealth University

• DORCAS U ATTAH, Workforce Intelligence Institute

Individual Presentations

• Presentation for Sodanid Phuong - Facilitator: Sodanid Phuong, Arizona State University

• Presentation for Albert Nii Noi Okwei - Facilitator: Albert Nii Noi Okwei, Virginia Commonwealth University

• Presentation for DORCAS U ATTAH - Facilitator: DORCAS U ATTAH, Workforce Intelligence Institute