AI Prompt Examples

Documenting and Sharing AI Agent Workflows

Brian Yandell (byandell.github.io)

2026-06-18

Introduction & Core Concepts

Documenting AI Workflows

How will you later remember how you used Artificial Intelligence (AI) to advance your projects? And how will you share your learning with others?

  • Saved prompts and walkthroughs can be used to reproduce your work.
  • Share prompts instead of just raw code.
  • Ask your AI agent to save prompts to a file (e.g., prompts.md).
  • Save Walkthrough artifacts in a separate walkthrough.md file.

Prompt Engineering Resources

Here are some guidebooks and frameworks to help refine your prompts:

Saving Prompts & Agent Conversations

Saving Prompts and Walkthroughs

Prompts to Save Progress:

  • "save the prompts used in this conversation to prompts.md"
  • "save the walkthrough to walkthrough.md"

When & How:

  • Run at the beginning or end of a conversation.
  • Click Save (top right) or Export on the walkthrough panel.

Real-world Examples:

AI Agent Conversation Structure

  • User interactions happen in conversations or chat sessions.
  • A session involves a chain of prompts:
    1. System Instruction: Establishes global constraints/roles.
    2. Framing Prompt: Broad definition of the task.
    3. Refinement Prompts: Step-by-step guidance.
  • Keep different conversations within a project separated.
  • Store projects as GitHub repos for version control.

System Instructions

Set global rules for prompt chaining in your session:

# Project Context
I'm a data scientist working on machine learning projects.
For Python code, always provide examples using `scikit-learn` and `pandas`.
For R code, always provide examples using `tidyverse`.
Include `docstrings` (Python) or `roxygen2` (R) and type hints in all code.

# Response Preferences
- Be concise but thorough
- Explain complex concepts with analogies
- Always cite sources when making factual claims

The CRAFT Framework

For complicated tasks, structure your prompt using the CRAFT framework:

  • Context: Background information.
  • Role: Person/identity the AI assumes.
  • Action: The specific task to perform.
  • Format: Deliverable structure.
  • Tone: Style of response.

Example: Grant Proposal

Context: Grant proposal for NSF funding on AI in education  
Role: Experienced grant writer & researcher  
Action: Review draft introduction and suggest improvements  
Format: Feedback as tracked changes with explanations  
Tone: Professional, constructive, and encouraging

Practical Prompt Examples

Documenting Files & Folders

Create a Folder README

  • Prompt:
    • "create a README.md document that concisely summarizes
      contents of this folder at a high level"
    • "update README.md with any additional files,
      obeying restrictions in .gitignore"

Create a Table of Contents (TOC)

  • Prompt:
    • "create TOC for README.md"
    • "update TOC for README.md"

Pro-Tip

Establishing a culture of updating these docs enables the AI to update files on the fly as your project evolves!

Convert DOCX to Markdown & Slides

Convert DOCX via Pandoc

  • Prompt:
    • "find docx files and use pandoc to convert them to md versions"
  • Prerequisites: Pandoc installed locally.

Generate Slides via Quarto

  • Prompt:
    • "Use Quarto to automatically generate a PowerPoint deck
      (.pptx) embedding all of the outputted PNG figures."
    • "Add a table of contents slide to the presentation."

Note

See Diana and Alan’s evolved prompt versions for creating PowerPoint slides.

GitHub Prompts: Versioning Files

Multi-Version Commit Prompt

Find files ending in `[ _]v[0-9]+.R$`
with matching base name.
- If no `v1` exists but `basename.R`
  does, copy to `v1.R`.
- Commit `basename.R` with message `v1`
  and original date.
- For subsequent versions, copy the
  next version to `basename.R` and commit.
- Keep versioned files intact.

Refined Prompts

Check out file versions for a streamlined, instruction-optimized version of this prompt.

GitHub Prompts: Git Conflict Resolution

When you run into conflicts (e.g., forgetting to pull before committing):

The “Git Hell” Prompt

  • “I messed up and forgot to do git pull before committing new files.
    Now I am in git hell. See the message with git status.
    Walk me through how to fix this.”

Outcome

The AI successfully parsed the status and sorted out the repository state, providing step-by-step guidance.

Organizing a Workflow

The Goal

Transform a complex organic R workflow into neat functions inside an R package, featuring: - A function to run the workflow. - A function to explore results.

The Prompt

  • "build a concise prompt that captures the essence of the walkthroughs in inst/doc/walkthrough.md"

Key Packages

Uses devtools, roxygen2, and Quarto to render markdown outputs (HTML, PDF, DOCX).

Folder Documentation in ResearchDrive

The Challenge

Data and code are kept securely and privately on ResearchDrive (RD), but sharing/version control is still needed.

The Solution

Maintain public markdown documentation and folder structures on GitHub (with data-less READMEs) while keeping raw data safe in RD.