AI Prompt Examples

How will you later remember how you used Artificial Intelligence (AI) to advance your projects? And how will you share your learning with others? Here are some examples of prompts that I have used with AI agents to help me, and teams in interacting with projects. They illustrate a few principles along the way.

If you just use prompts and don’t keep track of them, you are missing an opportunity to easily document your work flow. Not only that, saved prompts and walkthroughs can be used to reproduce your work (see Sharing Prompts Instead of Code). A simple way to begin is to ask your AI agent to save your prompts in a file, e.g. prompts.md; alternatively, or in addition, save the Walkthrough artifact that the AI agent creates into a separate walkthrough.md file. If you do more work within the same conversation, ask the AI agent to update these files. Below are examples of prompts I have been developing.

Table of Contents

byandell.github.io/Documentation

Saving Prompts and Walkthroughs

Prompts:

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

You can do this at the beginning or end of a conversation. You can do it by clicking the Save button in the top right corner of the conversation window, and the Export button on a walkthrough panel. You can also (by hand or by prompt to AI) update or append to these files.

Examples:

AI Agent Conversation

A user interacts with an AI agent in a “conversation” or “chat session”, which involves a chain of prompts, often starting with a system instruction and moving from a broad framing prompt to specific refinement prompts. See Tyson’s Advanced Techniques.

System Instructions

System instructions are prior prompts, entered before specific prompts. They are meant to act as global rule to subsequent prompt chaining in an AI conversation. Tyson Swetnam has a nice example that I expand below for python and R:

Please set up the following system instruction:
# 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

Prompt CRAFTs

A prompt for a complicated task can have improved focus with the CRAFT framework. To illustrate, here is Tyson’s example for a grant proposal. Note that this focuses on the “draft introduction” rather than the whole proposal.

Context: I'm preparing a grant proposal for NSF funding on AI in education  
Role: Act as an experienced grant writer and education researcher  
Action: Review my draft introduction and suggest improvements  
Format: Provide feedback as tracked changes with explanations  
Tone: Professional, constructive, and encouraging

Here is one for beginning to build a set of python notebooks:

Context: I'm preparing a set of `python` notebooks to document flooding impact  
Role: Act as an experienced python programmer and software project manager  
Action: Review my draft plan for building the setup notebook  
Format: Provide feedback as tracked changes with explanations  
Tone: Professional, constructive, and encouraging

Subsequent “chained” prompts would refine the Action framing to build the first notebook, then move on to build related notebooks.

Conversations and Projects

There could be several different conversations within a project, with a project usually in its own folder or set of folders. Best practices include saving projects as GitHub repos to both share with others and evolve the project over time through edits documented through version control. Of course, saving prompts for those conversation with a project will help with later recall of how the project developed.

Several Prompt Examples

Create a README.md for a Folder

Prompts:

  • “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”

Example:

Create a Table of Contents for a File

Just as you can create a README.md for a folder, you can create a table of contents for that README.md (or any markdown) file.

Prompts:

  • “create TOC for README.md”
  • “update TOC for README.md”

Example:

Once you have these and establish a “culture” of updating them, the AI will sometimes make the changes on the fly as you evolve your material.

Convert DOCX files to Markdown

This prompt use pandoc to convert docx files to md files. This app needs to be installed once on a laptop to use.

Prompt: “find docx files and use pandoc to convert them to md versions”

Example:

Powerpoint Presentation

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.”

This prompt assumes that you have installed Quarto (only need to do once) and have already generated the PNG figures using the workflow prompt. It could be part of a more involved prompt that also generates the figures. The AI will likely include a title slide; you can also ask AI to add other material such as references or overview of experiment.

Example:

This prompt was used to create the powerpoint presentation for a GTT project. For more detail, see Sharing Prompts instead of Code. The saved file includes Alan and Diana’s versions of the prompt . For details on how the prompt evolved, click on Blame (top left) or History (top right). The displayed version is the most recent (Diana’s); the same file contains Alan’s earlier version.

Another user could create a prompt that cites this prompt file and directs AI to substitute different data and code files. However, the prompt as written is rather specific to GTT for DO founder mice, and in particular to NZO and B6 strains. A user could further specify to replace step 2 with their own version of instructions.

GitHub Prompts

These prompts interact with Git and set up files to be pushed to GitHub. To use these, a GitHub account and working knowledge of how Git and GitHub work is assumed. See GitHub references for more information.

Commit to GitHub multiple versions of an R file

Prompt: Find files ending in [ _]v[0-9]+.R$ that have the same name before this version. If there is more than one set, do the following steps for each set. Denote by [basename] the part of the filename before the version number.

  1. If there is not a v1 version but there is the [basename] without a version, copy [basename].R to [basename] v1.R or [basename]_v1.R as appropriate. If there is already a [basename] v1.R or [basename]_v1.R, do not overwrite it, but instead copy the v1 version to [basename].R for this set.
  2. Commit [basename].R with message v1 and the date this file was created. (Be careful with files that have implicit v1.)
  3. For subsequent versions, copy the next highest version number to [basename].R and commit with the next version number and the date this file was created. Continue until all versions are committed.
  4. Do not include the singleton files such as source*v2.R.
  5. Make sure to keep the versioned files intact.

The above prompt is a great task for an AI or an automated script. The original prompt is logically sound but can be streamlined to be more “instructional” and clear about the desired state. The file file versions prompt has an improved, more concise version of the prompt.

Examples:

For details of versions, see for instance MafA_SNP_DEG_Integration.R and click on Blame (top left) or History (top right).

Resolve Git Hell Conflicts

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

Example:

This was used on the Multiple Versions prompt above. The first time, I forgot to git pull changes I made on GitHub before starting. I tried to solve by myself and got lost. AI sorted it out. See GitHub references for more information.

Organize a Workflow

This complicated prompt requires various tools on one’s laptop, notably devtools and roxygen2 in R. Installing the standalone app Quarto gives one a markdown engine that can render markdown files to html, docx, pdf, and other formats. Other R packages will likely be needed along the way.

Prompt: “build a concise prompt that captures the essence of the walkthroughs in inst/doc/walkthrough.md”

Examples:

The workflow.md is a refinement of the workflow developed for sysgenAnalysis. This particular prompt, which is described conceptually in Prompts to Organize Workflows, transforms an R workflow into functions in an R package, including a function to run the workflow and another function to explore the results.

This workflow for organizing workflows (!) was developed organically. Each day, I advanced the project through a series of prompts; at the end of the session, I asked the AI agent to save the workflow walkthroughs, including the date of that step. At some point, I asked AI the above prompt, which resulted in a comprehensive workflow prompt. I later asked the AI agent to modify the workflow prompt as I added subsequent workflow steps.

Documenting Folders in ResearchDrive

Several of the prompts shown earlier concern building documentation for data and code kept in folders in the ResearchDrive (RD). While the RD is secure and private, I have made public versions of some code and documentation via the Attie Lab Systems Genetics GitHub. Having these public allows us to share ideas with others and to use them as a reference for future projects, while still keeping research data secure.

This site uses Just the Docs, a documentation theme for Jekyll.