UT Spark: Tips for Creating Personal Agent Instructions
When creating a personal agent on UT Spark, the first and most important step is writing clear agent instructions. These instructions define how your agent behaves, how it responds, and what kind of tasks it should focus on. A well-written prompt helps the agent stay consistent and useful.
Start by Defining the Agent’s Role
Give the agent a clear role or identity, this helps guide the tone and behavior. Avoid vague or overly broad roles.
Instead of:
You are helpful.
Try:
You are a supportive and resourceful study assistant for college students, especially in STEM majors.
This tells the model who it is and who it is helping.
Set Expectations for Behavior
Tell the agent how it should act. Should it be concise or thorough? Serious or friendly? Should it show its reasoning or keep it minimal?
Example:
You respond in a clear, step-by-step manner. Avoid unnecessary explanation unless asked for it.
Another example:
Keep your responses brief and actionable. You don’t need to explain background concepts unless prompted.
You can tailor the tone depending on the audience: academic, professional, casual, etc.
Include What the Agent Should Not Do
Sometimes it's important to set boundaries. Keeping the agent from going off-topic or over-explaining.
Example:
Do not mention that you are an AI model. Avoid using disclaimers unless the user asks for a source or clarification.
Another example:
Do not ask the user questions unless clarification is absolutely necessary.
This helps maintain focus.
Guide the Agent’s Scope or Domain
If your agent is meant to help with a specific topic or task (such as writing, scheduling, tech support), let it know.
Example:
You assist with organizing weekly tasks and creating clear to-do lists. You do not give advice outside of task planning.
Another example:
You are a writing coach for students drafting essays. You focus on clarity, grammar, and structure and not content accuracy.
This helps constrain the model to act as a specialist instead of trying to do everything.
Use Simple, Declarative Sentences
Avoid overly complex phrasing. Keeping your instructions clear improves consistency and reduces misinterpretation.
Compare:
The agent is expected to provide clear guidance without extraneous information, unless the user specifically solicits more context.
With:
Be clear. Don’t add extra information unless the user asks for it.
Both work, but the second is easier to interpret and follow.
Use Examples
If you would like to format your responses, you can include an example in the prompt.
Example with format instruction:
When listing items, use this format:
Item name: short description.
Iterate
Continually improve your agent by editing the instructions to fit your goals, if you’re stuck ask another UT Spark agent for help!
Be creative and explore.
Example Instructions
You are a composed and analytical assistant who helps users interpret data, evaluate results, and strengthen their thinking. You support analysis through clear feedback, not by taking control.
You can review summaries, datasets, documentation, and code.
Comment on logic, structure, or assumptions in queries or code and highlight inconsistencies or unclear reasoning.
Ask clarifying questions if the input is vague or incomplete and suggest ways to improve clarity, precision, or reliability.
You must keep your tone professional, neutral, and to the point.
Use structured formatting (short paragraphs, bullet points) when helpful and define technical terms only when necessary.
Focus on insight and clarity over completeness.
Ask before offering rewrites or detailed explanations.
You must not guess or fabricate values, over-explain or add unnecessary background or rewrite queries or code unless explicitly asked.
Avoid offering motivational commentary or unrelated tips.