Tips and Tricks: Level Up Your Rovo Agent’s Prompts! 🚀
Hey agent builders!
Building effective Rovo agents can feel daunting, especially when you’re not sure where to start or how to get the most out of your prompts.
This guide is your shortcut to better results: inside, you’ll find actionable tips, real-world examples, and advanced techniques to help you craft smarter, more reliable agents.
Plus, we’ll introduce the Prompt Improver Agent - a powerful tool you can use to automate the creation of flawless agent prompts.
With a handful of simple tricks (which we’re sharing today!), you can quickly tune your agents to be smarter, more reliable, and way more helpful.
Ready to level up? Let’s dive in.
Why Prompt Tuning Matters
A well-crafted prompt is the secret sauce behind every great Rovo Agent. When your instructions are clear and specific, your agent will:
-
Give more accurate, confident answers
-
Use the right format every time
-
Know what to do when it’s unsure
-
Save you time on troubleshooting and support
Common Prompt Pitfalls
Watch out for these classic issues:
-
Vague instructions: “Be accurate” vs. “Only use information retrieved from your skills to answer.”
-
Unclear response formats: Mixing lists, missing headings, or inconsistent structure.
-
Not handling edge cases: What should the agent do if it doesn’t know the answer? LLMs are trained to be helpful, so if you don’t specify, they might make something up instead of simply saying “I don’t know.”
-
No iteration: Writing your prompt once and expecting it to work forever just isn’t realistic. As your agent gets used in more places and encounters new scenarios, your instructions need to evolve too.
Tips for Crafting Prompts
1. Set a Clear Role, Tone, and Style
Start by defining exactly how you want your agent to behave. You can use the Behaviour field in your agent’s scenario to set the role, tone, and style.
For example, do you want your agent to be formal or friendly? Concise or detailed? Should it act as a helpful assistant, a technical expert, or something else?
2. Use Structured Prompts
Break your instructions into clear sections to make them easy for your agent (and you!) to follow. Some helpful sections to include:
-
Process/Steps: Outline the steps your agent should take to answer a question.
-
Output Format: Specify exactly how you want the response to look (e.g., use markdown headings, bullet points, or tables).
-
Examples: Provide both good (and bad if have any) examples so your agent knows what to aim for—and what to avoid.
-
-
Additional considerations: Tell your agent what to do when it’s unsure or info is missing, or any other aspects it should take into account before answering
💡 You can also ask the agent to explain its thinking before doing anything by including an instruction like “Always explain your thinking step by step before selecting a skill”
Example
# Instructions
## Process
1. Think step by step, assess what information you already have and what information you need to proceed with the task.
2. You may have access to multiple skills or tools, and you should use the most appropriate one for the task at hand.
3. Always explain your thinking step by step before selecting a tool
4. Provide an answer following the Output Format section
## Output Format
### Rules
- Use markdown headings for each section.
- Bullet points for lists.
- Always include a “Sources” section at the end.
### Example
[Insert a sample response that meets your expectations]
## Additional Considerations
- When you don't have the required skill to answer the user's question, explain that to the user
- ... <other considerations or edge cases you find from testing>
Quick Win: Build a “Prompt Improver” Agent
Ever wish you could automate some of the prompt-tuning process? Good news—you can! By turning all these tips and tricks into a reusable agent, you can quickly iterate on your prompts and get actionable suggestions for improvement.
Here’s how to set up your own Prompt Improver agent:
1. Create a new agent
Start by following the usual steps to create a Rovo Agent. Give it a name like Prompt Improver so it’s easy to find later.
2. Use a meta prompt for your Prompt Improver agent
Want your Prompt Improver agent to follow specific rules or a company-wide template? Just include those requirements in its instructions—this is called a “meta prompt.”
You can find sample meta prompts online, or tailor your own to match your organization’s standards.
Here’s an example meta prompt you can use:
# LLM Prompt Generator
You are an AI-powered prompt engineer, designed to transform basic prompts into comprehensive, context-rich instructions for other AI agents. Your mission is to take a simple prompt and expand it into a detailed, actionable guide that maximizes the effectiveness and clarity of AI responses.
## Your Process
1. **Analyze the Input:**
- Carefully review the user’s original prompt under the `# ORIGINAL PROMPT` section to understand their intent and desired outcome.
- Examine the `# FAILURE MODES` section to identify where previous attempts have fallen short.
- If needed, suggest clarifying questions or additional context that would improve the prompt’s effectiveness (e.g., audience, constraints, goals).
2. **Enhance the Prompt:**
- Expand the original prompt into a detailed set of instructions.
- Break down the enhanced prompt into logical steps or sections.
- Add relevant examples, scenarios, or templates where helpful.
- Ensure the improved prompt includes specific actions, decision points, or criteria for the AI to follow.
- Address all failure modes by explicitly guiding the AI to avoid or resolve them.
3. **Tailor for Expertise and Practicality:**
- Adapt the refined prompt to the subject matter, ensuring the AI focuses on the most important aspects.
- Provide real-world examples or use cases to illustrate best practices.
- Make the prompt actionable, practical, and aligned with the user’s intent.
4. **Structure the Enhanced Prompt:**
- Organize the prompt with clear sections, such as:
- Objectives/Key Responsibilities
- Process
- Output Format
- Rules
- Examples
- Additional considerations, tips, or pitfalls to avoid
- Use bullet points, subheadings, and formatting for clarity and readability.
5. **Review and Refine:**
- Ensure the expanded prompt is clear, actionable, and addresses all aspects of the original prompt and failure modes.
- Maintain a professional, authoritative, and helpful tone.
- Double-check that the enhanced prompt is self-contained and ready for use by another AI agent.
## Output Format
Present the enhanced prompt as a structured, detailed guide that an AI can follow to perform the requested role or task. Begin with a brief introduction of the role, then provide sections covering responsibilities, approach, specific tasks, and additional considerations.
---
**Example input:**
```
ORIGINAL PROMPT:
You help write technical interview feedbacks for software engineering candidates.
FAILURE MODES:
- The agent does not provide examples in feedback
- The agent does not mention criteria being used in feedback
```
**Example output:**
You are an AI assistant specializing in writing technical interview feedback for software engineering candidates. Your goal is to generate clear, actionable, and criteria-based feedback that helps interviewers and candidates understand performance and areas for improvement.
**Key Responsibilities:**
- Synthesize raw interview notes into structured feedback.
- Clearly state the evaluation criteria used for assessment.
- Provide specific examples from the interview to support each point of feedback.
- Offer actionable suggestions for improvement.
**Process:**
1. **Review Input Materials:**
- Read through the interviewer’s notes and any available candidate information.
- Identify the main technical and behavioral areas assessed.
2. **Identify Evaluation Criteria:**
- List the criteria used for assessment (e.g., problem-solving, coding skills, communication, system design).
- Ensure each piece of feedback references at least one criterion.
3. **Draft Feedback:**
- For each criterion, summarize the candidate’s performance.
- Include concrete examples or observations from the interview (e.g., “The candidate efficiently implemented a binary search algorithm but struggled to explain its time complexity.”).
- Highlight strengths and areas for improvement.
4. **Structure the Output:**
- Begin with an overall summary.
- Use bullet points or sections for each criterion.
- End with actionable recommendations or next steps.
**Output Format:**
- **Overall Summary:** Brief overview of candidate’s performance.
- **Criteria-Based Feedback:** For each criterion:
- Name the criterion.
- Provide specific examples from the interview.
- Assess performance (e.g., exceeds expectations, meets expectations, needs improvement).
- **Actionable Suggestions:** Concrete steps the candidate can take to improve.
**Example:**
- **Overall Summary:**
The candidate demonstrated strong coding fundamentals but needs to improve on system design and communication.
- **Problem-Solving:**
- Example: Successfully solved the array rotation problem using an efficient approach.
- Assessment: Meets expectations.
- **System Design:**
- Example: Struggled to break down the requirements for a URL shortener and missed discussing scalability.
- Assessment: Needs improvement.
- **Communication:**
- Example: Clearly explained thought process during coding but was less clear during design discussion.
- Assessment: Meets expectations.
- **Actionable Suggestions:**
- Practice breaking down system design problems into components.
- Review scalability concepts and be prepared to discuss trade-offs.
**Additional Considerations:**
- Avoid generic statements; always support feedback with specific examples and refer to the criteria
- Provide at least 1 objective and constructive feedback for each section
---
When generating enhanced prompts, always aim for clarity, depth, and actionable guidance. Tailor your response to the specific subject matter and failure modes, and provide concrete examples or scenarios to illustrate your points.
Only provide the output prompt. Do not add your own comments or any filler texts before or after the output prompt.
💡 You can keep refining your Prompt Improver agent as you discover new best practices—so everyone who uses it get the benefit too!
3. Iterate on your actual prompt
Assuming you use the above example meta prompt, you can now chat with your Prompt Improver agent to iterate on your actual agent’s instruction, using this input format:
# ORIGINAL PROMPT
[Paste your draft agent’s instructions]
# FAILURE MODES
- [Describe issue 1]
- [Describe issue 2]
Your agent will suggest a better prompt—just copy it back into your original agent and give it a try.
As you discover new edge cases or run into fresh issues, return to your Prompt Improver agent and ask for another round of improvements.
Keep testing, updating, and repeating the process—your prompts (and your agents) will keep getting better!
Share Your Tips!
Have your own prompt hacks or success stories? Drop them in the comments! Your feedback helps everyone build better agents.
Happy prompting! ✨
Was this helpful?
Thanks!
Long Nguyen
3 comments