Prompting Best Practices

Overview

The quality of your prompts directly impacts Nova Act’s reliability and success rate. Nova Act works most reliably when tasks can be accomplished in fewer than 30 steps. This guide covers best practices for writing effective prompts.

Core Principles

1. Be Direct and Succinct

Tell the agent exactly what to do, not what you want to achieve. Avoid conversational or exploratory language.

Don't
Do

nova.act("Let's see what routes vta offers")

This is too conversational and unclear about the specific action needed.

nova.act("Navigate to the routes tab")

This is direct and specifies exactly what action to take.

Don't
Do

nova.act_get(
    "I want to go and meet a friend. I should figure out when the Orange Line comes next."
)

This includes unnecessary context and is not action-focused.

nova.act_get(
    f"Find the next departure time for the Orange Line from Government Center after {time}"
)

This is specific, direct, and includes relevant parameters.

2. Provide Complete Instructions

Specify all the details the agent needs to make decisions. Don’t leave choices or values ambiguous.

Don't
Do

nova.act("book me a hotel that costs less than $100 with the highest star rating")

Missing critical details: dates, location, number of guests, bed preferences.

nova.act(
    f"book a hotel for two adults in Houston between {startdate} and {enddate} "
    f"that costs less than $100 per night with the highest star rating. "
    f"two queen beds preferred but single king also ok. "
    f"stop when you get to the enter customer details or payment page."
)

Includes all necessary details and clear stopping criteria.

3. Break Up Large Tasks

Split complex multi-step workflows into smaller, focused act() calls. This improves reliability and makes debugging easier.

Don't
Do

nova.act(
    "book me a hotel that costs less than $100 with the highest star rating "
    "then find the closest car rental and get me car there, "
    "finally find a lunch spot nearby and book it at 12:30pm"
)

This combines three different tasks into one prompt.

hotel_address = nova.act_get(
    f"book a hotel for two adults in Houston between {startdate} and {enddate} "
    f"that costs less than $100 per night with the highest star rating. "
    f"two queen beds preferred but single king also ok. "
    f"return the address of the hotel you booked."
).response

nova.act(f"book a restaurant near {hotel_address} at 12:30pm for two people")

nova.act(
    f"rent a small sized car between {startdate} and {enddate} "
    f"from a car rental place near {hotel_address}"
)

Each task is separate, and results flow between them.

Further Breaking Down Tasks

If the agent still struggles, break it down even more:

# Search first
nova.act(
    f"search for hotels for two adults in Houston between {startdate} and {enddate}"
)

# Then sort
nova.act("sort by avg customer review")

# Then book and extract
hotel_address = nova.act_get(
    "book the first hotel that is $100 or less. "
    "prefer two queen beds if there is an option. "
    "return the address of the hotel you booked."
).response

# Continue with restaurant
nova.act(
    f"book a restaurant near {hotel_address} at 12:30pm on {startdate} for two people"
)

# Car rental
nova.act(f"search for car rental places near {hotel_address} and navigate to the closest one's website")
nova.act(
    f"rent a small sized car between {startdate} and {enddate}, pickup time 12pm, drop-off 12pm."
)

Writing Effective Prompts

Specify Actions Clearly

# Good - action-focused
nova.act("Click the submit button")
nova.act("Scroll down once")
nova.act("Type 'search query' into the search field")

# Bad - outcome-focused without clear action
nova.act("I need to submit this form")
nova.act("Show me more results")
nova.act("Look for something")

Include Contextual Hints

If Nova Act struggles with specific UI elements or navigation, add hints:

# Add hints about UI interaction
nova.act(
    "search for cats. type enter to initiate the search."
)

# Guide navigation paths
nova.act(
    "Navigate to the settings page. The settings link is in the top right corner menu."
)

# Specify what to avoid
nova.act(
    "Find the price of the first product. Don't click on any ads or sponsored results."
)

Use Variables for Dynamic Data

from datetime import datetime, timedelta

start_date = datetime.now()
end_date = start_date + timedelta(days=7)
destination = "San Francisco"
guests = 2

nova.act(
    f"Search for hotels in {destination} "
    f"from {start_date.strftime('%B %d')} to {end_date.strftime('%B %d')} "
    f"for {guests} guests"
)

Date Handling

Specify dates in absolute time format:

# Good - absolute dates
nova.act("select dates march 23 to march 28")
nova.act("enter departure date February 15, 2026")

# Avoid relative dates
# nova.act("select dates next week")  # Ambiguous

Handling Extraction

When you need information back from the page, be explicit about what you want:

# Clearly state what to return
result = nova.act_get(
    "Navigate to the confirmation page and return the booking confirmation number"
)

# Bad - unclear what should be returned
# result = nova.act_get("Go to the confirmation page")  # What should be returned?

Common Patterns

Search and Filter

# Clear, step-by-step approach
nova.act(
    "Close any cookie banners. "
    f"Search for apartments near {city}, "
    f"then filter for {bedrooms} bedrooms and {baths} bathrooms. "
    "Close any dialogs that get in the way of your task. "
    "Ensure the results mode is set to List."
)

Form Filling

# Be explicit about each field
nova.act(
    "Fill in the form with the following details: "
    "Name: John Doe, "
    "Email: [email protected], "
    "Phone: 555-1234. "
    "Do not submit yet."
)

# Specify the expected outcome
nova.act(
    "Click on the Products tab and wait for the product grid to load"
)

Handling Common Challenges

# Proactively handle common obstacles
nova.act(
    "Close any cookie banners or popups if they appear, "
    "then click the Sign In button"
)

Search Initiation

If the search button is hard to find:

nova.act("Type 'laptops' in the search box and press enter to search")

Multi-Step Forms

Break forms into stages:

# Step 1: Personal info
nova.act("Fill in name and email fields, then click Next")

# Step 2: Address
nova.act("Fill in the address form with [details], then click Next")

# Step 3: Review
confirmation = nova.act_get("Review the details and click Submit. Return the confirmation number.")

Testing and Iteration

When developing prompts:

Start simple: Begin with basic prompts and add detail as needed
Observe failures: When act() goes off track, enhance the prompt with hints
Break it down: If reliability is inconsistent, split into smaller steps
Add constraints: Specify stopping conditions and what to avoid

# Version 1 - Too vague
# nova.act("Find a flight")

# Version 2 - More specific but missing details
# nova.act("Search for flights from Boston to Seattle")

# Version 3 - Complete and actionable
nova.act(
    f"Search for round-trip flights from Boston to Seattle "
    f"departing {departure_date} and returning {return_date} "
    f"for {num_passengers} passengers. "
    f"Filter for non-stop flights under $500. "
    f"Sort by best value."
)

Best Practices Summary

Be direct - Use imperative commands, not conversational language
Be complete - Include all necessary details and constraints
Break it up - Split complex tasks into focused act() calls under 30 steps each
Add hints - Guide the agent with contextual information when needed
Use absolute values - Especially for dates and quantities
Specify returns - Clearly state what information you want back with act_get()
Iterate - Refine prompts based on observed behavior

Get Started

Core Concepts

Guides

Advanced

Overview

Core Principles

1. Be Direct and Succinct

2. Provide Complete Instructions

3. Break Up Large Tasks

Further Breaking Down Tasks

Writing Effective Prompts

Specify Actions Clearly

Include Contextual Hints

Use Variables for Dynamic Data

Date Handling

Handling Extraction

Common Patterns

Search and Filter

Form Filling

Navigation with Validation

Handling Common Challenges

Obstacle Navigation

Search Initiation

Multi-Step Forms

Testing and Iteration

Example: Iterative Refinement

Best Practices Summary

Build docs developers (and LLMs) love

Get Started

Core Concepts

Guides

Advanced

​Overview

​Core Principles

​1. Be Direct and Succinct

​2. Provide Complete Instructions

​3. Break Up Large Tasks

​Further Breaking Down Tasks

​Writing Effective Prompts

​Specify Actions Clearly

​Include Contextual Hints

​Use Variables for Dynamic Data

​Date Handling

​Handling Extraction

​Common Patterns

​Search and Filter

​Form Filling

​Navigation with Validation

​Handling Common Challenges

​Obstacle Navigation

​Search Initiation

​Multi-Step Forms

​Testing and Iteration

​Example: Iterative Refinement

​Best Practices Summary

Build docs developers (and LLMs) love

Overview

Core Principles

1. Be Direct and Succinct

2. Provide Complete Instructions

3. Break Up Large Tasks

Further Breaking Down Tasks

Writing Effective Prompts

Specify Actions Clearly

Include Contextual Hints

Use Variables for Dynamic Data

Date Handling

Handling Extraction

Common Patterns

Search and Filter

Form Filling

Navigation with Validation

Handling Common Challenges

Obstacle Navigation

Search Initiation

Multi-Step Forms

Testing and Iteration

Example: Iterative Refinement

Best Practices Summary