The NGA Mindset
What is the primary metric used to evaluate "conversational efficiency"?
The number of "turns" it takes to complete a task.
If an account has "GPT Support Opt Out" set to "True," what specific error message must the Agent display?
"Sorry, I can't process any data related to this account due to legal requirements".
What is the "Golden Rule" for every error message to ensure there are no "Dead Ends"?
Every error message must end with a question or a clear call to action.
What symbol prefix is used to access variables, actions, and topics in Agent Script?
The @ symbol (e.g., @variables, @actions, @topic)
What special syntax (three characters) tells the LLM to extract a value from the user's conversation when binding action inputs?
... (three dots, also called "slot-filling")
To cut down on initial validation steps, the Agent should use this type of recognition when a user is on a specific page or channel.
Context Recognition
What must the Agent do if a user mentions a specific record name in their utterance while they are already on a different record page?
It must override any existing system-identified context with the specific record mentioned.
If an action fails because the user forgot to provide a required parameter, what mode should the Agent shift into?
Data-gathering mode.
In Agent Script, what keyword declares a variable that the agent can modify during conversation?
mutable (e.g., user_name: mutable string = "")
What clause makes an action conditionally available based on a variable's state?
available when (e.g., available when @variables.report_type != "")
What "anti-pattern" should developers avoid when designing NGA instruction logic to ensure it doesn't feel like an old-fashioned phone tree?
Avoid using heavy "if/else" logic; instead, lead with general instructions and only add determinism where non-deterministic behavior is seen.
When is the Agent allowed to proceed without requiring the user to validate a record match?
Only when there is an exact match on the record name (e.g., Success Plan name or Account name).
How should the Agent handle a situation where a list output contains only one single result?
It should pick that single result and move forward automatically instead of asking the user to select it.
What does the -> symbol mean when placed after instructions: in a topic's reasoning block?
Answer: It creates a procedure that can contain conditionals (if/else), run actions, and build instructions dynamically.
What built-in utility action is used to navigate from one topic to another in Agent Script?
@utils.transition (e.g., @utils.transition to @topic.billing_specialist)
According to the NGA strategy, what should a developer do with "overloaded" or long instructions during the migration?
Refactor them by consolidating similar Topics or using the new "Topic selection" feature to reduce sprawl.
Under NGA guidelines, what must happen before the Agent invokes an action to create or update a record?
It must display the collected inputs to the user for final confirmation or changes.
When using "me" or "I" in a prompt, which record is the Agent required to identify first?
The user's own record.
What syntax is used to insert variable values into text strings (template expressions)?
{!@variables.variable_name} (the curly braces with exclamation point)
Name the three types of action input bindings in Agent Script.
Fixed/literal values (e.g., with format="pdf"), variable bindings (e.g., with user_id=@variables.user_id), and LLM slot-filling (e.g., with report_type=...)
What is the main risk identified for continuing to develop features in the "legacy" way during the migration period.
Rework (or having to redo everything after the cutover)
What is the "Redundancy" strategy for GPT Opt-Out in NGA (mention both layers)?
Layer 1 is a frontend hardcoded list for immediate blocks; Layer 2 is backend deterministic logic within functional flows.
How should "System-level" errors (like an API being down) be presented to the user?
They must be translated from technical terms into human-friendly language (e.g., "Sorry, it looks like that field is empty").
What are the three main blocks that every Agent Script file must contain?
config: (agent metadata), system: (global settings/messages), and at least one topic block (where the agent "lives")
In the BidirectionalNavigation pattern, how does a specialist topic automatically return control to the main topic after completing its task?
By including transition to @topic.general_support inside the action callback (after the set statements)